ÿ‚ �'Nœ 8@p€àÁ !‡ /hYW”cïõIyöI yöIDöIœDöI{IöI¤{IöI±LöI±LöI;VöIœ;VöIqYöIhqYöIKmöIôKmöIé­öIpé­öIM³öIxM³öIƒ¶öI䃶öI ÀöIp ÀöICÃöI<CÃöI×öIÈ×öI»÷ID »÷I÷IL ÷IU ÷I¸ U ÷Iß)÷ID ß)÷I-÷I -÷IWZ÷Iœ WZ÷Im\÷Ixm\÷Iû^÷Iû^÷I[ƒ÷I‚[ƒ÷I�°÷IÚ�°÷IÓ³÷IªÓ³÷I·½÷IÚ ·½÷Isò÷Iê sò÷IÿøIJÿøIÙøIÙøIøIJøI øIz øI£øIª£øIßxùIJßxùIeúIjeúI“rúIB“rúI=ûIr=ûI ÙûI  ÙûIµëûI2"€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxcïõIWöIèWöIõDöIdõDöI—KöIl—KöI!UöIè!UöIWXöI´WXöI)nöI@)nöIÇ®öI¼Ç®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIû×öI˜û×öI™÷I™÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷I¥`÷IÌ ¥`÷I…÷IP …÷IG²÷IpG²÷IÏÀ÷ILÏÀ÷I‹õ÷IØ‹õ÷I}øIü}øI{ øI({ øIû¢øI´û¢øIÓxùIxÓxùI eúI| eúI‡rúI\‡rúIûIèûIרûI,רûIëûI0€} 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$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=Õ=Õ?J@JB¿B¿D4E4G©G©I7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`KbËbËd?e?g³g³i'j'l›l›noqƒqƒs÷s÷ukvkxßxßzS{S}Ç}Ç;€;‚¯‚¯„#…#‡™‡™‰ŠŒƒŒƒŽøŽø�m‘m“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„cïõIyöIyöIDöI|DöI{IöI„{IöI±LöIð±LöI;VöI|;VöIqYöIHqYöIEmöIÔEmöIã­öIPã­öIM³öIXM³öIƒ¶öIă¶öI ÀöIP ÀöICÃöICÃöI×öI¨×öIµ÷I$ µ÷I÷I, ÷IU ÷I˜ U ÷Iß)÷I$ ß)÷I-÷Ið -÷IWZ÷I| WZ÷IÛ`÷IXÛ`÷I;…÷IÔ;…÷I}²÷Iô}²÷IoÀ÷IÐoÀ÷I+õ÷I\+õ÷I£øI€£øI¡øI¬¡øIõ¢øI8õ¢øIÍxùIüÍxùIeúIeúI�rúIà�rúI×ûIl×ûI§ØûI̧ØûIOëûIЀ} 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 })çxcïõI]öIè]öIûDöIdûDöI—KöIl—KöI!UöIè!UöIWXöI´WXöI)nöI@)nöIÇ®öI¼Ç®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIû×öI˜û×öI™÷I™÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I9Z÷Ið 9Z÷Ic`÷IÈ c`÷IÄ÷ID Ä÷I³÷Id³÷IóÀ÷I<óÀ÷I¯õ÷Iȯõ÷I1øIì1øI/!øI/!øIï¢øI¤ï¢øIÇxùIhÇxùIeúIleúI{rúIL{rúI¿ûIØ¿ûI�ØûI�ØûI7ëûI €} 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$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†]‡]‰ЉЋCŒC޶޶�)‘)“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxiïõI�öIè�öI;DöId;DöI£KöIl£KöI-UöIè-UöIcXöI´cXöIomöI@omöI ®öI¼ ®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIA×öI˜A×öIß÷Iß÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷IçÀ÷IDçÀ÷I£õ÷IУõ÷IÑøIôÑøIÏ øI Ï øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIûIàûIéØûI$éØûI‘ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIQöIèQöIïDöIdïDöI£KöIl£KöI-UöIè-UöIcXöI´cXöI#nöI@#nöIÁ®öI¼Á®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIõ×öI˜õ×öI“÷I“÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I±`÷IÌ ±`÷I…÷IP …÷IS²÷IpS²÷I�º÷IL�º÷I7þ÷IÈ7þ÷Iõ øIìõ øIóøIÈóøIÝ¢øITÝ¢øIµxùIµxùIïdúIïdúIirúIüirúIûIˆûIרûIÌרûIëûIЀ} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI�öIè�öI;DöId;DöI£KöIl£KöI-UöIè-UöIcXöI´cXöIimöI@imöI®öI¼®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöI;×öI˜;×öIÙ÷IÙ÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷IÿÀ÷IDÿÀ÷I»õ÷Iлõ÷I[øIô[øIY!øI Y!øIã¢øI¬ã¢øI»xùIp»xùIõdúItõdúIorúITorúI¡ûIà¡ûIqØûI$qØûIëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIQöIèQöIïDöIdïDöI£KöIl£KöI-UöIè-UöIcXöI´cXöI#nöI@#nöIÁ®öI¼Á®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIõ×öI˜õ×öI“÷I“÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IEZ÷Ið EZ÷I‡`÷IÈ ‡`÷Iç„÷ID ç„÷I%³÷Id%³÷IKº÷I<Kº÷Iõý÷I¸õý÷IÉ øIÜÉ øIÇøI¸ÇøIË¢øIDË¢øI£xùI£xùIÝdúI ÝdúIWrúIìWrúI­ûIx­ûI}ØûI¼}ØûI%ëûIÀ€} 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$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j llnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†P‡P‰Â‰Â‹4Œ4ަަ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxiïõI£öIè£öIADöIdADöI¯KöIl¯KöI9UöIè9UöIoXöIRoXöIomöI‚omöI ®öI’ ®öI�µöIº�µöI ¿öIÊ ¿öIAÂöIzAÂöIA×öIªA×öIß÷Iº ß÷IS÷Iâ S÷IÝ(÷Iò Ý(÷I,÷I¢,÷IUY÷IÒUY÷I«`÷I¢«`÷I …÷I² …÷IM²÷I M²÷I“º÷IÚ“º÷I=þ÷Iê=þ÷IøIJ"øIøI$øI颸IJ%颸IÁxùIê&ÁxùIûdúI )ûdúIurúIâ*urúICûI,CûIÙûI²0ÙûI»ëûIÒ2€} 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·liïõIKöIÐKöIéDöILéDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöInöI(nöI»®öI¤»®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIï×öI€ï×öI�÷Iü�÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I½`÷I´ ½`÷I…÷I8 …÷I_²÷IX_²÷I­¸÷I4­¸÷I3 øI 3 øI£øI,£øIÙxùIðÙxùIeúIôeúI7ûIÔ7ûIÙûIÙûI¯ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI�öIè�öI;DöId;DöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöIomöI@omöI ®öI¼ ®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIA×öI˜A×öIß÷Iß÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷IÃ`÷IÌ Ã`÷I#…÷IH #…÷Ie²÷Ihe²÷I#»÷ID#»÷IÍþ÷IÀÍþ÷I�øIä�øIøIÀøIÑ¢øILÑ¢øI©xùI©xùIãdúIãdúI]rúIô]rúIÑûI€ÑûI¡ØûIÄ¡ØûIIëûIÈ€} 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$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbäbädWeWgÊgÊi=j=l°l°n!o!q’q’stvtvtxåxåzV{V}Ç}Ç8€8‚©‚©„…‡s‡s‰ä‰ä‹UŒUŽÆŽÆ�7‘7“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·liïõIKöIÐKöIéDöILéDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöI»®öI¤»®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIï×öI€ï×öI�÷Iü�÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷I]Z÷IØ ]Z÷I«`÷I° «`÷I …÷I,  …÷II³÷ILI³÷I'º÷I$'º÷I« øI�« øI颸I颸IÁxùIàÁxùIûdúIäûdúI¿ûIÄ¿ûI�ØûI�ØûI7ëûI €} 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$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·loïõI©öIЩöIGDöILGDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIumöI(umöI®öI¤®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIG×öI€G×öIå÷Iüå÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I¡¸÷I,¡¸÷I!øI˜!øI £øI$ £øIåxùIèåxùIeúIìeúIsûIÌsûICÙûICÙûIëëûI€} 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·loïõIEöIÐEöIãDöILãDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI¯®öI¤¯®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIã×öI€ã×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IW`÷I´ W`÷I·„÷I8 ·„÷Iù±÷IXù±÷I§¸÷I4§¸÷I øI  øIï¢øI,ï¢øIÇxùIðÇxùIeúIôeúI¿ûIÔ¿ûI�ØûI�ØûI7ëûI€} 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·loïõI£öIУöIADöILADöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIumöI(umöI®öI¤®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIG×öI€G×öIå÷Iüå÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I»÷I,»÷I™øI˜™øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIûIÌûIÑØûIÑØûIyëûI€} 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,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡ЇЉþ‰þ‹rŒr޿޿�Z‘Z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·loïõIEöIÐEöIãDöILãDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöIµ®öI¤µ®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIé×öI€é×öI‡÷Iü‡÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IuZ÷IØ uZ÷I‡`÷I° ‡`÷Iç„÷I, ç„÷I%³÷IL%³÷I!º÷I$!º÷I� øI�� øI×¢øI×¢øI¯xùIà¯xùIédúIäédúI_ûIÄ_ûI/ØûI/ØûI×êûI €} 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$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI©öIЩöIGDöILGDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI{möI({möI®öI¤®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIM×öI€M×öIë÷Iüë÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷I¡¸÷I,¡¸÷IŸøI˜ŸøI£øI$£øIßxùIèßxùIeúIìeúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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·luïõI?öIÐ?öIÝDöILÝDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI nöI( nöI©®öI¤©®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIÝ×öI€Ý×öI{÷Iü{÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I?`÷I´ ?`÷IŸ„÷I8 Ÿ„÷Iá±÷IXá±÷I§¸÷I4§¸÷I øI  øI£øI,£øIÙxùIðÙxùIeúIôeúI ûIÔ ûIÝØûIÝØûI…ëûI€} 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·luïõI©öIЩöIGDöILGDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI{möI({möI®öI¤®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIM×öI€M×öIë÷Iüë÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I#»÷I,#»÷I#øI˜#øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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·luïõI?öIÐ?öIÝDöILÝDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI¯®öI¤¯®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIã×öI€ã×öI�÷Iü�÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IoZ÷IØ oZ÷I±`÷I° ±`÷I…÷I, …÷IO³÷ILO³÷I!º÷I$!º÷I“ øI�“ øIõ¢øIõ¢øIÍxùIàÍxùIeúIäeúIÅûIÄÅûI•ØûIR•ØûI=ëûIr€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1vªL b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}"}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements648input0Elements41472input1Elements5184invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 k«Ñä$ 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$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI¯öIЯöIMDöILMDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI®öI¤®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIS×öI€S×öIñ÷Iüñ÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷Ic`÷I´ c`÷IÄ÷I0 Ä÷I²÷IP²÷I§¸÷I,§¸÷ISøI˜SøI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIïØûIïØûI—ëûI€} 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$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâI[J[LÒLÒNIOIQÀQÀS7T7V®V®X%Y%[œ[œ]^`Š`ŠbcexexgïgïifjflÝlÝnOoOqÁqÁs3t3v¥v¥xy{‰{‰}û}ûm€m‚߂߄Q…Q‡ÇÉ5Š5Œ§Œ§Ž�‘‹‘‹“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI¯öIЯöIMDöILMDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI®öI¤®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIS×öI€S×öIñ÷Iüñ÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷I5»÷I,5»÷IãøI˜ãøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI­ûIÌ­ûI}ØûI}ØûI%ëûI€} 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$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Þ=Þ?P@PBÂBÂD4E4G¦G¦IJLŠLŠNüNüPnQnSàSàURVRXÄXÄZ6[6]¨]¨_`bŒbŒdþdþfpgpiâiâkTlTnÆnÆp8q8sªsªuvxŽxŽz{}r}rää�V‚V„ȄȆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI9öIÐ9öI×DöIL×DöI¯KöIT¯KöI9UöIÐ9UöIoXöIœoXöInöI(nöI£®öI¤£®öI�µöI¬�µöI ¿öI( ¿öIAÂöIôAÂöI××öI€××öIu÷Iüu÷IS÷I S÷IÝ(÷I€ Ý(÷I,÷IL ,÷IQZ÷IØ QZ÷I½`÷I° ½`÷I…÷I, …÷I[³÷IL[³÷I'º÷I$'º÷IÏ øI�Ï øIû¢øIû¢øIÓxùIàÓxùI eúIä eúIÝûIÄÝûI­ØûI­ØûIUëûI €} 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·l{ïõIµöIеöISDöILSDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI‡möI(‡möI%®öI¤%®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIY×öI€Y×öI÷÷Iü÷÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IW`÷I´ W`÷I·„÷I0 ·„÷Iù±÷IPù±÷I­¸÷I,­¸÷IøI˜øI£øI$£øIñxùIèñxùI+eúIì+eúIsûIÌsûICÙûICÙûIëëûI€} 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·l{ïõI-öIÐ-öIËDöILËDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIÿmöI(ÿmöI�®öI¤�®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÑ×öI€Ñ×öIo÷Iüo÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I`÷I´ `÷Iu„÷I8 u„÷I·±÷IX·±÷I­¸÷I4­¸÷I± øI ± øI£øI,£øIëxùIðëxùI%eúIô%eúIIûIÔIûIÙûIÙûIÁëûI€} 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·l{ïõIµöIеöISDöILSDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI�möI(�möI®öI¤®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIS×öI€S×öIñ÷Iüñ÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷IA»÷I,A»÷I�øI˜�øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIרûIרûIëûI€} 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·l{ïõI3öIÐ3öIÑDöILÑDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIÿmöI(ÿmöI�®öI¤�®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÑ×öI€Ñ×öIo÷Iüo÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷I]Z÷IØ ]Z÷I½`÷I° ½`÷I…÷I, …÷I[³÷IL[³÷I9º÷I$9º÷I5 øI�5 øIû¢øIû¢øIÓxùIàÓxùI eúIä eúIÑûIÄÑûI¡ØûI¡ØûIIëûI €} 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·l�ïõI»öIлöIYDöILYDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI�möI(�möI+®öI¤+®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI_×öI€_×öIý÷Iüý÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IE`÷I´ E`÷I¥„÷I0 ¥„÷Iç±÷IPç±÷I­¸÷I,­¸÷I»øI˜»øI£øI$£øI÷xùIè÷xùI1eúIì1eúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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·l�ïõI'öIÐ'öIÅDöILÅDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIùmöI(ùmöI—®öI¤—®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIË×öI€Ë×öIi÷Iüi÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÃ`÷I´ Ã`÷I#…÷I8 #…÷Ie²÷IXe²÷I­¸÷I4­¸÷I øI  øI£øI,£øIÙxùIðÙxùIeúIôeúI×ûIÔ×ûI§ØûI§ØûIOëûI€} 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·l�ïõI»öIлöIYDöILYDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI‡möI(‡möI%®öI¤%®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIY×öI€Y×öI÷÷Iü÷÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Ii`÷I´ i`÷IÉ„÷I0 É„÷I ²÷IP ²÷IS»÷I,S»÷I]øI˜]øI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI'öIÐ'öIÅDöILÅDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIùmöI(ùmöI—®öI¤—®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIË×öI€Ë×öIi÷Iüi÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IcZ÷IØ cZ÷IÉ`÷I° É`÷I)…÷I, )…÷Ig³÷ILg³÷IEº÷I$Eº÷I› øI�› øIï¢øIï¢øIÇxùIàÇxùIeúIäeúIqûIÄqûIAØûIAØûIéêûI €} 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$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¾=¾?/@/B B DEG‚G‚IóIóKdLdNÕNÕPFQFS·S·U(V(X™X™Z [ ]{]{_ì_ìa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI3öI�3öIÑDöIÑDöI÷HöI$÷HöInöI�nöI£®öI£®öIɲöI$ɲöI××öI�××öIu÷Iu÷I›÷I$›÷I+_÷I�+_÷I‹ƒ÷Iü‹ƒ÷Is¶÷I s¶÷I?øI| ?øI+£øIÜ +£øIyùI  yùI=eúI¤ =eúIgûI„ gûI7ÙûIÈ7ÙûIßëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI!öI�!öI¿DöI ¿DöIÇKöIÇKöIómöI€ómöI‘®öIü‘®öI™µöI™µöIÅ×öIpÅ×öIc÷Iìc÷Ik÷Iôk÷IÉ`÷I`É`÷I)…÷IÌ)…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI9öI�9öI×DöI×DöI÷HöI$÷HöInöI�nöI£®öI£®öIɲöI$ɲöI××öI�××öIu÷Iu÷I›÷I$›÷I+_÷I�+_÷I‹ƒ÷Iü‹ƒ÷Is¶÷I s¶÷I?øI| ?øI£øIÜ £øIñxùI  ñxùI+eúI¤ +eúIûûI„ ûûIËØûIÈËØûIsëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI!öI�!öI¿DöI ¿DöIÇKöIÇKöIómöI€ómöI‘®öIü‘®öI™µöI™µöIÅ×öIpÅ×öIc÷Iìc÷Ik÷Iôk÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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�ïõIÍöI�ÍöIkDöI kDöIIöIIöIŸmöI€ŸmöI=®öIü=®öIç²öIç²öIq×öIpq×öI÷Iì÷I¹÷Iô¹÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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�ïõI'öI�'öIÅDöI ÅDöI�KöI�KöIùmöI€ùmöI—®öIü—®öIoµöIoµöIË×öIpË×öIi÷Iìi÷IA÷IôA÷I½`÷I`½`÷I…÷IÌ…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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�ïõIÇöI�ÇöIeDöI eDöIIöIIöI™möI€™möI7®öIü7®öIç²öIç²öIk×öIpk×öI ÷Iì ÷I¹÷Iô¹÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúI¿ûIT ¿ûI�ØûI˜�ØûI7ëûIœ€} 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�ïõI'öI�'öIÅDöI ÅDöI�KöI�KöIùmöI€ùmöI—®öIü—®öIoµöIoµöIË×öIpË×öIi÷Iìi÷IA÷IôA÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI-öI�-öIËDöI ËDöIÇKöIÇKöIÿmöI€ÿmöI�®öIü�®öI™µöI™µöIÑ×öIpÑ×öIo÷Iìo÷Ik÷Iôk÷I·`÷I`·`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIÁöI�ÁöI_DöI _DöI÷HöI÷HöI“möI€“möI1®öIü1®öIɲöIɲöIe×öIpe×öI÷Iì÷I›÷Iô›÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI-öI�-öIËDöI ËDöIÇKöIÇKöIÿmöI€ÿmöI�®öIü�®öI™µöI™µöIÑ×öIpÑ×öIo÷Iìo÷Ik÷Iôk÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÁöI�ÁöI_DöI _DöI!IöI!IöI�möI€�möI+®öIü+®öIó²öIó²öI_×öIp_×öIý÷Iìý÷IÅ÷IôÅ÷I±`÷I`±`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI3öI�3öIÑDöI ÑDöI…KöI…KöInöI€nöI£®öIü£®öIWµöIWµöI××öIp××öIu÷Iìu÷I)÷Iô)÷I_÷I`_÷Iƒ÷Ĩ÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI»öI�»öIYDöI YDöI'IöI'IöI�möI€�möI+®öIü+®öIù²öIù²öI_×öIp_×öIý÷Iìý÷IË÷IôË÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI3öI�3öIÑDöI ÑDöI…KöI…KöInöI€nöI£®öIü£®öIWµöIWµöI××öIp××öIu÷Iìu÷I)÷Iô)÷I_÷I`_÷Iƒ÷Ĩ÷Ig¶÷Iìg¶÷I3øIL 3øI颸I¬ 颸IÁxùIp ÁxùIûdúIt ûdúIkûIT kûI;ØûI˜;ØûIãêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI»öI�»öIYDöI YDöI÷HöI÷HöI‡möI€‡möI%®öIü%®öIɲöIɲöIY×öIpY×öI÷÷Iì÷÷I›÷Iô›÷I¥`÷I`¥`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöI»KöI»KöInöI€nöI£®öIü£®öI�µöI�µöI××öIp××öIu÷Iìu÷I_÷Iô_÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI ûIT  ûIÝØûI´ ÝØûI…ëûI"€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIµöI�µöISDöI SDöI÷HöI÷HöI‡möI€‡möI%®öIü%®öIɲöIɲöIY×öIpY×öI÷÷Iì÷÷I›÷Iô›÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Ia¶÷Iìa¶÷I-øIL -øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöI»KöI»KöInöI€nöI£®öIü£®öI�µöI�µöI××öIp××öIu÷Iìu÷I_÷Iô_÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ia¶÷Iìa¶÷I-øIL -øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI¯öI�¯öIMDöI MDöIIöIIöI�möI€�möI®öIü®öIí²öIí²öIS×öIpS×öIñ÷Iìñ÷I¿÷Iô¿÷I_÷I`_÷Isƒ÷IÌsƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI?öI�?öIÝDöI ÝDöIKöIKöI nöI€ nöI©®öIü©®öIQµöIQµöIÝ×öIpÝ×öI{÷Iì{÷I#÷Iô#÷I_÷I`_÷Isƒ÷IÌsƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI¯öI�¯öIMDöI MDöIIöIIöI{möI€{möI®öIü®öIí²öIí²öIM×öIpM×öIë÷Iìë÷I¿÷Iô¿÷I_÷I`_÷Isƒ÷IÌsƒ÷I[¶÷Iì[¶÷I'øIL 'øI颸I¬ 颸IÁxùIp ÁxùIûdúIt ûdúI§ûIT §ûIwØûI˜wØûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI?öI�?öIÝDöI ÝDöIKöIKöI nöI€ nöI©®öIü©®öIQµöIQµöIÝ×öIpÝ×öI{÷Iì{÷I#÷Iô#÷I_÷I`_÷Isƒ÷IÌsƒ÷I[¶÷Iì[¶÷I'øIL 'øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LoïõI©öI�©öIGDöI GDöIëHöIëHöI{möI€{möI®öIü®öI½²öI½²öIM×öIpM×öIë÷Iìë÷I�÷Iô�÷I_÷I`_÷Isƒ÷IÌsƒ÷I[¶÷Iì[¶÷I'øIL 'øI £øI¬  £øIåxùIp åxùIeúIt eúIgûIT gûI7ÙûI˜7ÙûIßëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LoïõI?öI�?öIÝDöI ÝDöIµKöIµKöInöI€nöI¯®öIü¯®öI‡µöI‡µöIã×öIpã×öI�÷Iì�÷IY÷IôY÷IÃ`÷I`Ã`÷I#…÷IÔ#…÷I[¶÷Iô[¶÷I'øIT 'øI£øI´ £øI/wùIx /wùI GúI|  GúICûI\ CûI9×ûI 9×ûIáéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI©öI�©öIGDöI GDöIëHöIëHöIumöI€umöI®öIü®öI½²öI½²öIG×öIpG×öIå÷Iìå÷I�÷Iô�÷I_÷I`_÷Isƒ÷IÌsƒ÷I[¶÷Iì[¶÷I'øIL 'øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LoïõI?öI�?öIÝDöI ÝDöI¯KöI¯KöInöI€nöI¯®öIü¯®öI�µöI�µöIã×öIpã×öI�÷Iì�÷IS÷IôS÷Iÿ`÷I`ÿ`÷I_…÷IÔ_…÷I[¶÷Iô[¶÷I'øIT 'øIï¢øI´ ï¢øIwùIx wùIõFúI| õFúIËûI\ ËûIÁÖûI ÁÖûIiéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LiïõI£öI�£öIADöI ADöIIöIIöIumöI€umöI®öIü®öIá²öIá²öIG×öIpG×öIå÷Iìå÷I³÷Iô³÷I½`÷I`½`÷I…÷IÔ…÷IU¶÷IôU¶÷I!øIT !øI£øI´ £øIAwùIx AwùIGúI| GúIûI\ ûIu×ûI u×ûIêûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LiïõIEöI�EöIãDöI ãDöIsKöIsKöInöI€nöIµ®öIüµ®öIEµöIEµöIé×öIpé×öI‡÷Iì‡÷I÷Iô÷Ié^÷I`é^÷IIƒ÷IÌIƒ÷IU¶÷IìU¶÷I!øIL !øIõ¢øI¬ õ¢øIwùIp wùIûFúIt ûFúIÅûIT ÅûI»ÖûI˜»ÖûIcéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LiïõIEöI�EöIãDöI ãDöIsKöIsKöInöI€nöIµ®öIüµ®öIEµöIEµöIé×öIpé×öI‡÷Iì‡÷I÷Iô÷Ié^÷I`é^÷IIƒ÷IÌIƒ÷IU¶÷IìU¶÷I!øIL !øIã¢øI¬ ã¢øI wùIp  wùIéFúIt éFúIeûIT eûI[ÖûI˜[ÖûIéûIœ€} 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$?%?'²'²)%*%,˜,˜. / 1~1~3ñ3ñ5d6d8×8×:J;J=»=»?,@,B�B�DEGGIðIðKaLaNÒNÒPCQCS´S´U%V%X–X–Z[]x]x_é_éafbfd×d×fHgHi¹i¹k*l*n›n›p q s}s}uîuîw_x_zÐzÐ|A}A²²�#‚#„”„”†‡‰s‰s‹ã‹ã�SŽS�Ã�Ã’ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„cïõIöIöIDöI|DöI�IöI„�IöI·LöIð·LöIAVöI|AVöIwYöIHwYöIKmöIÔKmöIé­öIPé­öIS³öIXS³öI‰¶öIĉ¶öIÀöIPÀöIIÃöIIÃöI×öI¨×öI»÷I$ »÷I%÷I, %÷I[ ÷I˜ [ ÷Iå)÷I$ å)÷I-÷Ið -÷I]Z÷I| ]Z÷I±`÷IX±`÷I…÷IÔ…÷IS²÷IôS²÷I3À÷IÐ3À÷Iïô÷I\ïô÷IqøI€qøIoøI¬oøI£øI8£øIßxùIüßxùIeúIeúI“rúIà“rúICûIlCûIÙûI°ÙûI»ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxcïõIWöIèWöIõDöIdõDöI—KöIl—KöI!UöIè!UöIWXöI´WXöI#nöI@#nöIÁ®öI¼Á®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIõ×öI˜õ×öI“÷I“÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷Iç`÷IÌ ç`÷IG…÷IP G…÷I‰²÷Ip‰²÷I¥À÷IL¥À÷Iaõ÷IØaõ÷IŸøIüŸøI�øI(�øI £øI´ £øIåxùIxåxùIeúI|eúI™rúI\™rúImûIèmûI=ÙûI,=ÙûIåëûI0€} 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„cïõIyöIyöIDöI|DöI�IöI„�IöI·LöIð·LöIAVöI|AVöIwYöIHwYöIKmöIÔKmöIé­öIPé­öIS³öIXS³öI‰¶öIĉ¶öIÀöIPÀöIIÃöIIÃöI×öI¨×öI»÷I$ »÷I%÷I, %÷I[ ÷I˜ [ ÷Iå)÷I$ å)÷I-÷Ið -÷I]Z÷I| ]Z÷I«`÷IX«`÷I …÷IÔ …÷IM²÷IôM²÷IŸÀ÷IПÀ÷I[õ÷I\[õ÷I{øI€{øIyøI¬yøIï¢øI8ï¢øIÇxùIüÇxùIeúIeúI{rúIà{rúIËûIlËûI›ØûI°›ØûICëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxcïõIWöIèWöIõDöIdõDöI—KöIl—KöI!UöIè!UöIWXöI´WXöI)nöI@)nöIÇ®öI¼Ç®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIû×öI˜û×öI™÷I™÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷Iu`÷IÌ u`÷IÕ„÷IP Õ„÷I²÷Ip²÷IÉÀ÷ILÉÀ÷I…õ÷IØ…õ÷IYøIüYøIW øI(W øIû¢øI´û¢øIÓxùIxÓxùI eúI| eúI‡rúI\‡rúIûûIèûûIËØûI,ËØûIsëûI0€} 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$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=ê=ê?a@aBØBØDOEOGÆGÆIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡œ‡œ‰ Š Œ~Œ~ŽïŽï�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxiïõI£öIè£öIADöIdADöI£KöIl£KöI-UöIè-UöIcXöI´cXöIomöI@omöI ®öI¼ ®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIA×öI˜A×öIß÷Iß÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I«`÷IÌ «`÷I …÷IH  …÷IM²÷IhM²÷IÁ÷IDÁ÷IÍõ÷IÐÍõ÷I¯øIô¯øI­!øI ­!øI颸I¬颸IÁxùIpÁxùIûdúItûdúIurúITurúI¿ûIà¿ûI�ØûI$�ØûI7ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIQöIèQöIïDöIdïDöI£KöIl£KöI-UöIè-UöIcXöI´cXöInöI@nöI»®öI¼»®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIï×öI˜ï×öI�÷I�÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷IÛ`÷IÌ Û`÷I;…÷IP ;…÷I}²÷Ip}²÷IÛ½÷ILÛ½÷I…øIÈ…øI7øIì7øI5øIÈ5øI颸IT颸IÁxùIÁxùIûdúIûdúIurúIüurúIUûIˆUûI%ÙûIÌ%ÙûIÍëûIЀ} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI�öIè�öI;DöId;DöI£KöIl£KöI-UöIè-UöIcXöI´cXöIomöI@omöI ®öI¼ ®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIA×öI˜A×öIß÷Iß÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷I/Á÷ID/Á÷Iëõ÷IÐëõ÷I9øIô9øI7"øI 7"øI×¢øI¬×¢øI¯xùIp¯xùIédúItédúIcrúITcrúIYûIàYûI)ØûI$)ØûIÑêûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIQöIèQöIïDöIdïDöI£KöIl£KöI-UöIè-UöIcXöI´cXöI#nöI@#nöIÁ®öI¼Á®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIõ×öI˜õ×öI“÷I“÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I™`÷IÌ ™`÷Iù„÷IP ù„÷I;²÷Ip;²÷IcÀ÷ILcÀ÷I øIÈ øI»øIì»øI¹øIȹøI×¢øIT×¢øI¯xùI¯xùIédúIédúIcrúIücrúIÝûIˆÝûI­ØûIÌ­ØûIUëûIЀ} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI©öIè©öIGDöIdGDöIµKöIlµKöI?UöIè?UöIuXöI´uXöIumöI@umöI®öI¼®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöIG×öI˜G×öIå÷Iå÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷IŸ`÷IÌ Ÿ`÷Iÿ„÷IH ÿ„÷IA²÷IhA²÷I™½÷ID™½÷ICøIÀCøIùøIäùøI÷øIÀ÷øI×¢øIL×¢øI¯xùI¯xùIédúIédúIcrúIôcrúIÝûI€ÝûI­ØûIÄ­ØûIUëûIÈ€} 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·loïõIEöIÐEöIãDöILãDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöInöI(nöIµ®öI¤µ®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIé×öI€é×öI‡÷Iü‡÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÃ`÷I´ Ã`÷I#…÷I8 #…÷Ie²÷IXe²÷Iɽ÷I4ɽ÷IøI øI£øI,£øIÙxùIðÙxùIeúIôeúI7ûIÔ7ûIÙûIÙûI¯ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI£öIè£öIADöIdADöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöIumöI@umöI®öI¼®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIG×öI˜G×öIå÷Iå÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I«`÷IÌ «`÷I …÷IH  …÷IM²÷IhM²÷IÀ÷IDÀ÷I¿øIÀ¿øIqøIäqøIoøIÀoøIÅ¢øILÅ¢øI�xùI�xùI×dúI×dúIQrúIôQrúI›ûI€›ûIkØûIÄkØûIëûIÈ€} 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·loïõIKöIÐKöIéDöILéDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöIµ®öI¤µ®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIé×öI€é×öI‡÷Iü‡÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I«`÷I´ «`÷I …÷I8  …÷IM²÷IXM²÷IQÀ÷I4QÀ÷I‘øI ‘øIï¢øI,ï¢øIÇxùIðÇxùIeúIôeúIÑûIÔÑûI¡ØûI¡ØûIIëûI€} 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·loïõI©öIЩöIGDöILGDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI{möI({möI®öI¤®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIM×öI€M×öIë÷Iüë÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I™½÷I,™½÷IøI˜øI£øI$£øIÙxùIèÙxùIeúIìeúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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·loïõI?öIÐ?öIÝDöILÝDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI¯®öI¤¯®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIã×öI€ã×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I8 ÿ„÷IA²÷IXA²÷I½½÷I4½½÷IÝøI ÝøI£øI,£øIÙxùIðÙxùIeúIôeúIûIÔûIéØûIéØûI‘ëûI€} 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·loïõI©öIЩöIGDöILGDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI{möI({möI®öI¤®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIM×öI€M×öIë÷Iüë÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IÀ÷I,À÷I�øI˜�øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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·loïõIEöIÐEöIãDöILãDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI¯®öI¤¯®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIã×öI€ã×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IQ`÷I´ Q`÷I±„÷I8 ±„÷Ió±÷IXó±÷IEÀ÷I4EÀ÷IaøI aøI颸I,颸IÁxùIðÁxùIûdúIôûdúI¡ûIÔ¡ûIqØûIqØûIëûI€} 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·luïõI¯öIЯöIMDöILMDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI�möI(�möI®öI¤®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIS×öI€S×öIñ÷Iüñ÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I«½÷I,«½÷I§øI˜§øI£øI$£øIßxùIèßxùIeúIìeúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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·luïõI9öIÐ9öI×DöIL×DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI nöI( nöI©®öI¤©®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIÝ×öI€Ý×öI{÷Iü{÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I�`÷I´ �`÷Ií„÷I8 í„÷I/²÷IX/²÷I±½÷I4±½÷I­øI ­øI£øI,£øIëxùIðëxùI%eúIô%eúIsûIÔsûICÙûICÙûIëëûI€} 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$j™j™lmo�o�qôqôsgtgvÚvÚxMyM{À{À}3~3€¦€¦‚ƒ…Œ…Œ‡ÿ‡ÿ‰qŠqŒãŒãŽU�U‘ǑǓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI¯öIЯöIMDöILMDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI�möI(�möI®öI¤®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIS×öI€S×öIñ÷Iüñ÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I-À÷I,-À÷I+øI˜+øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI9öIÐ9öI×DöIL×DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI nöI( nöI©®öI¤©®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIÝ×öI€Ý×öI{÷Iü{÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IE`÷I´ E`÷I¥„÷I8 ¥„÷Iç±÷IXç±÷I3À÷I43À÷I1øI 1øI£øI,£øIÙxùIðÙxùIeúIôeúIûûIÔûûIËØûIËØûIsëûI€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1t¨H b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 i©Ë ä$ 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$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIµöIеöISDöILSDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI‡möI(‡möI%®öI¤%®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIY×öI€Y×öI÷÷Iü÷÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IW`÷I´ W`÷I·„÷I0 ·„÷Iù±÷IPù±÷Iý÷I,ý÷ImøI˜møIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIÅûIÌÅûI•ØûI•ØûI=ëûI€} 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·l{ïõI3öIÐ3öIÑDöILÑDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöInöI(nöI£®öI¤£®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöI××öI€××öIu÷Iüu÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I{`÷I´ {`÷IÛ„÷I8 Û„÷I²÷IX²÷I½½÷I4½½÷IøI øI£øI,£øI÷xùIð÷xùI1eúIô1eúI…ûIÔ…ûIUÙûIUÙûIýëûI€} 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·l{ïõI3öIÐ3öIÑDöILÑDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöInöI(nöI£®öI¤£®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöI××öI€××öIu÷Iüu÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I9`÷I´ 9`÷I™„÷I8 ™„÷IÛ±÷IXÛ±÷I?À÷I4?À÷I‹øI ‹øI£øI,£øIßxùIðßxùIeúIôeúI ûIÔ ûIÝØûIÝØûI…ëûI€} 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·l{ïõI»öIлöIYDöILYDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI�möI(�möI+®öI¤+®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöI_×öI€_×öIý÷Iüý÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IE`÷I´ E`÷I¥„÷I0 ¥„÷Iç±÷IPç±÷IÛ½÷I,Û½÷I9øI˜9øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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·l{ïõI-öIÐ-öIËDöILËDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIùmöI(ùmöI—®öI¤—®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIË×öI€Ë×öIi÷Iüi÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷Ii`÷I´ i`÷IÉ„÷I8 É„÷I ²÷IX ²÷IÕ½÷I4Õ½÷IÓøI ÓøI£øI,£øIëxùIðëxùI%eúIô%eúIOûIÔOûIÙûIÙûIÇëûI€} 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$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀn5o5qªqªstv”v”x y {~{~}ó}óh€h‚݂݄R…R‡LJlj<Š<Œ±Œ±Ž&�&‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI»öIлöIYDöILYDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI‡möI(‡möI%®öI¤%®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIY×öI€Y×öI÷÷Iü÷÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷IuÀ÷I,uÀ÷IÕøI˜ÕøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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·l{ïõI-öIÐ-öIËDöILËDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIÿmöI(ÿmöI�®öI¤�®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÑ×öI€Ñ×öIo÷Iüo÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I-`÷I´ -`÷I�„÷I8 �„÷Iϱ÷IXϱ÷IiÀ÷I4iÀ÷IcøI cøI£øI,£øIÙxùIðÙxùIeúIôeúIãûIÔãûI³ØûI³ØûI[ëûI€} 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):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunènèp[q[sÎsÎuAvAx´x´z'{'}š}š € ‚€‚€„ó„ó†f‡f‰ىًLŒLŽ¿Ž¿�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIÁöIÐÁöI_DöIL_DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI�möI(�möI+®öI¤+®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI_×öI€_×öIý÷Iüý÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I3`÷I´ 3`÷I“„÷I0 “„÷IÕ±÷IPÕ±÷Iù½÷I,ù½÷IøI˜øI£øI$£øIëxùIèëxùI%eúIì%eúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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�ïõI'öIÐ'öIÅDöILÅDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIómöI(ómöI‘®öI¤‘®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIÅ×öI€Å×öIc÷Iüc÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÉ`÷I´ É`÷I)…÷I8 )…÷Ik²÷IXk²÷Ií½÷I4í½÷I™øI ™øI£øI,£øIëxùIðëxùI%eúIô%eúI1ûIÔ1ûIÙûIÙûI©ëûI€} 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�ïõI»öIлöIYDöILYDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI�möI(�möI+®öI¤+®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI_×öI€_×öIý÷Iüý÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Ic`÷I´ c`÷IÄ÷I0 Ä÷I²÷IP²÷I™À÷I,™À÷I§øI˜§øI£øI$£øIÙxùIèÙxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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�ïõI'öIÐ'öIÅDöILÅDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIùmöI(ùmöI—®öI¤—®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIË×öI€Ë×öIi÷Iüi÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÏ`÷I´ Ï`÷I/…÷I8 /…÷Iq²÷IXq²÷I�À÷I4�À÷I5øI 5øIû¢øI,û¢øIÓxùIðÓxùI eúIô eúI¹ûIÔ¹ûI‰ØûI‰ØûI1ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI-öI�-öIËDöIËDöI÷HöI$÷HöIÿmöI�ÿmöI�®öI�®öIɲöI$ɲöIÑ×öI�Ñ×öIo÷Io÷I›÷I$›÷I+_÷I�+_÷I‹ƒ÷Iü‹ƒ÷Is¶÷I s¶÷I?øI| ?øI1£øIÜ 1£øI yùI   yùICeúI¤ CeúIgûI„ gûI7ÙûIÈ7ÙûIßëûIÌ€} 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$…—…—‡ˆŠzŠzŒìŒìŽ^�^‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI3öI�3öIÑDöIÑDöI÷HöI$÷HöInöI�nöI£®öI£®öIɲöI$ɲöI××öI�××öIu÷Iu÷I›÷I$›÷I+_÷I�+_÷I‹ƒ÷Iü‹ƒ÷Is¶÷I s¶÷I?øI| ?øI£øIÜ £øIñxùI  ñxùI+eúI¤ +eúIïûI„ ïûI¿ØûIÈ¿ØûIgëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI!öI�!öI¿DöI ¿DöIÁKöIÁKöIómöI€ómöI‘®öIü‘®öI“µöI“µöIÅ×öIpÅ×öIc÷Iìc÷Ie÷Iôe÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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‡ïõIÓöI�ÓöIqDöI qDöIIöIIöI¥möI€¥möIC®öIüC®öIá²öIá²öIw×öIpw×öI÷Iì÷I³÷Iô³÷IÉ`÷I`É`÷I)…÷IÌ)…÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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‡ïõIÍöI�ÍöIkDöI kDöIIöIIöIŸmöI€ŸmöI=®öIü=®öIá²öIá²öIq×öIpq×öI÷Iì÷I³÷Iô³÷IÉ`÷I`É`÷I)…÷IÌ)…÷Im¶÷Iìm¶÷I9øIL 9øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI}ûIT }ûIMØûI˜MØûIõêûIœ€} 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‡ïõI'öI�'öIÅDöI ÅDöI�KöI�KöIómöI€ómöI‘®öIü‘®öIoµöIoµöIÅ×öIpÅ×öIc÷Iìc÷IA÷IªA÷IÏ`÷IšÏ`÷I/…÷IŠ/…÷Im¶÷Iâm¶÷I9øIº9øI£øI’ £øIñxùI2 ñxùI+eúIR +eúIûI*ûIïØûIÊïØûI—ëûIê€} 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�ïõIÍöI�ÍöIkDöI kDöI÷HöI÷HöI™möI€™möI7®öIü7®öIɲöIɲöIk×öIpk×öI ÷Iì ÷I›÷Iô›÷I½`÷I`½`÷I…÷IÌ…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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�ïõI-öI�-öIËDöI ËDöIÁKöIÁKöIùmöI€ùmöI—®öIü—®öI“µöI“µöIË×öIpË×öIi÷Iìi÷Ie÷Iôe÷I·`÷I`·`÷I…÷IÌ…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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�ïõIÇöI�ÇöIeDöI eDöI÷HöI÷HöI™möI€™möI7®öIü7®öIɲöIɲöIk×öIpk×öI ÷Iì ÷I›÷Iô›÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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�ïõI-öI�-öIËDöI ËDöIÁKöIÁKöIùmöI€ùmöI—®öIü—®öI“µöI“µöIË×öIpË×öIi÷Iìi÷Ie÷Iôe÷IÉ`÷I`É`÷I)…÷IÌ)…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÇöI�ÇöIeDöI eDöI!IöI!IöI“möI€“möI1®öIü1®öIó²öIó²öIe×öIpe×öI÷Iì÷IÅ÷IôÅ÷I·`÷I`·`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIñxùIp ñxùI+eúIt +eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÁöI�ÁöI_DöI _DöI!IöI!IöI“möI€“möI1®öIü1®öIó²öIó²öIe×öIpe×öI÷Iì÷IÅ÷IôÅ÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöI…KöI…KöIÿmöI€ÿmöI�®öIü�®öIWµöIWµöIÑ×öIpÑ×öIo÷Iìo÷I)÷Iô)÷I_÷I`_÷Iƒ÷Ĩ÷Ig¶÷Iìg¶÷I3øIL 3øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI³ûIT ³ûIƒØûI˜ƒØûI+ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI»öI�»öIYDöI YDöI÷HöI÷HöI�möI€�möI+®öIü+®öIɲöIɲöI_×öIp_×öIý÷Iìý÷I›÷Iô›÷I«`÷I`«`÷I …÷IÌ …÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI3öI�3öIÑDöI ÑDöI»KöI»KöInöI€nöI£®öIü£®öI�µöI�µöI××öIp××öIu÷Iìu÷I_÷Iô_÷I_÷I`_÷Iƒ÷Ĩ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIsûIT sûICÙûI˜CÙûIëëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI»öI�»öIYDöI YDöI÷HöI÷HöI‡möI€‡möI%®öIü%®öIɲöIɲöIY×öIpY×öI÷÷Iì÷÷I›÷Iô›÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ia¶÷Iìa¶÷I-øIL -øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI3öI�3öIÑDöI ÑDöIµKöIµKöInöI€nöI£®öIü£®öI‡µöI‡µöI××öIp××öIu÷Iìu÷IY÷IôY÷I_÷I`_÷Iƒ÷Ĩ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIµöI�µöISDöI SDöIIöIIöI‡möI€‡möI%®öIü%®öIç²öIç²öIY×öIpY×öI÷÷Iì÷÷I¹÷Iô¹÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ia¶÷Iìa¶÷I-øIL -øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöIyKöIyKöI nöI€ nöI©®öIü©®öIKµöIKµöIÝ×öIpÝ×öI{÷Iì{÷I÷Iô÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIyûIT yûIIÙûI˜IÙûIñëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIµöI�µöISDöI SDöIIöIIöI�möI€�möI®öIü®öIç²öIç²öIS×öIpS×öIñ÷Iìñ÷I¹÷Iô¹÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ia¶÷Iìa¶÷I-øIL -øIÝ¢øI¬ Ý¢øIµxùIp µxùIïdúIt ïdúI_ûIT _ûI/ØûI˜/ØûI×êûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LoïõI¯öI�¯öIMDöI MDöIëHöIëHöI�möI€�möI®öIü®öI½²öI½²öIS×öIpS×öIñ÷Iìñ÷I�÷Iô�÷I_÷I`_÷Isƒ÷IÌsƒ÷I[¶÷Iì[¶÷I'øIL 'øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LoïõI?öI�?öIÝDöI ÝDöI¯KöI¯KöInöI€nöI¯®öIü¯®öI�µöI�µöIã×öIpã×öI�÷Iì�÷IS÷IôS÷IÉ`÷I`É`÷I)…÷IÔ)…÷I[¶÷Iô[¶÷I'øIT 'øI £øI´  £øI5wùIx 5wùIGúI| GúICûI\ CûI9×ûI 9×ûIáéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI©öI�©öIGDöI GDöIëHöIëHöI{möI€{möI®öIü®öI½²öI½²öIM×öIpM×öIë÷Iìë÷I�÷Iô�÷I_÷I`_÷Isƒ÷IÌsƒ÷I[¶÷Iì[¶÷I'øIL 'øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúI¿ûIT ¿ûI�ØûI˜�ØûI7ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LoïõI?öI�?öIÝDöI ÝDöI¯KöI¯KöInöI€nöI¯®öIü¯®öI�µöI�µöIã×öIpã×öI�÷Iì�÷IS÷IôS÷Ia÷I`a÷Iq…÷IÔq…÷I[¶÷Iô[¶÷I'øIT 'øIõ¢øI´ õ¢øIwùIx wùIûFúI| ûFúI×ûI\ ×ûIÍÖûI ÍÖûIuéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõIEöI�EöIãDöI ãDöIsKöIsKöInöI€nöI¯®öIü¯®öIEµöIEµöIã×öIpã×öI�÷Iì�÷I÷Iô÷Iï^÷I`ï^÷IOƒ÷IÌOƒ÷IU¶÷IìU¶÷I!øIL !øI£øI¬ £øI/wùIp /wùI GúIt  GúI%ûIT %ûI×ûI˜×ûIÃéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI£öI�£öIADöI ADöI IöI IöIumöI€umöI®öIü®öIÛ²öIÛ²öIG×öIpG×öIå÷Iìå÷I­÷Iô­÷IÕ`÷I`Õ`÷I5…÷IÔ5…÷IU¶÷IôU¶÷I!øIT !øIû¢øI´ û¢øI#wùIx #wùIGúI| GúIÝûI\ ÝûIÓÖûI ÓÖûI{éûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõIEöI�EöIãDöI ãDöIsKöIsKöInöI€nöI¯®öIü¯®öIEµöIEµöIã×öIpã×öI�÷Iì�÷I÷Iô÷Iï^÷I`ï^÷IOƒ÷IÌOƒ÷IU¶÷IìU¶÷I!øIL !øIï¢øI¬ ï¢øIwùIp wùIõFúIt õFúI­ûIT ­ûI£ÖûI˜£ÖûIKéûIœ€} 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$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„cïõIöIöIDöI|DöI�IöI„�IöI·LöIð·LöIAVöI|AVöIwYöIHwYöIQmöIÔQmöIï­öIPï­öIS³öIXS³öI‰¶öIĉ¶öIÀöIPÀöIIÃöIIÃöI#×öI¨#×öIÁ÷I$ Á÷I%÷I, %÷I[ ÷I˜ [ ÷Iå)÷I$ å)÷I-÷Ið -÷I]Z÷I| ]Z÷IÕ`÷IXÕ`÷I5…÷IÔ5…÷Iw²÷Iôw²÷I?À÷IÐ?À÷Iûô÷I\ûô÷I›øI€›øI™øI¬™øI£øI8£øIñxùIüñxùI+eúI+eúI¥rúIà¥rúIûIlûIOÙûI°OÙûI÷ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxcïõIQöIèQöIïDöIdïDöI—KöIl—KöI!UöIè!UöIWXöI´WXöI#nöI@#nöIÁ®öI¼Á®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIõ×öI˜õ×öI“÷I“÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷I±`÷IÌ ±`÷I…÷IH …÷IS²÷IhS²÷I{À÷ID{À÷I7õ÷IÐ7õ÷IÇøIôÇøIÅøI ÅøI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúI1ûIà1ûIÙûI$ÙûI©ëûI(€} 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„cïõIöIöIDöI|DöI�IöI„�IöI·LöIð·LöIAVöI|AVöIwYöIHwYöIKmöIÔKmöIé­öIPé­öIS³öIXS³öI‰¶öIĉ¶öIÀöIPÀöIIÃöIIÃöI×öI¨×öI»÷I$ »÷I%÷I, %÷I[ ÷I˜ [ ÷Iå)÷I$ å)÷I-÷Ið -÷I]Z÷I| ]Z÷IŸ`÷IXŸ`÷Iÿ„÷IÔÿ„÷IA²÷IôA²÷IÉÀ÷IÐÉÀ÷I…õ÷I\…õ÷IYøI€YøIW øI¬W øI£øI8£øIÙxùIüÙxùIeúIeúI�rúIà�rúIûIlûIרûI°רûIëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxcïõIWöIèWöIõDöIdõDöI—KöIl—KöI!UöIè!UöIWXöI´WXöI#nöI@#nöIÁ®öI¼Á®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIõ×öI˜õ×öI“÷I“÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷IÏ`÷IÌ Ï`÷I/…÷IH /…÷Iq²÷Ihq²÷IŸÀ÷IDŸÀ÷I[õ÷IÐ[õ÷I{øIô{øIyøI yøIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúI×ûIà×ûI§ØûI$§ØûIOëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI©öIè©öIGDöIdGDöI£KöIl£KöI-UöIè-UöIcXöI´cXöIumöI@umöI®öI¼®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIG×öI˜G×öIå÷Iå÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷IÛ`÷IÌ Û`÷I;…÷IH ;…÷I}²÷Ih}²÷IEÀ÷IDEÀ÷Iõ÷IÐõ÷I§øIô§øI¥øI ¥øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúI=ûIà=ûI ÙûI$ ÙûIµëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIKöIèKöIéDöIdéDöI£KöIl£KöI-UöIè-UöIcXöI´cXöInöI@nöI»®öI¼»®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIï×öI˜ï×öI�÷I�÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷IÉ`÷IÌ É`÷I)…÷IH )…÷Ik²÷Ihk²÷Iuº÷IDuº÷Iþ÷IÀþ÷I#øIä#øI!øIÀ!øIÝ¢øILÝ¢øIµxùIµxùIïdúIïdúIirúIôirúIûI€ûIãØûIÄãØûI‹ëûIÈ€} 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;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn@o@q´q´s(t(vœvœxy{„{„}ø}øl€l‚à‚à„T…T‡¬‡¬‰ Š Œ”Œ”Ž�‘|‘|“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxiïõI£öIè£öIADöIdADöI£KöIl£KöI-UöIè-UöIcXöI´cXöIumöI@umöI®öI¼®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIG×öI˜G×öIå÷Iå÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I¥`÷IÌ ¥`÷I…÷IH …÷IG²÷IhG²÷IEÀ÷IDEÀ÷Iõ÷IÐõ÷I¡øIô¡øIŸøI ŸøIï¢øI¬ï¢øIÇxùIpÇxùIeúIteúI{rúIT{rúIÑûIàÑûI¡ØûI$¡ØûIIëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIKöIèKöIéDöIdéDöI©KöIl©KöI3UöIè3UöIiXöI´iXöInöI@nöI»®öI¼»®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIï×öI˜ï×öI�÷I�÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷IM»÷IDM»÷I÷þ÷IÀ÷þ÷I­øIä­øI«øIÀ«øIÅ¢øILÅ¢øI�xùI�xùI×dúI×dúIQrúIôQrúI¡ûI€¡ûIqØûIÄqØûIëûIÈ€} 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·loïõIEöIÐEöIãDöILãDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöIµ®öI¤µ®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIé×öI€é×öI‡÷Iü‡÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I¿¸÷I,¿¸÷IùøI˜ùøI颸I$颸IÁxùIèÁxùIûdúIìûdúI¿ûIÌ¿ûI�ØûI�ØûI7ëûI€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1‡»X$ 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI©öIè©öIGDöIdGDöIµKöIlµKöI?UöIè?UöIuXöI´uXöI{möI@{möI®öI¼®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöIM×öI˜M×öIë÷Ië÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷IŸ`÷IÌ Ÿ`÷Iÿ„÷IH ÿ„÷IA²÷IhA²÷IA»÷IDA»÷Iëþ÷IÀëþ÷I›øI䛸I™øIÀ™øI¿¢øIL¿¢øI—xùI—xùIÑdúIÑdúIKrúIôKrúI}ûI€}ûIMØûIÄMØûIõêûIÈ€} 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·loïõIEöIÐEöIãDöILãDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöIµ®öI¤µ®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIé×öI€é×öI‡÷Iü‡÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IA»÷I,A»÷I}øI˜}øI×¢øI$×¢øI¯xùIè¯xùIédúIìédúIYûIÌYûI)ØûI)ØûIÑêûI€} 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·loïõI¯öIЯöIMDöILMDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI�möI(�möI®öI¤®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIS×öI€S×öIñ÷Iüñ÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I³¸÷I,³¸÷I/øI˜/øI £øI$ £øIåxùIèåxùIeúIìeúIgûIÌgûI7ÙûI7ÙûIßëûI€} 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·loïõI?öIÐ?öIÝDöILÝDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI nöI( nöI©®öI¤©®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIÝ×öI€Ý×öI{÷Iü{÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷I³¸÷I,³¸÷IÏøI˜ÏøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIûIÌûIרûIרûIëûI€} 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·loïõI¯öIЯöIMDöILMDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI�möI(�möI®öI¤®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIS×öI€S×öIñ÷Iüñ÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷I;»÷I,;»÷I¹øI˜¹øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIõûIÌõûIÅØûIÅØûImëûI€} 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·loïõI?öIÐ?öIÝDöILÝDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI¯®öI¤¯®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIã×öI€ã×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷I;»÷I,;»÷ISøI˜SøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIÅûIÌÅûI•ØûI•ØûI=ëûI€} 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·luïõIµöIеöISDöILSDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI‡möI(‡möI%®öI¤%®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIY×öI€Y×öI÷÷Iü÷÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I­¸÷I,­¸÷IGøI˜GøI£øI$£øIëxùIèëxùI%eúIì%eúIyûIÌyûIIÙûIIÙûIñëûI€} 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·luïõI9öIÐ9öI×DöIL×DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI£®öI¤£®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI××öI€××öIu÷Iüu÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷Iu`÷I´ u`÷IÕ„÷I0 Õ„÷I²÷IP²÷I­¸÷I,­¸÷I¥øI˜¥øI£øI$£øIßxùIèßxùIeúIìeúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·luïõIµöIеöISDöILSDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI�möI(�möI®öI¤®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIS×öI€S×öIñ÷Iüñ÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I;»÷I,;»÷I×øI˜×øI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIרûIרûIëûI€} 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·luïõI9öIÐ9öI×DöIL×DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI nöI( nöI©®öI¤©®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIÝ×öI€Ý×öI{÷Iü{÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I/»÷I,/»÷I)øI˜)øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1r¦D b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 g§Åä$ 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$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI»öIлöIYDöILYDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI+®öI¤+®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöI_×öI€_×öIý÷Iüý÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I§¸÷I,§¸÷I•øI˜•øI £øI$ £øIåxùIèåxùIeúIìeúICûIÌCûIÙûIÙûI»ëûI€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n0o0q¥q¥stv�v�xy{y{y}î}îc€c‚؂؄M…M‡‡‰7Š7Œ¬Œ¬Ž!�!‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI-öIÐ-öIËDöILËDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIÿmöI(ÿmöI�®öI¤�®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIÑ×öI€Ñ×öIo÷Iüo÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷I§¸÷I,§¸÷IóøI˜óøI£øI$£øIëxùIèëxùI%eúIì%eúIIûIÌIûIÙûIÙûIÁëûI€} 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·l{ïõI»öIлöIYDöILYDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI‡möI(‡möI%®öI¤%®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIY×öI€Y×öI÷÷Iü÷÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I;»÷I,;»÷I%øI˜%øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l{ïõI3öIÐ3öIÑDöILÑDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIÿmöI(ÿmöI�®öI¤�®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIÑ×öI€Ñ×öIo÷Iüo÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷I5»÷I,5»÷I}øI˜}øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$l%l'à'à)T*T,È,È.Œ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÁöIÐÁöI_DöIL_DöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI�möI(�möI+®öI¤+®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöI_×öI€_×öIý÷Iüý÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷I­¸÷I,­¸÷IIøI˜IøI £øI$ £øIåxùIèåxùIeúIìeúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI'öIÐ'öIÅDöILÅDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIùmöI(ùmöI—®öI¤—®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIË×öI€Ë×öIi÷Iüi÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I­¸÷I,­¸÷I§øI˜§øI£øI$£øIÙxùIèÙxùIeúIìeúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'bšbšd e g€g€ióiókflfnÙnÙpLqLs¿s¿u2v2x¥x¥z{}‹}‹þþ�q‚q„ä„ä†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI»öIлöIYDöILYDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI�möI(�möI+®öI¤+®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI_×öI€_×öIý÷Iüý÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I]`÷I´ ]`÷I½„÷I0 ½„÷Iÿ±÷IPÿ±÷IG»÷I,G»÷IåøI˜åøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI³ûI̳ûIƒØûIƒØûI+ëûI€} 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$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ;‡;‰¬‰¬‹ŒŽŽŽŽ�ÿ�ÿ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI-öIÐ-öIËDöILËDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIùmöI(ùmöI—®öI¤—®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIË×öI€Ë×öIi÷Iüi÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IA»÷I,A»÷I=øI˜=øI颸I$颸IÁxùIèÁxùIûdúIìûdúIkûIÌkûI;ØûI;ØûIãêûI€} 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�ïõIÇöIÐÇöIeDöILeDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI“möI(“möI1®öI¤1®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIe×öI€e×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Iu`÷I´ u`÷IÕ„÷I0 Õ„÷I²÷IP²÷I³¸÷I,³¸÷IýøI˜ýøI£øI$£øI÷xùIè÷xùI1eúIì1eúIûIÌûIOÙûIOÙûI÷ëûI€} 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�ïõI!öIÐ!öI¿DöIL¿DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIómöI(ómöI‘®öI¤‘®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIÅ×öI€Å×öIc÷Iüc÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I³¸÷I,³¸÷I[øI˜[øI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIïØûIïØûI—ëûI€} 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�ïõI!öIÐ!öI¿DöIL¿DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIómöI(ómöI‘®öI¤‘®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIÅ×öI€Å×öIc÷Iüc÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IS»÷I,S»÷IýøI˜ýøI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI-öI�-öIËDöIËDöI÷HöI$÷HöIùmöI�ùmöI—®öI—®öIɲöI$ɲöIË×öI�Ë×öIi÷Ii÷I›÷I$›÷I1_÷I�1_÷I‘ƒ÷Iü‘ƒ÷Is¶÷I s¶÷I?øI| ?øI=£øIÜ =£øIyùI  yùIOeúI¤ OeúI£ûI„ £ûIsÙûIÈsÙûIìûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¹DöI ¹DöIÁKöIÁKöIímöI€ímöI‹®öIü‹®öI“µöI“µöI¿×öIp¿×öI]÷Iì]÷Ie÷Iôe÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI-öI�-öIËDöIËDöI÷HöI$÷HöIÿmöI�ÿmöI�®öI�®öIɲöI$ɲöIÑ×öI�Ñ×öIo÷Io÷I›÷I$›÷I1_÷I�1_÷I‘ƒ÷Iü‘ƒ÷Is¶÷I s¶÷I?øI| ?øI%£øIÜ %£øIýxùI  ýxùI7eúI¤ 7eúI+ûI„ +ûIûØûIÈûØûI£ëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¹DöI ¹DöIÁKöIÁKöIímöI€ímöI‹®öIü‹®öI“µöI“µöI¿×öIp¿×öI]÷Iì]÷Ie÷Iôe÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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‡ïõIÙöI�ÙöIwDöI wDöIIöIIöI¥möI€¥möIC®öIüC®öIá²öIá²öIw×öIpw×öI÷Iì÷I³÷Iô³÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI+£øI¬ +£øIyùIp yùI=eúIt =eúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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‡ïõI!öI�!öI¿DöI ¿DöI—KöI—KöIómöI€ómöI‘®öIü‘®öIiµöIiµöIÅ×öIpÅ×öIc÷Iìc÷I;÷Iô;÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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‡ïõIÓöI�ÓöIqDöI qDöIIöIIöI¥möI€¥möIC®öIüC®öIá²öIá²öIw×öIpw×öI÷Iì÷I³÷Iô³÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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‡ïõI!öI�!öI¿DöI ¿DöI�KöI�KöIómöI€ómöI‘®öIü‘®öIoµöIoµöIÅ×öIpÅ×öIc÷Iìc÷IA÷IôA÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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�ïõIÓöI�ÓöIqDöI qDöI÷HöI÷HöIŸmöI€ŸmöI=®öIü=®öIɲöIɲöIq×öIpq×öI÷Iì÷I›÷Iô›÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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�ïõI'öI�'öIÅDöI ÅDöIÁKöIÁKöIùmöI€ùmöI—®öIü—®öI“µöI“µöIË×öIpË×öIi÷Iìi÷Ie÷Iôe÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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�ïõIÍöI�ÍöIkDöI kDöI÷HöI÷HöIŸmöI€ŸmöI=®öIü=®öIɲöIɲöIq×öIpq×öI÷Iì÷I›÷Iô›÷IÉ`÷I`É`÷I)…÷IÌ)…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI¿ûIT ¿ûI�ØûI˜�ØûI7ëûIœ€} 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�ïõI'öI�'öIÅDöI ÅDöI»KöI»KöIùmöI€ùmöI—®öIü—®öI�µöI�µöIË×öIpË×öIi÷Iìi÷I_÷Iô_÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Im¶÷Iìm¶÷I9øIL 9øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIwûIT wûIGØûI˜GØûIïêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÇöI�ÇöIeDöI eDöIIöIIöI™möI€™möI7®öIü7®öIí²öIí²öIk×öIpk×öI ÷Iì ÷I¿÷Iô¿÷I·`÷I`·`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIOÙûI˜OÙûI÷ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöI…KöI…KöIÿmöI€ÿmöI�®öIü�®öIWµöIWµöIÑ×öIpÑ×öIo÷Iìo÷I)÷Iô)÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÇöI�ÇöIeDöI eDöI!IöI!IöI“möI€“möI1®öIü1®öIó²öIó²öIe×öIpe×öI÷Iì÷IÅ÷IôÅ÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöI…KöI…KöIÿmöI€ÿmöI�®öIü�®öIWµöIWµöIÑ×öIpÑ×öIo÷Iìo÷I)÷Iô)÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÁöI�ÁöI_DöI _DöI÷HöI÷HöI“möI€“möI1®öIü1®öIɲöIɲöIe×öIpe×öI÷Iì÷I›÷Iô›÷I±`÷I`±`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI…ûIT …ûIUÙûI˜UÙûIýëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÁöI�ÁöI_DöI _DöI÷HöI÷HöI�möI€�möI+®öIü+®öIɲöIɲöI_×öIp_×öIý÷Iìý÷I›÷Iô›÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIßxùIp ßxùIeúIt eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI»öI�»öIYDöI YDöIIöIIöI�möI€�möI+®öIü+®öIç²öIç²öI_×öIp_×öIý÷Iìý÷I¹÷Iô¹÷I¥`÷I`¥`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI £øI¬  £øIåxùIp åxùIeúIt eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöIyKöIyKöInöI€nöI£®öIü£®öIKµöIKµöI××öIp××öIu÷Iìu÷I÷Iô÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIßxùIp ßxùIeúIt eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIµöI�µöISDöI SDöIIöIIöI‡möI€‡möI%®öIü%®öIç²öIç²öIY×öIpY×öI÷÷Iì÷÷I¹÷Iô¹÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ia¶÷Iìa¶÷I-øIL -øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöIyKöIyKöInöI€nöI£®öIü£®öIKµöIKµöI××öIp××öIu÷Iìu÷I÷Iô÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ia¶÷Iìa¶÷I-øIL -øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LoïõIµöI�µöISDöI SDöIëHöIëHöI‡möI€‡möI%®öIü%®öI½²öI½²öIY×öIpY×öI÷÷Iì÷÷I�÷Iô�÷I_÷I`_÷Iyƒ÷IÌyƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI¯öI�¯öIMDöI MDöIëHöIëHöI�möI€�möI®öIü®öI½²öI½²öIS×öIpS×öIñ÷Iìñ÷I�÷Iô�÷I_÷I`_÷Iyƒ÷IÌyƒ÷I[¶÷Iì[¶÷I'øIL 'øI颸I¬ 颸IÁxùIp ÁxùIûdúIt ûdúI§ûIT §ûIwØûI˜wØûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LoïõI9öI�9öI×DöI ×DöI©KöI©KöI nöI€ nöI©®öIü©®öI{µöI{µöIÝ×öIpÝ×öI{÷Iì{÷IM÷IôM÷I{`÷I`{`÷IÛ„÷IÌÛ„÷I[¶÷Iì[¶÷I'øIL 'øIÝ¢øI¬ Ý¢øIwùIp wùIãFúIt ãFúI_ûIT _ûIUÖûI˜UÖûIýèûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI¯öI�¯öIMDöI MDöI IöI IöI{möI€{möI®öIü®öIÛ²öIÛ²öIM×öIpM×öIë÷Iìë÷I­÷Iô­÷I¥`÷I`¥`÷I…÷IÔ…÷IU¶÷IôU¶÷I!øIT !øI£øI´ £øI;wùIx ;wùIGúI| GúImûI\ mûIc×ûI c×ûI êûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI?öI�?öIÝDöI ÝDöIsKöIsKöInöI€nöI¯®öIü¯®öIEµöIEµöIã×öIpã×öI�÷Iì�÷I÷Iô÷Iï^÷I`ï^÷IOƒ÷IÌOƒ÷IU¶÷IìU¶÷I!øIL !øI£øI¬ £øI/wùIp /wùI GúIt  GúI ûIT  ûI×ûI˜×ûI«éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI©öI�©öIGDöI GDöI IöI IöI{möI€{möI®öIü®öIÛ²öIÛ²öIM×öIpM×öIë÷Iìë÷I­÷Iô­÷IÉ`÷I`É`÷I)…÷IÔ)…÷IU¶÷IôU¶÷I!øIT !øI£øI´ £øI)wùIx )wùIGúI| GúIûI\ ûI÷ÖûI ÷ÖûIŸéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI?öI�?öIÝDöI ÝDöIsKöIsKöInöI€nöI¯®öIü¯®öIEµöIEµöIã×öIpã×öI�÷Iì�÷I÷Iô÷Iï^÷I`ï^÷IOƒ÷IÌOƒ÷IU¶÷IìU¶÷I!øIL !øIõ¢øI¬ õ¢øIwùIp wùIûFúIt ûFúIËûIT ËûIÁÖûI˜ÁÖûIiéûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIQöIèQöIïDöIdïDöI—KöIl—KöI!UöIè!UöIWXöI´WXöInöI@nöI»®öI¼»®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIï×öI˜ï×öI�÷I�÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷I½`÷IÌ ½`÷I…÷IH …÷I_²÷Ih_²÷IQÀ÷IDQÀ÷I õ÷IÐ õ÷I‡øIô‡øI…øI …øI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúIyûIàyûIIÙûI$IÙûIñëûI(€} 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$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=ÿ=ÿ?w@wBïBïDgEgGßGßIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélénkokqáqásWtWvÍvÍxCyC{¹{¹}/~/€¥€¥‚ƒ…‘…‘‡ˆŠ}Š}ŒóŒóŽi�i‘ߑߓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„iïõIöIöIDöI|DöI‡IöI„‡IöI½LöIð½LöIGVöI|GVöI}YöIH}YöIQmöIÔQmöIï­öIPï­öIY³öIXY³öI�¶öIÄ�¶öIÀöIPÀöIOÃöIOÃöI#×öI¨#×öIÁ÷I$ Á÷I+÷I, +÷Ia ÷I˜ a ÷Ië)÷I$ ë)÷I!-÷Ið !-÷IcZ÷I| cZ÷IÏ`÷IXÏ`÷I/…÷IÔ/…÷Iq²÷Iôq²÷IùÀ÷IÐùÀ÷Iµõ÷I\µõ÷I1øI€1øI/!øI¬/!øIõ¢øI8õ¢øIÍxùIüÍxùIeúIeúI�rúIà�rúI×ûIl×ûI§ØûI°§ØûIOëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIQöIèQöIïDöIdïDöI—KöIl—KöI!UöIè!UöIWXöI´WXöI#nöI@#nöIÁ®öI¼Á®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIõ×öI˜õ×öI“÷I“÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷IÉ`÷IÌ É`÷I)…÷IH )…÷Ik²÷Ihk²÷I¥À÷ID¥À÷Iaõ÷IÐaõ÷I#øIô#øI! øI ! øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIûIàûIרûI$רûIëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI©öIè©öIGDöIdGDöI£KöIl£KöI-UöIè-UöIcXöI´cXöI{möI@{möI®öI¼®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIM×öI˜M×öIë÷Ië÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I±`÷IÌ ±`÷I…÷IH …÷IS²÷IhS²÷IKÀ÷IDKÀ÷Iõ÷IÐõ÷IøIôøIøI øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúI=ûIà=ûI ÙûI$ ÙûIµëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIKöIèKöIéDöIdéDöI©KöIl©KöI3UöIè3UöIiXöI´iXöInöI@nöIµ®öI¼µ®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIé×öI˜é×öI‡÷I‡÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷IÏ`÷IÌ Ï`÷I/…÷IH /…÷Iq²÷Ihq²÷I¾÷ID¾÷I¯øIÀ¯øIcøIäcøIaøIÀaøIÝ¢øILÝ¢øIµxùIµxùIïdúIïdúIirúIôirúIûI€ûIãØûIÄãØûI‹ëûIÈ€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI©öIè©öIGDöIdGDöI£KöIl£KöI-UöIè-UöIcXöI´cXöI{möI@{möI®öI¼®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIM×öI˜M×öIë÷Ië÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷IQÀ÷IDQÀ÷I õ÷IÐ õ÷I‹øIô‹øI‰øI ‰øIï¢øI¬ï¢øIÇxùIpÇxùIeúIteúI{rúIT{rúIÅûIàÅûI•ØûI$•ØûI=ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIKöIèKöIéDöIdéDöI©KöIl©KöI3UöIè3UöIiXöI´iXöInöI@nöI»®öI¼»®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIï×öI˜ï×öI�÷I�÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷IÏ`÷IÌ Ï`÷I/…÷IH /…÷Iq²÷Ihq²÷IŸÀ÷IDŸÀ÷IIøIÀIøIÿøIäÿøIýøIÀýøIË¢øILË¢øI£xùI£xùIÝdúIÝdúIWrúIôWrúI­ûI€­ûI}ØûIÄ}ØûI%ëûIÈ€} 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·loïõI?öIÐ?öIÝDöILÝDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöI¯®öI¤¯®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIã×öI€ã×öI�÷Iü�÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷Iù½÷I,ù½÷I3øI˜3øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIûIÌûIéØûIéØûI‘ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI¯öIè¯öIMDöIdMDöIµKöIlµKöI?UöIè?UöIuXöI´uXöI�möI@�möI®öI¼®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöIS×öI˜S×öIñ÷Iñ÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷IÛ`÷IÌ Û`÷I;…÷IH ;…÷I}²÷Ih}²÷IQÀ÷IDQÀ÷IûøIÀûøI¯øI䝸I­øIÀ­øI³¢øIL³¢øI‹xùI‹xùIÅdúIÅdúI?rúIô?rúI5ûI€5ûIØûIÄØûI­êûIÈ€} 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$9%9'¬'¬)*,’,’./1x1x3ë3ë5^6^8Ñ8Ñ:D;D=·=·?(@(B™B™D E G{G{IìIìK]L]NÎNÎP?Q?S°S°U!V!X’X’Z[]t]t_å_åaVbVdÇdÇf8g8i©i©kln‹n‹püpürmsmuÞuÞwOxOzÀzÀ|1}1¢¢�‚„„„„†Ö†ÖˆF‰F‹¶‹¶�&Ž&�–�–’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·loïõIEöIÐEöIãDöILãDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöI¯®öI¤¯®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIã×öI€ã×öI�÷Iü�÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I“À÷I,“À÷IÏøI˜ÏøIã¢øI$ã¢øI»xùIè»xùIõdúIìõdúI¡ûIÌ¡ûIqØûIqØûIëûI€} 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$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIµöIеöISDöILSDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI‡möI(‡möI%®öI¤%®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIY×öI€Y×öI÷÷Iü÷÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷Iɽ÷I,ɽ÷ICøI˜CøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIûIÌûIרûIרûIëûI€} 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·luïõI9öIÐ9öI×DöIL×DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI nöI( nöI©®öI¤©®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIÝ×öI€Ý×öI{÷Iü{÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷Ií½÷I,í½÷I øI˜ øI£øI$£øIëxùIèëxùI%eúIì%eúImûIÌmûI=ÙûI=ÙûIåëûI€} 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·luïõIµöIеöISDöILSDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI�möI(�möI®öI¤®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIS×öI€S×öIñ÷Iüñ÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IQÀ÷I,QÀ÷IÍøI˜ÍøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI¿ûIÌ¿ûI�ØûI�ØûI7ëûI€} 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$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,bžbždeg‚g‚iôiôkflfnØnØpJqJs¼s¼u.v.x x z{}„}„öö�h‚h„Ú„Ú†N‡N‰Á‰Á‹4Œ4ާާ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI?öIÐ?öIÝDöILÝDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI nöI( nöI©®öI¤©®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIÝ×öI€Ý×öI{÷Iü{÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I�À÷I,�À÷IŸøI˜ŸøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIûûIÌûûIËØûIËØûIsëûI€} 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$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI»öIлöIYDöILYDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI�möI(�möI+®öI¤+®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI_×öI€_×öIý÷Iüý÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷Iý÷I,ý÷IaøI˜aøI£øI$£øIßxùIèßxùIeúIìeúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnJoJq¼q¼s.t.v v xy{„{„}ö}öh€h‚Ú‚Ú„L…L‡¾‡¾‰0Š0Œ¢Œ¢Ž�‘†‘†“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI3öIÐ3öIÑDöILÑDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI£®öI¤£®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI××öI€××öIu÷Iüu÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷Iá½÷I,á½÷IÙøI˜ÙøI£øI$£øIñxùIèñxùI+eúIì+eúIûIÌûIOÙûIOÙûI÷ëûI€} 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��  "÷"÷$o%o'ç'ç)_*_,×,×.O/O1Ç1Ç3?4?6·6·8/9/;§;§=>@—@—BCE‡E‡GÿGÿIwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënaoaq×q×sMtMvÃvÃx9y9{¯{¯}%~%€›€›‚ƒ…‡…‡‡ý‡ý‰sŠsŒéŒéŽ_�_‘Õ‘Õ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI»öIлöIYDöILYDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI‡möI(‡möI%®öI¤%®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIY×öI€Y×öI÷÷Iü÷÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷IQÀ÷I,QÀ÷IñøI˜ñøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI3öIÐ3öIÑDöILÑDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI£®öI¤£®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI××öI€××öIu÷Iüu÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IuÀ÷I,uÀ÷IoøI˜oøI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIרûIרûIëûI€} 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$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÁöIÐÁöI_DöIL_DöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI+®öI¤+®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöI_×öI€_×öIý÷Iüý÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷IϽ÷I,Ͻ÷I»øI˜»øI £øI$ £øIåxùIèåxùIeúIìeúICûIÌCûIÙûIÙûI»ëûI€} 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;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn>o>q²q²s&t&všvšxy{‚{‚}ö}öj€j‚Þ‚Þ„R…R‡ƇƉ:Š:Œ®Œ®Ž"�"‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI-öIÐ-öIËDöILËDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIÿmöI(ÿmöI�®öI¤�®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIÑ×öI€Ñ×öIo÷Iüo÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷IÕ½÷I,Õ½÷I¯øI˜¯øI£øI$£øIëxùIèëxùI%eúIì%eúIIûIÌIûIÙûIÙûIÁëûI€} 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$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâIYJYLÎLÎNCOCQ¸Q¸S-T-V¢V¢XY[Œ[Œ]^`v`vbëbëd`e`gÕgÕiJjJl¿l¿n4o4q©q©stv“v“xy{}{}}ò}òg€g‚܂܄Q…Q‡ƇƉ;Š;Œ°Œ°Ž%�%‘š‘š“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI»öIлöIYDöILYDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI+®öI¤+®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöI_×öI€_×öIý÷Iüý÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I]À÷I,]À÷IKøI˜KøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI6J6L¨L¨NOQŒQŒSþSþUpVpXâXâZT[T]Æ]Æ_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wºoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z9z="=zA"AzE*’ªreasonOperatorKernelvalue10kernelSpecoutputElements576input0Elements640input1Elements62722Planned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })binvocations1·l{ïõI-öIÐ-öIËDöILËDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIÿmöI(ÿmöI�®öI¤�®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIÑ×öI€Ñ×öIo÷Iüo÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷IcÀ÷I,cÀ÷I?øI˜?øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIÇöIÐÇöIeDöILeDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI“möI(“möI1®öI¤1®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIe×öI€e×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷Iç½÷I,ç½÷I�øI˜�øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI'öIÐ'öIÅDöILÅDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIómöI(ómöI‘®öI¤‘®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÅ×öI€Å×öIc÷Iüc÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷Iá½÷I,á½÷IøI˜øI£øI$£øIëxùIèëxùI%eúIì%eúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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·l�ïõIÁöIÐÁöI_DöIL_DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI“möI(“möI1®öI¤1®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIe×öI€e×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷I‡À÷I,‡À÷I#øI˜#øI颸I$颸IÁxùIèÁxùIûdúIìûdúIkûIÌkûI;ØûI;ØûIãêûI€} 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$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=á=á?S@SBÅBÅD7E7G©G©IJL�L�NÿNÿPqQqSãSãUUVUXÇXÇZ9[9]«]«_`b�b�degsgsiåiåkWlWnÉnÉp;q;s­s­uvx‘x‘z{}u}uçç�Y‚Y„˄ˆ=‡=‰®‰®‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIÇöIÐÇöIeDöILeDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI™möI(™möI7®öI¤7®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIk×öI€k×öI ÷Iü ÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷Iÿ½÷I,ÿ½÷IMøI˜MøI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIéØûIéØûI‘ëûI€} 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$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI!öIÐ!öI¿DöIL¿DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIímöI(ímöI‹®öI¤‹®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI¿×öI€¿×öI]÷Iü]÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷Iù½÷I,ù½÷IçøI˜çøI%£øI$%£øIýxùIèýxùI7eúIì7eúIûIÌûIOÙûIOÙûI÷ëûI€} 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�ïõIÇöIÐÇöIeDöILeDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI™möI(™möI7®öI¤7®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIk×öI€k×öI ÷Iü ÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷I«À÷I,«À÷IõøI˜õøI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI!öIÐ!öI¿DöIL¿DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIómöI(ómöI‘®öI¤‘®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIÅ×öI€Å×öIc÷Iüc÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I™À÷I,™À÷IƒøI˜ƒøI £øI$ £øIåxùIèåxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡“‡“‰ŠŒuŒu޿޿�W‘W“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöIÅDöI÷HöI$÷HöIómöI�ómöI‘®öI‘®öIɲöI$ɲöIÅ×öI�Å×öIc÷Ic÷I›÷I$›÷I1_÷I�1_÷I‘ƒ÷Iü‘ƒ÷Is¶÷I s¶÷I?øI| ?øI+£øIÜ +£øIyùI  yùI=eúI¤ =eúI[ûI„ [ûI+ÙûIÈ+ÙûIÓëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöI»KöI»KöIímöI€ímöI‹®öIü‹®öI�µöI�µöI¿×öIp¿×öI]÷Iì]÷I_÷Iô_÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Is¶÷Iìs¶÷I?øIL ?øI7£øI¬ 7£øIyùIp yùIIeúIt IeúI�ûIT �ûImÙûI˜mÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöIÅDöI÷HöI$÷HöIùmöI�ùmöI—®öI—®öIɲöI$ɲöIË×öI�Ë×öIi÷Ii÷I›÷I$›÷I1_÷I�1_÷I‘ƒ÷Iü‘ƒ÷Is¶÷I s¶÷I?øI| ?øI£øIÜ £øIñxùI  ñxùI+eúI¤ +eúIûI„ ûIÑØûIÈÑØûIyëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöI»KöI»KöIímöI€ímöI‹®öIü‹®öI�µöI�µöI¿×öIp¿×öI]÷Iì]÷I_÷Iô_÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIßöI�ßöI}DöI }DöI IöI IöI«möI€«möII®öIüI®öIÛ²öIÛ²öI}×öIp}×öI÷Iì÷I­÷Iô­÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI!öI�!öI¿DöI ¿DöI—KöI—KöIímöI€ímöI‹®öIü‹®öIiµöIiµöI¿×öIp¿×öI]÷Iì]÷I;÷Iô;÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIÙöI�ÙöIwDöI wDöI IöI IöI«möI€«möII®öIüI®öIÛ²öIÛ²öI}×öIp}×öI÷Iì÷I­÷Iô­÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI!öI�!öI¿DöI ¿DöI—KöI—KöIímöI€ímöI‹®öIü‹®öIiµöIiµöI¿×öIp¿×öI]÷Iì]÷I;÷Iô;÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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�ïõIÓöI�ÓöIqDöI qDöI÷HöI÷HöI¥möI€¥möIC®öIüC®öIɲöIɲöIw×öIpw×öI÷Iì÷I›÷Iô›÷I½`÷I`½`÷I…÷IÌ…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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�ïõI'öI�'öIÅDöI ÅDöI»KöI»KöIómöI€ómöI‘®öIü‘®öI�µöI�µöIÅ×öIpÅ×öIc÷Iìc÷I_÷Iô_÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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�ïõIÓöI�ÓöIqDöI qDöIýHöIýHöIŸmöI€ŸmöI=®öIü=®öIϲöIϲöIq×öIpq×öI÷Iì÷I¡÷Iô¡÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Im¶÷Iìm¶÷I9øIL 9øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIwûIT wûIGØûI˜GØûIïêûIœ€} 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�ïõI'öI�'öIÅDöI ÅDöI»KöI»KöIómöI€ómöI‘®öIü‘®öI�µöI�µöIÅ×öIpÅ×öIc÷Iìc÷I_÷Iô_÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI¹ûIT ¹ûI‰ØûI˜‰ØûI1ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIÍöI�ÍöIkDöI kDöIIöIIöIŸmöI€ŸmöI=®öIü=®öIí²öIí²öIq×öIpq×öI÷Iì÷I¿÷Iô¿÷I½`÷I`½`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöIKöIKöIùmöI€ùmöI—®öIü—®öIQµöIQµöIË×öIpË×öIi÷Iìi÷I#÷Iô#÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIyûIT yûIIÙûI˜IÙûIñëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöIKöIKöIùmöI€ùmöI—®öIü—®öIQµöIQµöIË×öIpË×öIi÷Iìi÷I#÷Iô#÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÇöI�ÇöIeDöI eDöI÷HöI÷HöI™möI€™möI7®öIü7®öIɲöIɲöIk×öIpk×öI ÷Iì ÷I›÷Iô›÷I±`÷I`±`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöI¯KöI¯KöIÿmöI€ÿmöI�®öIü�®öI�µöI�µöIÑ×öIpÑ×öIo÷Iìo÷IS÷IôS÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI…ûIT …ûIUÙûI˜UÙûIýëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÁöI�ÁöI_DöI _DöI÷HöI÷HöI“möI€“möI1®öIü1®öIɲöIɲöIe×öIpe×öI÷Iì÷I›÷Iô›÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÁöI�ÁöI_DöI _DöIIöIIöI“möI€“möI1®öIü1®öIá²öIá²öIe×öIpe×öI÷Iì÷I³÷Iô³÷I«`÷I`«`÷I …÷IÌ …÷Ia¶÷Iìa¶÷I-øIL -øI £øI¬  £øIåxùIp åxùIeúIt eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI3öI�3öIÑDöI ÑDöIyKöIyKöInöI€nöI£®öIü£®öIKµöIKµöI××öIp××öIu÷Iìu÷I÷Iô÷I_÷I`_÷Iƒ÷Ĩ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIßxùIp ßxùIeúIt eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI»öI�»öIYDöI YDöIIöIIöI�möI€�möI+®öIü+®öIá²öIá²öI_×öIp_×öIý÷Iìý÷I³÷Iô³÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ia¶÷Iìa¶÷I-øIL -øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI3öI�3öIÑDöI ÑDöIyKöIyKöInöI€nöI£®öIü£®öIKµöIKµöI××öIp××öIu÷Iìu÷I÷Iô÷I_÷I`_÷Iƒ÷Ĩ÷Ia¶÷Iìa¶÷I-øIL -øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI»öI�»öIYDöI YDöIëHöIëHöI‡möI€‡möI%®öIü%®öI½²öI½²öIY×öIpY×öI÷÷Iì÷÷I�÷Iô�÷I_÷I`_÷Iyƒ÷IÌyƒ÷I[¶÷Iì[¶÷I'øIL 'øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúI¿ûIT ¿ûI�ØûI˜�ØûI7ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöI£KöI£KöI nöI€ nöI©®öIü©®öIuµöIuµöIÝ×öIpÝ×öI{÷Iì{÷IG÷IôG÷Ic`÷I`c`÷IÄ÷IÌÄ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øI/wùIp /wùI GúIt  GúIûIT ûI×ûI˜×ûI½éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIµöI�µöISDöI SDöIëHöIëHöI‡möI€‡möI%®öIü%®öI½²öI½²öIY×öIpY×öI÷÷Iì÷÷I�÷Iô�÷I_÷I`_÷Iyƒ÷IÌyƒ÷I[¶÷Iì[¶÷I'øIL 'øIÝ¢øI¬ Ý¢øIµxùIp µxùIïdúIt ïdúI_ûIT _ûI/ØûI˜/ØûI×êûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöI£KöI£KöI nöI€ nöI©®öIü©®öIuµöIuµöIÝ×öIpÝ×öI{÷Iì{÷IG÷IôG÷I‡`÷I`‡`÷Iç„÷IÌç„÷I[¶÷Iì[¶÷I'øIL 'øIï¢øI¬ ï¢øIwùIp wùIõFúIt õFúI§ûIT §ûI�ÖûI˜�ÖûIEéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI?öI�?öIÝDöI ÝDöIsKöIsKöI nöI€ nöI©®öIü©®öIEµöIEµöIÝ×öIpÝ×öI{÷Iì{÷I÷Iô÷Iõ^÷I`õ^÷IUƒ÷IÌUƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øIAwùIp AwùIGúIt GúIsûIT sûIi×ûI˜i×ûIêûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI¯öI�¯öIMDöI MDöIIöIIöI�möI€�möI®öIü®öIÕ²öIÕ²öIS×öIpS×öIñ÷Iìñ÷I§÷Iô§÷I½`÷I`½`÷I…÷IÔ…÷I[¶÷Iô[¶÷I'øIT 'øIõ¢øI´ õ¢øIwùIx wùIûFúI| ûFúIËûI\ ËûIÁÖûI ÁÖûIiéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI?öI�?öIÝDöI ÝDöIsKöIsKöI nöI€ nöI©®öIü©®öIEµöIEµöIÝ×öIpÝ×öI{÷Iì{÷I÷Iô÷Iõ^÷I`õ^÷IUƒ÷IÌUƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øI)wùIp )wùIGúIt GúIûIT ûI÷ÖûI˜÷ÖûIŸéûIœ€} 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„iïõI…öI…öI#DöI|#DöI‡IöI„‡IöI½LöIð½LöIGVöI|GVöI}YöIH}YöIWmöIÔWmöIõ­öIPõ­öIY³öIXY³öI�¶öIÄ�¶öIÀöIPÀöIOÃöIOÃöI)×öI¨)×öIÇ÷I$ Ç÷I+÷I, +÷Ia ÷I˜ a ÷Ië)÷I$ ë)÷I!-÷Ið !-÷IcZ÷I| cZ÷I“`÷IX“`÷Ió„÷IÔó„÷I5²÷Iô5²÷IiÀ÷IÐiÀ÷I%õ÷I\%õ÷IgøI€gøIeøI¬eøI£øI8£øIñxùIüñxùI+eúI+eúI¥rúIà¥rúIsûIlsûICÙûI°CÙûIëëûI´€} 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„iïõI…öI…öI#DöI|#DöI‡IöI„‡IöI½LöIð½LöIGVöI|GVöI}YöIH}YöIQmöIÔQmöIï­öIPï­öIY³öIXY³öI�¶öIÄ�¶öIÀöIPÀöIOÃöIOÃöI#×öI¨#×öIÁ÷I$ Á÷I+÷I, +÷Ia ÷I˜ a ÷Ië)÷I$ ë)÷I!-÷Ið !-÷IcZ÷I| cZ÷I«`÷IX«`÷I …÷IÔ …÷IM²÷IôM²÷I#Á÷IÐ#Á÷Ißõ÷I\ßõ÷IøI€øI "øI¬ "øI£øI8£øIÙxùIüÙxùIeúIeúI�rúIà�rúIûIlûIÑØûI°ÑØûIyëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIQöIèQöIïDöIdïDöI—KöIl—KöI!UöIè!UöIWXöI´WXöInöI@nöI»®öI¼»®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIï×öI˜ï×öI�÷I�÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷IŸ`÷IÌ Ÿ`÷Iÿ„÷IH ÿ„÷IA²÷IhA²÷IQÀ÷IDQÀ÷I õ÷IÐ õ÷IëøIôëøIéøI éøIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúIËûIàËûI›ØûI$›ØûICëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI¯öIè¯öIMDöIdMDöI©KöIl©KöI3UöIè3UöIiXöI´iXöI�möI@�möI®öI¼®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIS×öI˜S×öIñ÷Iñ÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I“`÷IÌ “`÷Ió„÷IH ó„÷I5²÷Ih5²÷IWÀ÷IDWÀ÷Iõ÷IÐõ÷IøIôøIøI øI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúIyûIàyûIIÙûI$IÙûIñëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI¯öIè¯öIMDöIdMDöI£KöIl£KöI-UöIè-UöIcXöI´cXöI�möI@�möI®öI¼®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIS×öI˜S×öIñ÷Iñ÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷IWÀ÷IDWÀ÷Iõ÷IÐõ÷I�øIô�øI›øI ›øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIûIàûIÑØûI$ÑØûIyëûI(€} 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$J%J'À'À)6*6,¬,¬."/"1˜1˜346„6„8ú8ú:p;p=æ=æ?\@\BÒBÒDHEHG¾G¾IJL‡L‡NûNûPoQoSãSãUxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒxŒxŽìŽì�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxiïõIEöIèEöIãDöIdãDöI©KöIl©KöI3UöIè3UöIiXöI´iXöInöI@nöIµ®öI¼µ®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIé×öI˜é×öI‡÷I‡÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷IÛÀ÷IDÛÀ÷I—õ÷IЗõ÷IƒøIôƒøI� øI � øI×¢øI¬×¢øI¯xùIp¯xùIédúItédúIcrúITcrúIYûIàYûI)ØûI$)ØûIÑêûI(€} 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$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTz¤z¤|}††�÷�÷ƒh„h†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·loïõIµöIеöISDöILSDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI‡möI(‡möI%®öI¤%®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIY×öI€Y×öI÷÷Iü÷÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷IŸ÷I,Ÿ÷IøI˜øI£øI$£øIÙxùIèÙxùIeúIìeúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·loïõIµöIеöISDöILSDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI®öI¤®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIS×öI€S×öIñ÷Iüñ÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷I_»÷I,_»÷I»øI˜»øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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·loïõI?öIÐ?öIÝDöILÝDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöI¯®öI¤¯®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIã×öI€ã×öI�÷Iü�÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷I_»÷I,_»÷I�øI˜�øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI¿ûIÌ¿ûI�ØûI�ØûI7ëûI€} 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·luïõI»öIлöIYDöILYDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI�möI(�möI+®öI¤+®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI_×öI€_×öIý÷Iüý÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I¿¸÷I,¿¸÷I=øI˜=øI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIéØûIéØûI‘ëûI€} 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·luïõI9öIÐ9öI×DöIL×DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI£®öI¤£®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI××öI€××öIu÷Iüu÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I¿¸÷I,¿¸÷IÝøI˜ÝøI£øI$£øIßxùIèßxùIeúIìeúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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·luïõI»öIлöIYDöILYDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI‡möI(‡möI%®öI¤%®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIY×öI€Y×öI÷÷Iü÷÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IY»÷I,Y»÷IÙøI˜ÙøI颸I$颸IÁxùIèÁxùIûdúIìûdúI§ûI̧ûIwØûIwØûIëûI€} 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$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŠbŠdüdüfngniàiàkRlRnÄnÄp6q6s¨s¨uvxŒxŒzþzþ|p}pââ�T‚T„ƄƆ8‡8‰©‰©‹ŒŽ‹Ž‹�ü�ü’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI9öIÐ9öI×DöIL×DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI nöI( nöI©®öI¤©®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIÝ×öI€Ý×öI{÷Iü{÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷IY»÷I,Y»÷IsøI˜søIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÁöIÐÁöI_DöIL_DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI�möI(�möI+®öI¤+®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI_×öI€_×öIý÷Iüý÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I¹¸÷I,¹¸÷IUøI˜UøI£øI$£øIëxùIèëxùI%eúIì%eúImûIÌmûI=ÙûI=ÙûIåëûI€} 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·luïõI3öIÐ3öIÑDöILÑDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöIÿmöI(ÿmöI�®öI¤�®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIÑ×öI€Ñ×öIo÷Iüo÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I¹¸÷I,¹¸÷I³øI˜³øI £øI$ £øIåxùIèåxùIeúIìeúICûIÌCûIÙûIÙûI»ëûI€} 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·luïõI»öIлöIYDöILYDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI�möI(�möI+®öI¤+®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI_×öI€_×öIý÷Iüý÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IY»÷I,Y»÷IñøI˜ñøI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIÑØûIÑØûIyëûI€} 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·luïõI3öIÐ3öIÑDöILÑDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI£®öI¤£®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI××öI€××öIu÷Iüu÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷IM»÷I,M»÷IIøI˜IøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1n¢< b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 c£¹ùä$ 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��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÇöIÐÇöIeDöILeDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI“möI(“möI1®öI¤1®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIe×öI€e×öI÷Iü÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I³¸÷I,³¸÷IsøI˜søI£øI$£øIñxùIèñxùI+eúIì+eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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·l{ïõI'öIÐ'öIÅDöILÅDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIùmöI(ùmöI—®öI¤—®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIË×öI€Ë×öIi÷Iüi÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I­¸÷I,­¸÷I‰øI˜‰øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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·l{ïõIÁöIÐÁöI_DöIL_DöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI“möI(“möI1®öI¤1®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIe×öI€e×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IS»÷I,S»÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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·l{ïõI-öIÐ-öIËDöILËDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIùmöI(ùmöI—®öI¤—®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIË×öI€Ë×öIi÷Iüi÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷IA»÷I,A»÷IøI˜øI颸I$颸IÁxùIèÁxùIûdúIìûdúIkûIÌkûI;ØûI;ØûIãêûI€} 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·l�ïõIÇöIÐÇöIeDöILeDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI™möI(™möI7®öI¤7®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIk×öI€k×öI ÷Iü ÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I³¸÷I,³¸÷I—øI˜—øI£øI$£øIëxùIèëxùI%eúIì%eúIIûIÌIûIÙûIÙûIÁëûI€} 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·l�ïõI!öIÐ!öI¿DöIL¿DöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIómöI(ómöI‘®öI¤‘®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÅ×öI€Å×öIc÷Iüc÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I­¸÷I,­¸÷IéøI˜éøI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIéØûIéØûI‘ëûI€} 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·l�ïõIÇöIÐÇöIeDöILeDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI™möI(™möI7®öI¤7®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIk×öI€k×öI ÷Iü ÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IS»÷I,S»÷I9øI˜9øI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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·l�ïõI'öIÐ'öIÅDöILÅDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIómöI(ómöI‘®öI¤‘®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÅ×öI€Å×öIc÷Iüc÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IG»÷I,G»÷I…øI˜…øI£øI$£øIÙxùIèÙxùIeúIìeúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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�ïõIöIÐöI¹DöIL¹DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIímöI(ímöI‹®öI¤‹®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI¿×öI€¿×öI]÷Iü]÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I³¸÷I,³¸÷I�øI˜�øI£øI$£øIñxùIèñxùI+eúIì+eúICûIÌCûIÙûIÙûI»ëûI€} 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�ïõIÍöIÐÍöIkDöILkDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIŸmöI(ŸmöI=®öI¤=®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIq×öI€q×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷IY»÷I,Y»÷I¥øI˜¥øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúI¹ûI̹ûI‰ØûI‰ØûI1ëûI€} 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�ïõIöIÐöI¹DöIL¹DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIímöI(ímöI‹®öI¤‹®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI¿×öI€¿×öI]÷Iü]÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷IY»÷I,Y»÷IEøI˜EøI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI¿DöI÷HöI$÷HöIómöI�ómöI‘®öI‘®öIɲöI$ɲöIÅ×öI�Å×öIc÷Ic÷I›÷I$›÷I7_÷I�7_÷I—ƒ÷Iü—ƒ÷Is¶÷I s¶÷I?øI| ?øI%£øIÜ %£øIýxùI  ýxùI7eúI¤ 7eúIûI„ ûIïØûIÈïØûI—ëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI ³DöIµKöIµKöIçmöI€çmöI…®öIü…®öI‡µöI‡µöI¹×öIp¹×öIW÷IìW÷IY÷IôY÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIßöI�ßöI}DöI }DöI IöI IöI±möI€±möIO®öIüO®öIÛ²öIÛ²öIƒ×öIpƒ×öI!÷Iì!÷I­÷Iô­÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Is¶÷Iìs¶÷I?øIL ?øI7£øI¬ 7£øIyùIp yùIIeúIt IeúI�ûIT �ûImÙûI˜mÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¹DöI ¹DöI—KöI—KöIímöI€ímöI‹®öIü‹®öIiµöIiµöI¿×öIp¿×öI]÷Iì]÷I;÷Iô;÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¹DöI ¹DöI—KöI—KöIímöI€ímöI‹®öIü‹®öIiµöIiµöI¿×öIp¿×öI]÷Iì]÷I;÷Iô;÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Is¶÷Iìs¶÷I?øIL ?øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI}ûIT }ûIMØûI˜MØûIõêûIœ€} 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�ïõIÙöI�ÙöIwDöI wDöIýHöIýHöI«möI€«möII®öIüI®öIϲöIϲöI}×öIp}×öI÷Iì÷I¡÷Iô¡÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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�ïõIÙöI�ÙöIwDöI wDöIýHöIýHöI¥möI€¥möIC®öIüC®öIϲöIϲöIw×öIpw×öI÷Iì÷I¡÷Iô¡÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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�ïõI!öI�!öI¿DöI ¿DöIµKöIµKöIómöI€ómöI‘®öIü‘®öI‡µöI‡µöIÅ×öIpÅ×öIc÷Iìc÷IY÷IôY÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIÓöI�ÓöIqDöI qDöIIöIIöI¥möI€¥möIC®öIüC®öIç²öIç²öIw×öIpw×öI÷Iì÷I¹÷Iô¹÷I·`÷I`·`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöIKöIKöIùmöI€ùmöI—®öIü—®öIQµöIQµöIË×öIpË×öIi÷Iìi÷I#÷Iô#÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIÍöI�ÍöIkDöI kDöIIöIIöIŸmöI€ŸmöI=®öIü=®öIí²öIí²öIq×öIpq×öI÷Iì÷I¿÷Iô¿÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI¹ûIT ¹ûI‰ØûI˜‰ØûI1ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöIKöIKöIùmöI€ùmöI—®öIü—®öIQµöIQµöIË×öIpË×öIi÷Iìi÷I#÷Iô#÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÍöI�ÍöIkDöI kDöIýHöIýHöIŸmöI€ŸmöI=®öIü=®öIϲöIϲöIq×öIpq×öI÷Iì÷I¡÷Iô¡÷I·`÷I`·`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIyûIT yûIIÙûI˜IÙûIñëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöI©KöI©KöIùmöI€ùmöI—®öIü—®öI{µöI{µöIË×öIpË×öIi÷Iìi÷IM÷IôM÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÇöI�ÇöIeDöI eDöI÷HöI÷HöI™möI€™möI7®öIü7®öIɲöIɲöIk×öIpk×öI ÷Iì ÷I›÷Iô›÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöI©KöI©KöIùmöI€ùmöI—®öIü—®öI{µöI{µöIË×öIpË×öIi÷Iìi÷IM÷IôM÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÇöI�ÇöIeDöI eDöIIöIIöI“möI€“möI1®öIü1®öIá²öIá²öIe×öIpe×öI÷Iì÷I³÷Iô³÷I«`÷I`«`÷I …÷IÌ …÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIOÙûI˜OÙûI÷ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI3öI�3öIÑDöI ÑDöIyKöIyKöIÿmöI€ÿmöI�®öIü�®öIKµöIKµöIÑ×öIpÑ×öIo÷Iìo÷I÷Iô÷I“`÷I`“`÷Ió„÷IÌó„÷Ia¶÷Iìa¶÷I-øIL -øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÁöI�ÁöI_DöI _DöIIöIIöI“möI€“möI1®öIü1®öIá²öIá²öIe×öIpe×öI÷Iì÷I³÷Iô³÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI3öI�3öIÑDöI ÑDöIyKöIyKöIÿmöI€ÿmöI�®öIü�®öIKµöIKµöIÑ×öIpÑ×öIo÷Iìo÷I÷Iô÷I±`÷I`±`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øIÝ¢øI¬ Ý¢øIµxùIp µxùIïdúIt ïdúI_ûIT _ûI/ØûI˜/ØûI×êûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÁöI�ÁöI_DöI _DöIñHöIñHöI�möI€�möI+®öIü+®öIòöIòöI_×öIp_×öIý÷Iìý÷I•÷Iô•÷I¥`÷I`¥`÷I…÷IÌ…÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øIßxùIp ßxùIeúIt eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI3öI�3öIÑDöI ÑDöI�KöI�KöInöI€nöI£®öIü£®öIoµöIoµöI××öIp××öIu÷Iìu÷IA÷IôA÷Iu`÷I`u`÷IÕ„÷IÌÕ„÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øI)wùIp )wùIGúIt GúI ûIT  ûI×ûI˜×ûI«éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõI»öI�»öIYDöI YDöIëHöIëHöI�möI€�möI+®öIü+®öI½²öI½²öI_×öIp_×öIý÷Iìý÷I�÷Iô�÷IÉ`÷I`É`÷I)…÷IÌ)…÷I[¶÷Iì[¶÷I'øIL 'øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI3öI�3öIÑDöI ÑDöI�KöI�KöInöI€nöI£®öIü£®öIoµöIoµöI××öIp××öIu÷Iìu÷IA÷IôA÷I“`÷I`“`÷Ió„÷IÌó„÷I[¶÷Iì[¶÷I'øIL 'øIõ¢øI¬ õ¢øIwùIp wùIûFúIt ûFúIÅûIT ÅûI»ÖûI˜»ÖûIcéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõIµöI�µöISDöI SDöIIöIIöI‡möI€‡möI%®öIü%®öIÕ²öIÕ²öIY×öIpY×öI÷÷Iì÷÷I§÷Iô§÷I�`÷I`�`÷Ií„÷IÔí„÷I[¶÷Iô[¶÷I'øIT 'øI£øI´ £øI/wùIx /wùI GúI|  GúIûI\ ûI×ûI ×ûI½éûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI9öI�9öI×DöI ×DöIsKöIsKöI nöI€ nöI©®öIü©®öIEµöIEµöIÝ×öIpÝ×öI{÷Iì{÷I÷Iô÷Iõ^÷I`õ^÷IUƒ÷IÌUƒ÷I[¶÷Iì[¶÷I'øIL 'øI £øI¬  £øI5wùIp 5wùIGúIt GúI=ûIT =ûI3×ûI˜3×ûIÛéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõIµöI�µöISDöI SDöIIöIIöI�möI€�möI®öIü®öIÕ²öIÕ²öIS×öIpS×öIñ÷Iìñ÷I§÷Iô§÷I±`÷I`±`÷I…÷IÔ…÷I[¶÷Iô[¶÷I'øIT 'øIï¢øI´ ï¢øIwùIx wùIõFúI| õFúI­ûI\ ­ûI£ÖûI £ÖûIKéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI9öI�9öI×DöI ×DöIsKöIsKöI nöI€ nöI©®öIü©®öIEµöIEµöIÝ×öIpÝ×öI{÷Iì{÷I÷Iô÷Iõ^÷I`õ^÷IUƒ÷IÌUƒ÷I[¶÷Iì[¶÷I'øIL 'øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúIãûIT ãûIÙÖûI˜ÙÖûI�éûIœ€} 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„iïõI‹öI‹öI)DöI|)DöI�IöI„�IöIÃLöIðÃLöIMVöI|MVöIƒYöIHƒYöIWmöIÔWmöIõ­öIPõ­öI_³öIX_³öI•¶öIÄ•¶öIÀöIPÀöIUÃöIUÃöI)×öI¨)×öIÇ÷I$ Ç÷I1÷I, 1÷Ig ÷I˜ g ÷Iñ)÷I$ ñ)÷I'-÷Ið '-÷IiZ÷I| iZ÷IÉ`÷IXÉ`÷I)…÷IÔ)…÷Ik²÷Iôk²÷I“À÷IГÀ÷IOõ÷I\Oõ÷I?øI€?øI=øI¬=øI£øI8£øIÙxùIüÙxùIeúIeúI�rúIà�rúI ûIl ûIÝØûI°ÝØûI…ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIKöIèKöIéDöIdéDöI—KöIl—KöI!UöIè!UöIWXöI´WXöInöI@nöI»®öI¼»®öIiµöIÄiµöIó¾öI@ó¾öI)ÂöI )ÂöIï×öI˜ï×öI�÷I�÷I;÷I ;÷IÅ(÷I˜ Å(÷Iû+÷Id û+÷I=Y÷Ið =Y÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷IWÀ÷IDWÀ÷Iõ÷IÐõ÷IcøIôcøIaøI aøI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúICûIàCûIÙûI$ÙûI»ëûI(€} 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„iïõI…öI…öI#DöI|#DöI�IöI„�IöIÃLöIðÃLöIMVöI|MVöIƒYöIHƒYöIWmöIÔWmöIõ­öIPõ­öI_³öIX_³öI•¶öIÄ•¶öIÀöIPÀöIUÃöIUÃöI)×öI¨)×öIÇ÷I$ Ç÷I1÷I, 1÷Ig ÷I˜ g ÷Iñ)÷I$ ñ)÷I'-÷Ið '-÷IiZ÷I| iZ÷I‡`÷IX‡`÷Iç„÷IÔç„÷I)²÷Iô)²÷ISÁ÷IÐSÁ÷Iö÷I\ö÷IíøI€íøIë"øI¬ë"øIõ¢øI8õ¢øIÍxùIüÍxùIeúIeúI�rúIà�rúIËûIlËûI›ØûI°›ØûICëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIµöIèµöISDöIdSDöI©KöIl©KöI3UöIè3UöIiXöI´iXöI‡möI@‡möI%®öI¼%®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIY×öI˜Y×öI÷÷I÷÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷IÉ`÷IÌ É`÷I)…÷IH )…÷Ik²÷Ihk²÷IuÀ÷IDuÀ÷I1õ÷IÐ1õ÷I3øIô3øI1øI 1øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúI1ûIà1ûIÙûI$ÙûI©ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIEöIèEöIãDöIdãDöI©KöIl©KöI3UöIè3UöIiXöI´iXöInöI@nöIµ®öI¼µ®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIé×öI˜é×öI‡÷I‡÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I“`÷IÌ “`÷Ió„÷IH ó„÷I5²÷Ih5²÷I)Á÷ID)Á÷Iåõ÷IÐåõ÷IøIôøI"øI "øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIûIàûIéØûI$éØûI‘ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIKöIèKöIéDöIdéDöI©KöIl©KöI3UöIè3UöIiXöI´iXöInöI@nöIµ®öI¼µ®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIé×öI˜é×öI‡÷I‡÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷I«À÷ID«À÷Igõ÷IÐgõ÷IøIôøI øI  øI颸I¬颸IÁxùIpÁxùIûdúItûdúIurúITurúI¡ûIà¡ûIqØûI$qØûIëûI(€} 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$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�zßzß|Q}QÃÃ�5‚5„§„§†<‡<‰­‰­‹ŒŽ�Ž��‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·loïõI»öIлöIYDöILYDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI+®öI¤+®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöI_×öI€_×öIý÷Iüý÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷Ió½÷I,ó½÷IQøI˜QøI£øI$£øIÙxùIèÙxùIeúIìeúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·loïõI?öIÐ?öIÝDöILÝDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI nöI( nöI©®öI¤©®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÝ×öI€Ý×öI{÷Iü{÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I¾÷I,¾÷I¥øI˜¥øI £øI$ £øIåxùIèåxùIeúIìeúImûIÌmûI=ÙûI=ÙûIåëûI€} 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·loïõI»öIлöIYDöILYDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI‡möI(‡möI%®öI¤%®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIY×öI€Y×öI÷÷Iü÷÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷I�À÷I,�À÷IíøI˜íøI颸I$颸IÁxùIèÁxùIûdúIìûdúIÅûIÌÅûI•ØûI•ØûI=ëûI€} 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·loïõI?öIÐ?öIÝDöILÝDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöI¯®öI¤¯®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIã×öI€ã×öI�÷Iü�÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I¥À÷I,¥À÷IAøI˜AøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIõûIÌõûIÅØûIÅØûImëûI€} 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·luïõIÁöIÐÁöI_DöIL_DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI�möI(�möI+®öI¤+®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI_×öI€_×öIý÷Iüý÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷Ió½÷I,ó½÷IoøI˜oøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI¿ûIÌ¿ûI�ØûI�ØûI7ëûI€} 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·luïõI9öIÐ9öI×DöIL×DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI£®öI¤£®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI××öI€××öIu÷Iüu÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷I¾÷I,¾÷IÃøI˜ÃøI£øI$£øIñxùIèñxùI+eúIì+eúIûIÌûIOÙûIOÙûI÷ëûI€} 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·luïõI»öIлöIYDöILYDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI�möI(�möI+®öI¤+®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI_×öI€_×öIý÷Iüý÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I�À÷I,�À÷I øI˜ øIÝ¢øI$Ý¢øIµxùIèµxùIïdúIìïdúI_ûIÌ_ûI/ØûI/ØûI×êûI€} 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·luïõI9öIÐ9öI×DöIL×DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI nöI( nöI©®öI¤©®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIÝ×öI€Ý×öI{÷Iü{÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I¥À÷I,¥À÷I_øI˜_øI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIרûIרûIëûI€} 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·l{ïõIÇöIÐÇöIeDöILeDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI“möI(“möI1®öI¤1®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIe×öI€e×öI÷Iü÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷Ió½÷I,ó½÷I�øI˜�øI£øI$£øIÙxùIèÙxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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·l{ïõI-öIÐ-öIËDöILËDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöIÿmöI(ÿmöI�®öI¤�®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIÑ×öI€Ñ×öIo÷Iüo÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I¾÷I,¾÷IáøI˜áøI £øI$ £øIåxùIèåxùIeúIìeúIIûIÌIûIÙûIÙûIÁëûI€} 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·l{ïõIÁöIÐÁöI_DöIL_DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI“möI(“möI1®öI¤1®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIe×öI€e×öI÷Iü÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷I“À÷I,“À÷I/øI˜/øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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·l{ïõI3öIÐ3öIÑDöILÑDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI£®öI¤£®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI××öI€××öIu÷Iüu÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I¥À÷I,¥À÷IƒøI˜ƒøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l{ïõIÇöIÐÇöIeDöILeDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI™möI(™möI7®öI¤7®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIk×öI€k×öI ÷Iü ÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷Ió½÷I,ó½÷I±øI˜±øI £øI$ £øIåxùIèåxùIeúIìeúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·l{ïõI'öIÐ'öIÅDöILÅDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIùmöI(ùmöI—®öI¤—®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIË×öI€Ë×öIi÷Iüi÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷Iÿ½÷I,ÿ½÷IÛøI˜ÛøI £øI$ £øIåxùIèåxùIeúIìeúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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·l{ïõI-öIÐ-öIËDöILËDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIÿmöI(ÿmöI�®öI¤�®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIÑ×öI€Ñ×öIo÷Iüo÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IŸÀ÷I,ŸÀ÷I}øI˜}øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI³ûI̳ûIƒØûIƒØûI+ëûI€} 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�ïõIÍöIÐÍöIkDöILkDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIŸmöI(ŸmöI=®öI¤=®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIq×öI€q×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷Ió½÷I,ó½÷IÕøI˜ÕøI£øI$£øIëxùIèëxùI%eúIì%eúIOûIÌOûIÙûIÙûIÇëûI€} 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�ïõI!öIÐ!öI¿DöIL¿DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIómöI(ómöI‘®öI¤‘®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIÅ×öI€Å×öIc÷Iüc÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷Ió½÷I,ó½÷I±øI˜±øI%£øI$%£øIýxùIèýxùI7eúIì7eúIûIÌûIOÙûIOÙûI÷ëûI€} 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��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI'öIÐ'öIÅDöILÅDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIómöI(ómöI‘®öI¤‘®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÅ×öI€Å×öIc÷Iüc÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷I“À÷I,“À÷IMøI˜MøI £øI$ £øIåxùIèåxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbêbêd]e]gÐgÐiCjCl¶l¶n'o'q˜q˜s t vzvzxëxëz\{\}Í}Í>€>‚¯‚¯„ … ‡’‡’‰ŠŒtŒtŽåŽå�V‘V“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIÓöIÐÓöIqDöILqDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI¥möI(¥möIC®öI¤C®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIw×öI€w×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷I¾÷I,¾÷IMøI˜MøI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIöIÐöI¹DöIL¹DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIímöI(ímöI‹®öI¤‹®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI¿×öI€¿×öI]÷Iü]÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷I¾÷I,¾÷I/øI˜/øI+£øI$+£øIyùIèyùI=eúIì=eúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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‡ïõI!öIÐ!öI¿DöIL¿DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIímöI(ímöI‹®öI¤‹®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI¿×öI€¿×öI]÷Iü]÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I¥À÷I,¥À÷IÑøI˜ÑøI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIéØûIéØûI‘ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI ³DöI¯KöI¯KöIçmöI€çmöI…®öIü…®öI�µöI�µöI¹×öIp¹×öIW÷IìW÷IS÷IôS÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúIgûIT gûI7ÙûI˜7ÙûIßëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI¿DöIýHöI$ýHöIímöI�ímöI‹®öI‹®öIϲöI$ϲöI¿×öI�¿×öI]÷I]÷I¡÷I$¡÷I7_÷I�7_÷I—ƒ÷Iü—ƒ÷Iy¶÷I y¶÷IEøI| EøI£øIÜ £øIñxùI  ñxùI+eúI¤ +eúIéûI„ éûI¹ØûIȹØûIaëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöI¯KöI¯KöIçmöI€çmöI…®öIü…®öI�µöI�µöI¹×öIp¹×öIW÷IìW÷IS÷IôS÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIåöI�åöIƒDöI ƒDöIIöIIöI·möI€·möIU®öIüU®öIÕ²öIÕ²öI‰×öIp‰×öI'÷Iì'÷I§÷Iô§÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¹DöI ¹DöI—KöI—KöIímöI€ímöI‹®öIü‹®öIiµöIiµöI¿×öIp¿×öI]÷Iì]÷I;÷Iô;÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIåöI�åöIƒDöI ƒDöIIöIIöI±möI€±möIO®öIüO®öIÕ²öIÕ²öIƒ×öIpƒ×öI!÷Iì!÷I§÷Iô§÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¹DöI ¹DöI—KöI—KöIímöI€ímöI‹®öIü‹®öIiµöIiµöI¿×öIp¿×öI]÷Iì]÷I;÷Iô;÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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‡ïõIßöI�ßöI}DöI }DöIýHöIýHöI±möI€±möIO®öIüO®öIϲöIϲöIƒ×öIpƒ×öI!÷Iì!÷I¡÷Iô¡÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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‡ïõI!öI�!öI¿DöI ¿DöI¯KöI¯KöIómöI€ómöI‘®öIü‘®öI�µöI�µöIÅ×öIpÅ×öIc÷Iìc÷IS÷IôS÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI+£øI¬ +£øIyùIp yùI=eúIt =eúI…ûIT …ûIUÙûI˜UÙûIýëûIœ€} 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‡ïõIÙöI�ÙöIwDöI wDöIýHöIýHöI«möI€«möII®öIüI®öIϲöIϲöI}×öIp}×öI÷Iì÷I¡÷Iô¡÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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‡ïõI!öI�!öI¿DöI ¿DöI¯KöI¯KöIómöI€ómöI‘®öIü‘®öI�µöI�µöIÅ×öIpÅ×öIc÷Iìc÷IS÷IôS÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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�ïõIÙöI�ÙöIwDöI wDöIIöIIöI«möI€«möII®öIüI®öIç²öIç²öI}×öIp}×öI÷Iì÷I¹÷Iô¹÷I±`÷I`±`÷I…÷IÌ…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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�ïõI'öI�'öIÅDöI ÅDöIKöIKöIómöI€ómöI‘®öIü‘®öIQµöIQµöIÅ×öIpÅ×öIc÷Iìc÷I#÷Iô#÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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�ïõIÓöI�ÓöIqDöI qDöIIöIIöI¥möI€¥möIC®öIüC®öIç²öIç²öIw×öIpw×öI÷Iì÷I¹÷Iô¹÷I½`÷I`½`÷I…÷IÌ…÷Im¶÷Iìm¶÷I9øIL 9øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIqûIT qûIAØûI˜AØûIéêûIœ€} 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�ïõI'öI�'öIÅDöI ÅDöIKöIKöIùmöI€ùmöI—®öIü—®öIQµöIQµöIË×öIpË×öIi÷Iìi÷I#÷Iô#÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÓöI�ÓöIqDöI qDöIýHöIýHöIŸmöI€ŸmöI=®öIü=®öIϲöIϲöIq×öIpq×öI÷Iì÷I¡÷Iô¡÷I±`÷I`±`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÍöI�ÍöIkDöI kDöIýHöIýHöIŸmöI€ŸmöI=®öIü=®öIϲöIϲöIq×öIpq×öI÷Iì÷I¡÷Iô¡÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöI£KöI£KöIÿmöI€ÿmöI�®öIü�®öIuµöIuµöIÑ×öIpÑ×öIo÷Iìo÷IG÷IôG÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÍöI�ÍöIkDöI kDöI IöI IöI™möI€™möI7®öIü7®öIÛ²öIÛ²öIk×öIpk×öI ÷Iì ÷I­÷Iô­÷I±`÷I`±`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI £øI¬  £øIåxùIp åxùIeúIt eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöIyKöIyKöIÿmöI€ÿmöI�®öIü�®öIKµöIKµöIÑ×öIpÑ×öIo÷Iìo÷I÷Iô÷I¥`÷I`¥`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÇöI�ÇöIeDöI eDöI IöI IöI™möI€™möI7®öIü7®öIÛ²öIÛ²öIk×öIpk×öI ÷Iì ÷I­÷Iô­÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ia¶÷Iìa¶÷I-øIL -øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI3öI�3öIÑDöI ÑDöIyKöIyKöIÿmöI€ÿmöI�®öIü�®öIKµöIKµöIÑ×öIpÑ×öIo÷Iìo÷I÷Iô÷I½`÷I`½`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúI§ûIT §ûIwØûI˜wØûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÁöI�ÁöI_DöI _DöIñHöIñHöI“möI€“möI1®öIü1®öIòöIòöIe×öIpe×öI÷Iì÷I•÷Iô•÷I¥`÷I`¥`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI £øI¬  £øIåxùIp åxùIeúIt eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI3öI�3öIÑDöI ÑDöI—KöI—KöInöI€nöI£®öIü£®öIiµöIiµöI××öIp××öIu÷Iìu÷I;÷Iô;÷I�`÷I`�`÷Iá„÷IÌá„÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIAwùIp AwùIGúIt GúIsûIT sûIi×ûI˜i×ûIêûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÁöI�ÁöI_DöI _DöIñHöIñHöI�möI€�möI+®öIü+®öIòöIòöI_×öIp_×öIý÷Iìý÷I•÷Iô•÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Ia¶÷Iìa¶÷I-øIL -øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöI—KöI—KöInöI€nöI£®öIü£®öIiµöIiµöI××öIp××öIu÷Iìu÷I;÷Iô;÷I¥`÷I`¥`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øI)wùIp )wùIGúIt GúIûIT ûI÷ÖûI˜÷ÖûIŸéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI9öI�9öI×DöI ×DöImKöImKöI nöI€ nöI©®öIü©®öI?µöI?µöIÝ×öIpÝ×öI{÷Iì{÷I÷Iô÷Iû^÷I`û^÷I[ƒ÷IÌ[ƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øIGwùIp GwùI%GúIt %GúI…ûIT …ûI{×ûI˜{×ûI#êûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI»öI�»öIYDöI YDöIýHöIýHöI‡möI€‡möI%®öIü%®öIϲöIϲöIY×öIpY×öI÷÷Iì÷÷I¡÷Iô¡÷I¥`÷I`¥`÷I…÷IÔ…÷I[¶÷Iô[¶÷I'øIT 'øIã¢øI´ ã¢øI wùIx  wùIéFúI| éFúIeûI\ eûI[ÖûI [ÖûIéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI9öI�9öI×DöI ×DöImKöImKöI nöI€ nöI©®öIü©®öI?µöI?µöIÝ×öIpÝ×öI{÷Iì{÷I÷Iô÷Iû^÷I`û^÷I[ƒ÷IÌ[ƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øI/wùIp /wùI GúIt  GúI ûIT  ûI×ûI˜×ûI«éûIœ€} 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„iïõI‹öI‹öI)DöI|)DöI�IöI„�IöIÃLöIðÃLöIMVöI|MVöIƒYöIHƒYöI]möIÔ]möIû­öIPû­öI_³öIX_³öI•¶öIÄ•¶öIÀöIPÀöIUÃöIUÃöI/×öI¨/×öIÍ÷I$ Í÷I1÷I, 1÷Ig ÷I˜ g ÷Iñ)÷I$ ñ)÷I'-÷Ið '-÷IiZ÷I| iZ÷I«`÷IX«`÷I …÷IÔ …÷IM²÷IôM²÷IÃÀ÷IÐÃÀ÷Iõ÷I\õ÷IøI€øI øI¬ øI£øI8£øIßxùIüßxùIeúIeúI“rúIà“rúI%ûIl%ûIõØûI°õØûI�ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIQöIèQöIïDöIdïDöI�KöIl�KöI'UöIè'UöI]XöI´]XöI#nöI@#nöIÁ®öI¼Á®öIoµöIÄoµöIù¾öI@ù¾öI/ÂöI /ÂöIõ×öI˜õ×öI“÷I“÷IA÷I A÷IË(÷I˜ Ë(÷I,÷Id ,÷ICY÷Ið CY÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷IÁ÷IDÁ÷IÙõ÷IÐÙõ÷IëøIôëøIé!øI é!øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúIÅûIàÅûI•ØûI$•ØûI=ëûI(€} 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„iïõI‹öI‹öI)DöI|)DöI�IöI„�IöIÃLöIðÃLöIMVöI|MVöIƒYöIHƒYöIWmöIÔWmöIõ­öIPõ­öI_³öIX_³öI•¶öIÄ•¶öIÀöIPÀöIUÃöIUÃöI)×öI¨)×öIÇ÷I$ Ç÷I1÷I, 1÷Ig ÷I˜ g ÷Iñ)÷I$ ñ)÷I'-÷Ið '-÷IiZ÷I| iZ÷I·`÷IX·`÷I…÷IÔ…÷IY²÷IôY²÷I}Á÷IÐ}Á÷I9ö÷I\9ö÷IËøI€ËøIÉ#øI¬É#øIï¢øI8ï¢øIÇxùIüÇxùIeúIeúI{rúIà{rúI­ûIl­ûI}ØûI°}ØûI%ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIWöIèWöIõDöIdõDöI�KöIl�KöI'UöIè'UöI]XöI´]XöI#nöI@#nöIÁ®öI¼Á®öIoµöIÄoµöIù¾öI@ù¾öI/ÂöI /ÂöIõ×öI˜õ×öI“÷I“÷IA÷I A÷IË(÷I˜ Ë(÷I,÷Id ,÷ICY÷Ið CY÷I½`÷IÌ ½`÷I…÷IH …÷I_²÷Ih_²÷IeÁ÷IDeÁ÷I!ö÷IÐ!ö÷IAøIôAøI?#øI ?#øIã¢øI¬ã¢øI»xùIp»xùIõdúItõdúIorúITorúIeûIàeûI5ØûI$5ØûIÝêûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI»öIè»öIYDöIdYDöI©KöIl©KöI3UöIè3UöIiXöI´iXöI�möI@�möI+®öI¼+®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöI_×öI˜_×öIý÷Iý÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷I¥À÷ID¥À÷Iaõ÷IÐaõ÷IoøIôoøImøI møI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúImûIàmûI=ÙûI$=ÙûIåëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIKöIèKöIéDöIdéDöI©KöIl©KöI3UöIè3UöIiXöI´iXöInöI@nöIµ®öI¼µ®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIé×öI˜é×öI‡÷I‡÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I¥`÷IÌ ¥`÷I…÷IH …÷IG²÷IhG²÷IÿÀ÷IDÿÀ÷I»õ÷Iлõ÷I=øIô=øI;!øI ;!øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIûIàûIרûI$רûIëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI»öIè»öIYDöIdYDöI©KöIl©KöI3UöIè3UöIiXöI´iXöI‡möI@‡möI%®öI¼%®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIY×öI˜Y×öI÷÷I÷÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷IŸÀ÷IDŸÀ÷I[õ÷IÐ[õ÷I]øIô]øI[øI [øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIûûIàûûIËØûI$ËØûIsëûI(€} 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·loïõIÁöIÐÁöI_DöIL_DöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI+®öI¤+®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöI_×öI€_×öIý÷Iüý÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷I˸÷I,˸÷ICøI˜CøI £øI$ £øIåxùIèåxùIeúIìeúIsûIÌsûICÙûICÙûIëëûI€} 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·loïõIEöIÐEöIãDöILãDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöI¯®öI¤¯®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIã×öI€ã×öI�÷Iü�÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I˸÷I,˸÷IgøI˜gøI£øI$£øIÙxùIèÙxùIeúIìeúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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·loïõI»öIлöIYDöILYDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI�möI(�möI+®öI¤+®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöI_×öI€_×öIý÷Iüý÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷Ie»÷I,e»÷IßøI˜ßøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIûIÌûIÑØûIÑØûIyëûI€} 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·loïõIEöIÐEöIãDöILãDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöInöI(nöIµ®öI¤µ®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIé×öI€é×öI‡÷Iü‡÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷Ik»÷I,k»÷IøI˜øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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·luïõIÇöIÐÇöIeDöILeDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI“möI(“möI1®öI¤1®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIe×öI€e×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I¿¸÷I,¿¸÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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·luïõI9öIÐ9öI×DöIL×DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI nöI( nöI©®öI¤©®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIÝ×öI€Ý×öI{÷Iü{÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷IŸ÷I,Ÿ÷I…øI˜…øI £øI$ £øIåxùIèåxùIeúIìeúICûIÌCûIÙûIÙûI»ëûI€} 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·luïõIÁöIÐÁöI_DöIL_DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI“möI(“möI1®öI¤1®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIe×öI€e×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IY»÷I,Y»÷IµøI˜µøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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·l{ïõIÇöIÐÇöIeDöILeDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI™möI(™möI7®öI¤7®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIk×öI€k×öI ÷Iü ÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I¹¸÷I,¹¸÷IõøI˜õøI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIïØûIïØûI—ëûI€} 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·l{ïõI3öIÐ3öIÑDöILÑDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI£®öI¤£®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI××öI€××öIu÷Iüu÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IŸ÷I,Ÿ÷I�øI˜�øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIËûIÌËûI›ØûI›ØûICëûI€} 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·l{ïõIÇöIÐÇöIeDöILeDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI™möI(™möI7®öI¤7®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIk×öI€k×öI ÷Iü ÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IM»÷I,M»÷I‹øI˜‹øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI­ûIÌ­ûI}ØûI}ØûI%ëûI€} 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·l{ïõI9öIÐ9öI×DöIL×DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI nöI( nöI©®öI¤©®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIÝ×öI€Ý×öI{÷Iü{÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷Ie»÷I,e»÷I?øI˜?øIã¢øI$ã¢øI»xùIè»xùIõdúIìõdúIeûIÌeûI5ØûI5ØûIÝêûI€} 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·l{ïõI-öIÐ-öIËDöILËDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIÿmöI(ÿmöI�®öI¤�®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÑ×öI€Ñ×öIo÷Iüo÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I¿¸÷I,¿¸÷I—øI˜—øI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIéØûIéØûI‘ëûI€} 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·l{ïõIÍöIÐÍöIkDöILkDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIŸmöI(ŸmöI=®öI¤=®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIq×öI€q×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IA»÷I,A»÷ImøI˜møI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIרûIרûIëûI€} 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·l{ïõI3öIÐ3öIÑDöILÑDöIµKöITµKöI?UöIÐ?UöIuXöIœuXöIÿmöI(ÿmöI�®öI¤�®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIÑ×öI€Ñ×öIo÷Iüo÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I_»÷I,_»÷I9øI˜9øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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�ïõIÓöIÐÓöIqDöILqDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI¥möI(¥möIC®öI¤C®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIw×öI€w×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I³¸÷I,³¸÷I‹øI˜‹øI%£øI$%£øIýxùIèýxùI7eúIì7eúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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�ïõI'öIÐ'öIÅDöILÅDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIùmöI(ùmöI—®öI¤—®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIË×öI€Ë×öIi÷Iüi÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I³¸÷I,³¸÷IsøI˜søI£øI$£øIñxùIèñxùI+eúIì+eúICûIÌCûIÙûIÙûI»ëûI€} 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�ïõIÓöIÐÓöIqDöILqDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIŸmöI(ŸmöI=®öI¤=®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIq×öI€q×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷IS»÷I,S»÷I-øI˜-øI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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�ïõI-öIÐ-öIËDöILËDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIùmöI(ùmöI—®öI¤—®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIË×öI€Ë×öIi÷Iüi÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷IS»÷I,S»÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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$A%A'¶'¶)+*+, , ./1Š1Š3ÿ3ÿ5t6t8é8é:^;^=Ó=Ó?H@HB½B½D2E2G§G§IJL“L“NOQ}Q}SòSòUgVgXÜXÜZQ[Q]Æ]Æ_;`;b°b°d%e%gšgšijl„l„nönöphqhsÚsÚuLvLx¾x¾z0{0}¢}¢€‚†‚†„ø„ø†j‡j‰܉܋NŒNŽÀŽÀ�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIÙöIÐÙöIwDöILwDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI«möI(«möII®öI¤I®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI}×öI€}×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I¹¸÷I,¹¸÷IEøI˜EøI£øI$£øIñxùIèñxùI+eúIì+eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI!öIÐ!öI¿DöIL¿DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIímöI(ímöI‹®öI¤‹®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI¿×öI€¿×öI]÷Iü]÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷I³¸÷I,³¸÷IßøI˜ßøI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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‡ïõIÙöIÐÙöIwDöILwDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI¥möI(¥möIC®öI¤C®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIw×öI€w×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I_»÷I,_»÷IíøI˜íøI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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‡ïõI!öIÐ!öI¿DöIL¿DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIómöI(ómöI‘®öI¤‘®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIÅ×öI€Å×öIc÷Iüc÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I_»÷I,_»÷I‡øI˜‡øI£øI$£øIßxùIèßxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI³DöIýHöI$ýHöIçmöI�çmöI…®öI…®öIϲöI$ϲöI¹×öI�¹×öIW÷IW÷I¡÷I$¡÷I=_÷I�=_÷I�ƒ÷Iü�ƒ÷Iy¶÷I y¶÷IEøI| EøI+£øIÜ +£øIyùI  yùI=eúI¤ =eúIIûI„ IûIÙûIÈÙûIÁëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöI©KöI©KöIímöI€ímöI‹®öIü‹®öI{µöI{µöI¿×öIp¿×öI]÷Iì]÷IM÷IôM÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI¹DöIýHöI$ýHöIçmöI�çmöI…®öI…®öIϲöI$ϲöI¹×öI�¹×öIW÷IW÷I¡÷I$¡÷I=_÷I�=_÷I�ƒ÷Iü�ƒ÷Iy¶÷I y¶÷IEøI| EøI£øIÜ £øIëxùI  ëxùI%eúI¤ %eúIÑûI„ ÑûI¡ØûIÈ¡ØûIIëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöI©KöI©KöIímöI€ímöI‹®öIü‹®öI{µöI{µöI¿×öIp¿×öI]÷Iì]÷IM÷IôM÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIÙxùIp ÙxùIeúIt eúI‰ûIT ‰ûIYØûI˜YØûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIëöI�ëöI‰DöI ‰DöIIöIIöI½möI€½möI[®öIü[®öIÕ²öIÕ²öI�×öIp�×öI-÷Iì-÷I§÷Iô§÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI7£øI¬ 7£øIyùIp yùIIeúIt IeúI‘ûIT ‘ûIaÙûI˜aÙûI ìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI!öI�!öI¿DöI ¿DöI—KöI—KöIómöI€ómöI‘®öIü‘®öIiµöIiµöIÅ×öIpÅ×öIc÷Iìc÷I;÷Iô;÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIåöI�åöIƒDöI ƒDöIIöIIöI·möI€·möIU®öIüU®öIÕ²öIÕ²öI‰×öIp‰×öI'÷Iì'÷I§÷Iô§÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI!öI�!öI¿DöI ¿DöI—KöI—KöIómöI€ómöI‘®öIü‘®öIiµöIiµöIÅ×öIpÅ×öIc÷Iìc÷I;÷Iô;÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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‡ïõIåöI�åöIƒDöI ƒDöIýHöIýHöI·möI€·möIU®öIüU®öIϲöIϲöI‰×öIp‰×öI'÷Iì'÷I¡÷Iô¡÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI1£øI¬ 1£øI yùIp  yùICeúIt CeúI—ûIT —ûIgÙûI˜gÙûIìûIœ€} 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‡ïõI'öI�'öIÅDöI ÅDöI©KöI©KöIómöI€ómöI‘®öIü‘®öI{µöI{µöIÅ×öIpÅ×öIc÷Iìc÷IM÷IôM÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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‡ïõIßöI�ßöI}DöI }DöIýHöIýHöI±möI€±möIO®öIüO®öIϲöIϲöIƒ×öIpƒ×öI!÷Iì!÷I¡÷Iô¡÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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‡ïõI'öI�'öIÅDöI ÅDöI©KöI©KöIùmöI€ùmöI—®öIü—®öI{µöI{µöIË×öIpË×öIi÷Iìi÷IM÷IôM÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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�ïõIßöI�ßöI}DöI }DöIIöIIöI«möI€«möII®öIüI®öIá²öIá²öI}×öIp}×öI÷Iì÷I³÷Iô³÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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�ïõI-öI�-öIËDöI ËDöIKöIKöIùmöI€ùmöI—®öIü—®öIQµöIQµöIË×öIpË×öIi÷Iìi÷I#÷Iô#÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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�ïõIÙöI�ÙöIwDöI wDöIIöIIöI«möI€«möII®öIüI®öIç²öIç²öI}×öIp}×öI÷Iì÷I¹÷Iô¹÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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�ïõI-öI�-öIËDöI ËDöIKöIKöIÿmöI€ÿmöI�®öIü�®öIQµöIQµöIÑ×öIpÑ×öIo÷Iìo÷I#÷Iô#÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÙöI�ÙöIwDöI wDöIýHöIýHöI¥möI€¥möIC®öIüC®öIϲöIϲöIw×öIpw×öI÷Iì÷I¡÷Iô¡÷I«`÷I`«`÷I …÷IÌ …÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI-öI�-öIËDöI ËDöI�KöI�KöIÿmöI€ÿmöI�®öIü�®öIoµöIoµöIÑ×öIpÑ×öIo÷Iìo÷IA÷IôA÷I±`÷I`±`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÓöI�ÓöIqDöI qDöIýHöIýHöI¥möI€¥möIC®öIüC®öIϲöIϲöIw×öIpw×öI÷Iì÷I¡÷Iô¡÷I½`÷I`½`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI³ûIT ³ûIƒØûI˜ƒØûI+ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÍöI�ÍöIkDöI kDöI IöI IöIŸmöI€ŸmöI=®öIü=®öIÛ²öIÛ²öIq×öIpq×öI÷Iì÷I­÷Iô­÷I«`÷I`«`÷I …÷IÌ …÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIsûIT sûICÙûI˜CÙûIëëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI3öI�3öIÑDöI ÑDöIyKöIyKöInöI€nöI£®öIü£®öIKµöIKµöI××öIp××öIu÷Iìu÷I÷Iô÷I±`÷I`±`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÍöI�ÍöIkDöI kDöI IöI IöI™möI€™möI7®öIü7®öIÛ²öIÛ²öIk×öIpk×öI ÷Iì ÷I­÷Iô­÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI9öI�9öI×DöI ×DöIyKöIyKöInöI€nöI£®öIü£®öIKµöIKµöI××öIp××öIu÷Iìu÷I÷Iô÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Ia¶÷Iìa¶÷I-øIL -øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÇöI�ÇöIeDöI eDöIñHöIñHöI™möI€™möI7®öIü7®öIòöIòöIk×öIpk×öI ÷Iì ÷I•÷Iô•÷I«`÷I`«`÷I …÷IÌ …÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIOÙûI˜OÙûI÷ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöI—KöI—KöI nöI€ nöI©®öIü©®öIiµöIiµöIÝ×öIpÝ×öI{÷Iì{÷I;÷Iô;÷I�`÷I`�`÷Ií„÷IÌí„÷Ia¶÷Iìa¶÷I-øIL -øI £øI¬  £øI5wùIp 5wùIGúIt GúI7ûIT 7ûI-×ûI˜-×ûIÕéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÇöI�ÇöIeDöI eDöIñHöIñHöI“möI€“möI1®öIü1®öIòöIòöIe×öIpe×öI÷Iì÷I•÷Iô•÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI9öI�9öI×DöI ×DöI‘KöI‘KöI nöI€ nöI©®öIü©®öIcµöIcµöIÝ×öIpÝ×öI{÷Iì{÷I5÷Iô5÷I±`÷I`±`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúIÝûIT ÝûIÓÖûI˜ÓÖûI{éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõIÁöI�ÁöI_DöI _DöIýHöIýHöI“möI€“möI1®öIü1®öIϲöIϲöIe×öIpe×öI÷Iì÷I¡÷Iô¡÷Iû^÷I`û^÷I[ƒ÷IÌ[ƒ÷I[¶÷I€[¶÷I'øIà'øI£øI@£øI;wùIÚ ;wùIGúIú GúIIûIÒ IûI?×ûIr ?×ûIçéûI’ €} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI?öI�?öIÝDöI ÝDöImKöImKöInöI€nöI¯®öIü¯®öI?µöI?µöIã×öIpã×öI�÷Iì�÷I÷Iô÷Iû^÷I`û^÷I[ƒ÷IÌ[ƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øI;wùIp ;wùIGúIt GúIIûIT IûI?×ûI˜?×ûIçéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI»öI�»öIYDöI YDöIýHöIýHöI�möI€�möI+®öIü+®öIϲöIϲöI_×öIp_×öIý÷Iìý÷I¡÷Iô¡÷Iû^÷I`û^÷I[ƒ÷IÌ[ƒ÷I[¶÷Iì[¶÷I'øIL 'øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúIãûIT ãûIÙÖûI˜ÙÖûI�éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LoïõI?öI�?öIÝDöI ÝDöImKöImKöInöI€nöI¯®öIü¯®öI?µöI?µöIã×öIpã×öI�÷Iì�÷I÷Iô÷Iû^÷I`û^÷I[ƒ÷IÌ[ƒ÷I[¶÷Iì[¶÷I'øIL 'øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúIÑûIT ÑûIÇÖûI˜ÇÖûIoéûIœ€} 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„iïõI‘öI‘öI/DöI|/DöI“IöI„“IöIÉLöIðÉLöISVöI|SVöI‰YöIH‰YöI]möIÔ]möIû­öIPû­öIe³öIXe³öI›¶öIÄ›¶öI%ÀöIP%ÀöI[ÃöI[ÃöI/×öI¨/×öIÍ÷I$ Í÷I7÷I, 7÷Im ÷I˜ m ÷I÷)÷I$ ÷)÷I--÷Ið --÷IoZ÷I| oZ÷I�`÷IX�`÷Iá„÷IÔá„÷I#²÷Iô#²÷IíÀ÷IÐíÀ÷I©õ÷I\©õ÷IûøI€ûøIù øI¬ù øIõ¢øI8õ¢øIÍxùIüÍxùIeúIeúI�rúIà�rúIËûIlËûI›ØûI°›ØûICëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIWöIèWöIõDöIdõDöI£KöIl£KöI-UöIè-UöIcXöI´cXöI#nöI@#nöIÁ®öI¼Á®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIõ×öI˜õ×öI“÷I“÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷IóÀ÷IDóÀ÷I¯õ÷IЯõ÷I øIô øI !øI  !øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúIûIàûIïØûI$ïØûI—ëûI(€} 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„iïõI‹öI‹öI)DöI|)DöI“IöI„“IöIÉLöIðÉLöISVöI|SVöI‰YöIH‰YöI]möIÔ]möIû­öIPû­öIe³öIXe³öI›¶öIÄ›¶öI%ÀöIP%ÀöI[ÃöI[ÃöI/×öI¨/×öIÍ÷I$ Í÷I7÷I, 7÷Im ÷I˜ m ÷I÷)÷I$ ÷)÷I--÷Ið --÷IoZ÷I| oZ÷I±`÷IX±`÷I…÷IÔ…÷IS²÷IôS²÷I­Á÷IЭÁ÷Iiö÷I\iö÷I©øI€©øI§$øI¬§$øIã¢øI8ã¢øI»xùIü»xùIõdúIõdúIorúIàorúIeûIleûI5ØûI°5ØûIÝêûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõIWöIèWöIõDöIdõDöI£KöIl£KöI-UöIè-UöIcXöI´cXöI)nöI@)nöIÇ®öI¼Ç®öIuµöIÄuµöIÿ¾öI@ÿ¾öI5ÂöI 5ÂöIû×öI˜û×öI™÷I™÷IG÷I G÷IÑ(÷I˜ Ñ(÷I,÷Id ,÷IIY÷Ið IY÷I�`÷IÌ �`÷Iá„÷IH á„÷I#²÷Ih#²÷I5Á÷ID5Á÷Iñõ÷IÐñõ÷IcøIôcøIa"øI a"øIï¢øI¬ï¢øIÇxùIpÇxùIeúIteúI{rúIT{rúI­ûIà­ûI}ØûI$}ØûI%ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIÁöIèÁöI_DöId_DöI©KöIl©KöI3UöIè3UöIiXöI´iXöI�möI@�möI+®öI¼+®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöI_×öI˜_×öIý÷Iý÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I‡`÷IÌ ‡`÷Iç„÷IH ç„÷I)²÷Ih)²÷IÏÀ÷IDÏÀ÷I‹õ÷IЋõ÷IMøIôMøIK øI K øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIûIàûIרûI$רûIëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIQöIèQöIïDöIdïDöI©KöIl©KöI3UöIè3UöIiXöI´iXöInöI@nöI»®öI¼»®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIï×öI˜ï×öI�÷I�÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I½`÷IÌ ½`÷I…÷IH …÷I_²÷Ih_²÷IÕÀ÷IDÕÀ÷I‘õ÷IБõ÷I_øIô_øI] øI ] øI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúIgûIàgûI7ÙûI$7ÙûIßëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI»öIè»öIYDöIdYDöI©KöIl©KöI3UöIè3UöIiXöI´iXöI�möI@�möI+®öI¼+®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöI_×öI˜_×öIý÷Iý÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷IÉÀ÷IDÉÀ÷I…õ÷IÐ…õ÷IGøIôGøIE øI E øIï¢øI¬ï¢øIÇxùIpÇxùIeúIteúI{rúIT{rúIÅûIàÅûI•ØûI$•ØûI=ëûI(€} 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$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?K@KBÀBÀD5E5GªGªIøIøKkLkNÞNÞPQQQSÄSÄU^V^XÑXÑZD[D]·]·_*`*b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†T‡T‰ljNj:Œ:Ž­Ž­� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxoïõIQöIèQöIïDöIdïDöI©KöIl©KöI3UöIè3UöIiXöI´iXöI#nöI@#nöIÁ®öI¼Á®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIõ×öI˜õ×öI“÷I“÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I‡`÷IÌ ‡`÷Iç„÷IH ç„÷I)²÷Ih)²÷IoÀ÷IDoÀ÷I+õ÷IÐ+õ÷IßøIôßøIÝøI ÝøIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIûûIàûûIËØûI$ËØûIsëûI(€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n&o&q—q—stvyvyxêxêz4{4}¥}¥€‚‡‚‡„ø„ø†‘‡‘‰ŠŒsŒsŽäŽä�U‘U“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÇöIÐÇöIeDöILeDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI“möI(“möI1®öI¤1®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIe×öI€e×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷Iÿ½÷I,ÿ½÷I{øI˜{øI£øI$£øIÙxùIèÙxùIeúIìeúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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$Y%Y'Ð'Ð)G*G,¾,¾.5/51¬1¬3#4#6š6š89;ˆ;ˆ=ÿ=ÿ?v@vBíBíDdEdGÛGÛITJTLËLËNBOBQ¹Q¹S0T0V§V§XY[•[•] ^ `ƒ`ƒbúbúdqeqgègèi_j_lÖlÖnHoHqºqºs,t,vžvžxy{‚{‚}ô}ôf€f‚؂؄J…J‡¼‡¼‰.Š.Œ Œ Ž�‘„‘„“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIEöIÐEöIãDöILãDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöIµ®öI¤µ®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIé×öI€é×öI‡÷Iü‡÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IÕ½÷I,Õ½÷IsøI˜søI£øI$£øIëxùIèëxùI%eúIì%eúIyûIÌyûIIÙûIIÙûIñëûI€} 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$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélén_o_qÕqÕsKtKvÁvÁx7y7{­{­}#~#€™€™‚ƒ…………‡û‡û‰qŠqŒçŒçŽ]�]‘Ó‘Ó“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÁöIÐÁöI_DöIL_DöIµKöITµKöI?UöIÐ?UöIuXöIœuXöI“möI(“möI1®öI¤1®öI‡µöI¬‡µöI¿öI(¿öIGÂöIôGÂöIe×öI€e×öI÷Iü÷IY÷I Y÷Iã(÷I€ ã(÷I,÷IL ,÷I[Y÷IØ [Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I™À÷I,™À÷IøI˜øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$;%;'°'°)%*%,š,š./1„1„3ù3ù5n6n8ã8ã:X;X=Í=Í?B@BB·B·D,E,G¡G¡IJL�L�NOQwQwSìSìUaVaXÖXÖZK[K]À]À_5`5bªbªdeg”g”i j l~l~nðnðpbqbsÔsÔuFvFx¸x¸z*{*}œ}œ€‚€‚€„ò„ò†d‡d‰Ö‰Ö‹HŒHŽºŽº�,‘,“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIKöIÐKöIéDöILéDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöInöI(nöI»®öI¤»®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIï×öI€ï×öI�÷Iü�÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷IcÀ÷I,cÀ÷IýøI˜ýøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIûIÌûIÑØûIÑØûIyëûI€} 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"×"×$M%M'Ã'Ã)9*9,¯,¯.%/%1›1›346‡6‡8ý8ý:s;s=é=é?_@_BÕBÕDKEKGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÇöIÐÇöIeDöILeDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI™möI(™möI7®öI¤7®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIk×öI€k×öI ÷Iü ÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷Ió½÷I,ó½÷IKøI˜KøI£øI$£øIßxùIèßxùIeúIìeúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIWJWLÌLÌNAOAQ¶Q¶S+T+V V XY[Š[Š]ÿ]ÿ_t`tbébéd^e^gÓgÓiHjHl½l½n2o2q§q§stv‘v‘xy{{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÇöIÐÇöIeDöILeDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI™möI(™möI7®öI¤7®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIk×öI€k×öI ÷Iü ÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I‡À÷I,‡À÷IçøI˜çøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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·luïõIEöIÐEöIãDöILãDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöIµ®öI¤µ®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIé×öI€é×öI‡÷Iü‡÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷IcÀ÷I,cÀ÷I!øI˜!øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l{ïõIÍöIÐÍöIkDöILkDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIŸmöI(ŸmöI=®öI¤=®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIq×öI€q×öI÷Iü÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷Iá½÷I,á½÷I!øI˜!øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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·l{ïõI9öIÐ9öI×DöIL×DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI nöI( nöI©®öI¤©®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIÝ×öI€Ý×öI{÷Iü{÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷IÕ½÷I,Õ½÷I¯øI˜¯øI £øI$ £øIåxùIèåxùIeúIìeúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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·l{ïõIÍöIÐÍöIkDöILkDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIŸmöI(ŸmöI=®öI¤=®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIq×öI€q×öI÷Iü÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I{À÷I,{À÷I·øI˜·øIã¢øI$ã¢øI»xùIè»xùIõdúIìõdúIeûIÌeûI5ØûI5ØûIÝêûI€} 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·l{ïõI?öIÐ?öIÝDöILÝDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI nöI( nöI©®öI¤©®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIÝ×öI€Ý×öI{÷Iü{÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IcÀ÷I,cÀ÷I?øI˜?øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI­ûIÌ­ûI}ØûI}ØûI%ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations1Çp�ïõIÓöIØÓöIqDöITqDöI»KöI\»KöIEUöIØEUöI{XöI¤{XöI¥möI0¥möIC®öI¬C®öI�µöI´�µöI¿öI0¿öIMÂöIüMÂöIw×öIˆw×öI÷I÷I_÷I _÷Ié(÷Iˆ é(÷I,÷IT ,÷IaY÷Ià aY÷Iw[÷I¼ w[÷Iï^÷IH ï^÷IOƒ÷I´ Oƒ÷I‘°÷IÔ‘°÷IÛ½÷I°Û½÷IøIøI£øI¨£øIßxùIlßxùIeúIpeúIûIPûIãØûI”ãØûI‹ëûI˜€} 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$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI3öIÐ3öIÑDöILÑDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöI£®öI¤£®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöI××öI€××öIu÷Iüu÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷IÛ½÷I,Û½÷I øI˜ øI£øI$£øI÷xùIè÷xùI1eúIì1eúIyûIÌyûIIÙûIIÙûIñëûI€} 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·l�ïõIÓöIÐÓöIqDöILqDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIŸmöI(ŸmöI=®öI¤=®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIq×öI€q×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷IoÀ÷I,oÀ÷I™øI˜™øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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·l�ïõI9öIÐ9öI×DöIL×DöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöInöI(nöI£®öI¤£®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöI××öI€××öIu÷Iüu÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷IoÀ÷I,oÀ÷IŸøI˜ŸøI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIרûIרûIëûI€} 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�ïõIÙöIÐÙöIwDöILwDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI«möI(«möII®öI¤I®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI}×öI€}×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷Ió½÷I,ó½÷IÏøI˜ÏøI£øI$£øIëxùIèëxùI%eúIì%eúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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�ïõI-öIÐ-öIËDöILËDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIùmöI(ùmöI—®öI¤—®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIË×öI€Ë×öIi÷Iüi÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷Iù½÷I,ù½÷IÕøI˜ÕøI%£øI$%£øIýxùIèýxùI7eúIì7eúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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�ïõIÙöIÐÙöIwDöILwDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI¥möI(¥möIC®öI¤C®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIw×öI€w×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I“À÷I,“À÷IqøI˜qøI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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�ïõI-öIÐ-öIËDöILËDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIÿmöI(ÿmöI�®öI¤�®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIÑ×öI€Ñ×öIo÷Iüo÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I™À÷I,™À÷IqøI˜qøI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIéØûIéØûI‘ëûI€} 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$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡�‡�‰ŠŒwŒwŽëŽë�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIßöIÐßöI}DöIL}DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI«möI(«möII®öI¤I®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI}×öI€}×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I ¾÷I, ¾÷I›øI˜›øI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l‡ïõI'öIÐ'öIÅDöILÅDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIómöI(ómöI‘®öI¤‘®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIÅ×öI€Å×öIc÷Iüc÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I¾÷I,¾÷I›øI˜›øI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l‡ïõIÙöIÐÙöIwDöILwDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI«möI(«möII®öI¤I®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI}×öI€}×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I½À÷I,½À÷IIøI˜IøI£øI$£øIßxùIèßxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l‡ïõI'öIÐ'öIÅDöILÅDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIùmöI(ùmöI—®öI¤—®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIË×öI€Ë×öIi÷Iüi÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I½À÷I,½À÷IIøI˜IøI £øI$ £øIåxùIèåxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I,J,LŸLŸNOQ…Q…SøSøUkVkXÞXÞZQ[Q]Ä]Ä_7`7bªbªdeg�g�ijlvlvnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI­DöI­DöIýHöI$ýHöIámöI�ámöI®öI®öIϲöI$ϲöI³×öI�³×öIQ÷IQ÷I¡÷I$¡÷I=_÷I�=_÷I�ƒ÷Iü�ƒ÷Iy¶÷I y¶÷IEøI| EøI£øIÜ £øIñxùI  ñxùI+eúI¤ +eúIïûI„ ïûI¿ØûIÈ¿ØûIgëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI ¿DöI©KöI©KöIómöI€ómöI‘®öIü‘®öI{µöI{µöIÅ×öIpÅ×öIc÷Iìc÷IM÷IôM÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI³DöIýHöI$ýHöIçmöI�çmöI…®öI…®öIϲöI$ϲöI¹×öI�¹×öIW÷IW÷I¡÷I$¡÷I=_÷I�=_÷I�ƒ÷Iü�ƒ÷Iy¶÷I y¶÷IEøI| EøI£øIÜ £øIßxùI  ßxùIeúI¤ eúI‰ûI„ ‰ûIYØûIÈYØûIëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI ¿DöI©KöI©KöIómöI€ómöI‘®öIü‘®öI{µöI{µöIÅ×öIpÅ×öIc÷Iìc÷IM÷IôM÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIñöI�ñöI�DöI �DöIýHöIýHöIÃmöI€ÃmöIa®öIüa®öIϲöIϲöI•×öIp•×öI3÷Iì3÷I¡÷Iô¡÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöI—KöI—KöIómöI€ómöI‘®öIü‘®öIiµöIiµöIÅ×öIpÅ×öIc÷Iìc÷I;÷Iô;÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI1£øI¬ 1£øI yùIp  yùICeúIt CeúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIëöI�ëöI‰DöI ‰DöIýHöIýHöI½möI€½möI[®öIü[®öIϲöIϲöI�×öIp�×öI-÷Iì-÷I¡÷Iô¡÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöI—KöI—KöIùmöI€ùmöI—®öIü—®öIiµöIiµöIË×öIpË×öIi÷Iìi÷I;÷Iô;÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIëöI�ëöI‰DöI ‰DöIýHöIýHöI·möI€·möIU®öIüU®öIϲöIϲöI‰×öIp‰×öI'÷Iì'÷I¡÷Iô¡÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI-öI�-öIËDöI ËDöI£KöI£KöIùmöI€ùmöI—®öIü—®öIuµöIuµöIË×öIpË×öIi÷Iìi÷IG÷IôG÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI1£øI¬ 1£øI yùIp  yùICeúIt CeúI—ûIT —ûIgÙûI˜gÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIåöI�åöIƒDöI ƒDöIýHöIýHöI·möI€·möIU®öIüU®öIϲöIϲöI‰×öIp‰×öI'÷Iì'÷I¡÷Iô¡÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI-öI�-öIËDöI ËDöI£KöI£KöIÿmöI€ÿmöI�®öIü�®öIuµöIuµöIÑ×öIpÑ×öIo÷Iìo÷IG÷IôG÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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�ïõI-öI�-öIËDöI ËDöIKöIKöIÿmöI€ÿmöI�®öIü�®öIQµöIQµöIÑ×öIpÑ×öIo÷Iìo÷I#÷Iô#÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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�ïõIßöI�ßöI}DöI }DöIIöIIöI±möI€±möIO®öIüO®öIá²öIá²öIƒ×öIpƒ×öI!÷Iì!÷I³÷Iô³÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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�ïõI3öI�3öIÑDöI ÑDöIKöIKöIÿmöI€ÿmöI�®öIü�®öIQµöIQµöIÑ×öIpÑ×öIo÷Iìo÷I#÷Iô#÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIÙöI�ÙöIwDöI wDöIýHöIýHöI«möI€«möII®öIüI®öIϲöIϲöI}×öIp}×öI÷Iì÷I¡÷Iô¡÷IŸ`÷I`Ÿ`÷Iÿ„÷IÌÿ„÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI3öI�3öIÑDöI ÑDöI�KöI�KöInöI€nöI£®öIü£®öIoµöIoµöI××öIp××öIu÷Iìu÷IA÷IôA÷I½`÷I`½`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIÙöI�ÙöIwDöI wDöIýHöIýHöI¥möI€¥möIC®öIüC®öIϲöIϲöIw×öIpw×öI÷Iì÷I¡÷Iô¡÷I±`÷I`±`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI颸I¬ 颸IÁxùIp ÁxùIûdúIt ûdúIkûIT kûI;ØûI˜;ØûIãêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI9öI�9öI×DöI ×DöI—KöI—KöInöI€nöI£®öIü£®öIiµöIiµöI××öIp××öIu÷Iìu÷I;÷Iô;÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Ig¶÷Iìg¶÷I3øIL 3øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI³ûIT ³ûIƒØûI˜ƒØûI+ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÓöI�ÓöIqDöI qDöIIöIIöI¥möI€¥möIC®öIüC®öIÕ²öIÕ²öIw×öIpw×öI÷Iì÷I§÷Iô§÷I¥`÷I`¥`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI9öI�9öI×DöI ×DöIsKöIsKöI nöI€ nöI©®öIü©®öIEµöIEµöIÝ×öIpÝ×öI{÷Iì{÷I÷Iô÷I½`÷I`½`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIsûIT sûICÙûI˜CÙûIëëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÓöI�ÓöIqDöI qDöIIöIIöIŸmöI€ŸmöI=®öIü=®öIÕ²öIÕ²öIq×öIpq×öI÷Iì÷I§÷Iô§÷I½`÷I`½`÷I…÷IÌ…÷Ig¶÷Iìg¶÷I3øIL 3øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÍöI�ÍöIkDöI kDöIñHöIñHöIŸmöI€ŸmöI=®öIü=®öIòöIòöIq×öIpq×öI÷Iì÷I•÷Iô•÷I¥`÷I`¥`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIßxùIp ßxùIeúIt eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI?öI�?öIÝDöI ÝDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷IŸ`÷I`Ÿ`÷Iÿ„÷IÌÿ„÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIGwùIp GwùI%GúIt %GúI…ûIT …ûI{×ûI˜{×ûI#êûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõI?öI�?öIÝDöI ÝDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷I½`÷I`½`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øI/wùIp /wùI GúIt  GúI ûIT  ûI×ûI˜×ûI«éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÇöI�ÇöIeDöI eDöI÷HöI÷HöI™möI€™möI7®öIü7®öIɲöIɲöIk×öIpk×öI ÷Iì ÷I›÷Iô›÷I_÷I`_÷Iaƒ÷IÌaƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øI;wùIp ;wùIGúIt GúIIûIT IûI?×ûI˜?×ûIçéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIEöI�EöIãDöI ãDöImKöImKöInöI€nöI¯®öIü¯®öI?µöI?µöIã×öIpã×öI�÷Iì�÷I÷Iô÷I_÷I`_÷Iaƒ÷IÌaƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øI;wùIp ;wùIGúIt GúIOûIT OûIE×ûI˜E×ûIíéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÁöI�ÁöI_DöI _DöI÷HöI÷HöI“möI€“möI1®öIü1®öIɲöIɲöIe×öIpe×öI÷Iì÷I›÷Iô›÷I_÷I`_÷Iaƒ÷IÌaƒ÷I[¶÷Iì[¶÷I'øIL 'øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúIÑûIT ÑûIÇÖûI˜ÇÖûIoéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIEöI�EöIãDöI ãDöImKöImKöInöI€nöIµ®öIüµ®öI?µöI?µöIé×öIpé×öI‡÷Iì‡÷I÷Iô÷I_÷I`_÷Iaƒ÷IÌaƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øI)wùIp )wùIGúIt GúIãûIT ãûIÙÖûI˜ÙÖûI�éûIœ€} 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):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„iïõI‘öI‘öI/DöI|/DöI“IöI„“IöIÉLöIðÉLöISVöI|SVöI‰YöIH‰YöIcmöIÔcmöI®öIP®öIe³öIXe³öI›¶öIÄ›¶öI%ÀöIP%ÀöI[ÃöI[ÃöI5×öI¨5×öIÓ÷I$ Ó÷I7÷I, 7÷Im ÷I˜ m ÷I÷)÷I$ ÷)÷I--÷Ið --÷IoZ÷I| oZ÷I¥`÷IX¥`÷I…÷IÔ…÷IG²÷IôG²÷IuÀ÷IÐuÀ÷I1õ÷I\1õ÷I‘øI€‘øI�øI¬�øI£øI8£øIëxùIüëxùI%eúI%eúIŸrúIàŸrúIIûIlIûIÙûI°ÙûIÁëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI]öIè]öIûDöIdûDöI©KöIl©KöI3UöIè3UöIiXöI´iXöI)nöI@)nöIÇ®öI¼Ç®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIû×öI˜û×öI™÷I™÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I�`÷IÌ �`÷Iá„÷IH á„÷I#²÷Ih#²÷IÉÀ÷IDÉÀ÷I…õ÷IÐ…õ÷I5øIô5øI3 øI 3 øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúI ûIà ûIÝØûI$ÝØûI…ëûI(€} 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„iïõI‘öI‘öI/DöI|/DöI“IöI„“IöIÉLöIðÉLöISVöI|SVöI‰YöIH‰YöI]möIÔ]möIû­öIPû­öIe³öIXe³öI›¶öIÄ›¶öI%ÀöIP%ÀöI[ÃöI[ÃöI/×öI¨/×öIÍ÷I$ Í÷I7÷I, 7÷Im ÷I˜ m ÷I÷)÷I$ ÷)÷I--÷Ið --÷IoZ÷I| oZ÷I™`÷IX™`÷Iù„÷IÔù„÷I;²÷Iô;²÷IuÀ÷IÐuÀ÷I1õ÷I\1õ÷I‘øI€‘øI�øI¬�øIû¢øI8û¢øIÓxùIüÓxùI eúI eúI‡rúIà‡rúIÝûIlÝûI­ØûI°­ØûIUëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxiïõI]öIè]öIûDöIdûDöI©KöIl©KöI3UöIè3UöIiXöI´iXöI/nöI@/nöIÍ®öI¼Í®öI{µöIÄ{µöI¿öI@¿öI;ÂöI ;ÂöIØöI˜ØöIŸ÷IŸ÷IM÷I M÷I×(÷I˜ ×(÷I ,÷Id  ,÷IOY÷Ið OY÷I‡`÷IÌ ‡`÷Iç„÷IH ç„÷I)²÷Ih)²÷I Á÷ID Á÷IÇõ÷IÐÇõ÷I‹øIô‹øI‰!øI ‰!øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúIËûIàËûI›ØûI$›ØûICëûI(€} 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$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†]‡]‰ЉЋCŒC޶޶�)‘)“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxoïõIÇöIèÇöIeDöIdeDöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI“möI@“möI1®öI¼1®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIe×öI˜e×öI÷I÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I«`÷IÌ «`÷I …÷IH  …÷IM²÷IhM²÷IÿÀ÷IDÿÀ÷I»õ÷Iлõ÷I+øIô+øI)!øI )!øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIûIàûIïØûI$ïØûI—ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIQöIèQöIïDöIdïDöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI#nöI@#nöIÁ®öI¼Á®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIõ×öI˜õ×öI“÷I“÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I‡`÷IÌ ‡`÷Iç„÷IH ç„÷I)²÷Ih)²÷I«À÷ID«À÷Igõ÷IÐgõ÷I‡øIô‡øI…øI …øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúI+ûIà+ûIûØûI$ûØûI£ëûI(€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnMoMq¿q¿s1t1v£v£xy{_{_}Ñ}ÑC€C‚µ‚µ„'…'‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxoïõIÁöIèÁöI_DöId_DöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI“möI@“möI1®öI¼1®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIe×öI˜e×öI÷I÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I�`÷IÌ �`÷Iá„÷IH á„÷I#²÷Ih#²÷IùÀ÷IDùÀ÷Iµõ÷Iеõ÷IøIôøI!øI !øI颸I¬颸IÁxùIpÁxùIûdúItûdúIurúITurúI§ûIà§ûIwØûI$wØûIëûI(€} 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$R%R'Æ'Æ):*:,®,®."/"1–1–3 4 6~6~8ò8ò:f;f=Ú=Ú?L@LB¾B¾D0E0G¢G¢IëIëK]L]NÏNÏPAQAS³S³UNVNXÀXÀZ2[2]¤]¤_`b�b�degtgtiæiækXlXnÊnÊp‡>‰¯‰¯‹ Œ ޑޑ�‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxoïõIWöIèWöIõDöIdõDöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI)nöI@)nöIÇ®öI¼Ç®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIû×öI˜û×öI™÷I™÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I“`÷IÌ “`÷Ió„÷IH ó„÷I5²÷Ih5²÷I{À÷ID{À÷I7õ÷IÐ7õ÷I—øIô—øI•øI •øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúI×ûIà×ûI§ØûI$§ØûIOëûI(€} 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·luïõIÇöIÐÇöIeDöILeDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI™möI(™möI7®öI¤7®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIk×öI€k×öI ÷Iü ÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷I¹¸÷I,¹¸÷I5øI˜5øI £øI$ £øIåxùIèåxùIeúIìeúIgûIÌgûI7ÙûI7ÙûIßëûI€} 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·luïõIKöIÐKöIéDöILéDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöInöI(nöI»®öI¤»®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIï×öI€ï×öI�÷Iü�÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷I¹¸÷I,¹¸÷IYøI˜YøI£øI$£øIßxùIèßxùIeúIìeúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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·luïõIÇöIÐÇöIeDöILeDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI™möI(™möI7®öI¤7®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIk×öI€k×öI ÷Iü ÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷IA»÷I,A»÷I¿øI˜¿øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIûûIÌûûIËØûIËØûIsëûI€} 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$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbäbädWeWgÊgÊi=j=l°l°n!o!q’q’stvtvtxåxåzV{V}Ç}Ç8€8‚©‚©„…‡Œ‡Œ‰ý‰ý‹nŒnŽߎß�P‘P“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIQöIÐQöIïDöILïDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI#nöI(#nöIÁ®öI¤Á®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIõ×öI€õ×öI“÷Iü“÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IG»÷I,G»÷IãøI˜ãøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIÅûIÌÅûI•ØûI•ØûI=ëûI€} 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$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÍöIÐÍöIkDöILkDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöIŸmöI(ŸmöI=®öI¤=®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIq×öI€q×öI÷Iü÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷I­¸÷I,­¸÷I øI˜ øI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIOÙûIOÙûI÷ëûI€} 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$j%j'â'â)Z*Z,Ò,Ò.J/J1Â1Â3:4:6²6²8*9*;¢;¢=>@’@’B C E‚E‚GúGúIrJrLèLèN^O^QÔQÔSJTJVÀVÀX6Y6[¬[¬]"^"`˜`˜bce„e„gúgúipjplælæn\o\qÒqÒsHtHv¾v¾x4y4{ª{ª} ~ €–€–‚ ƒ …‚…‚‡ø‡ø‰nŠnŒäŒäŽZ�Z‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIEöIÐEöIãDöILãDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöIµ®öI¤µ®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIé×öI€é×öI‡÷Iü‡÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I³¸÷I,³¸÷IqøI˜qøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÍöIÐÍöIkDöILkDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIŸmöI(ŸmöI=®öI¤=®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIq×öI€q×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷I5»÷I,5»÷I•øI˜•øI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIרûIרûIëûI€} 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$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡ЇЉþ‰þ‹rŒr޿޿�Z‘Z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIKöIÐKöIéDöILéDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöIµ®öI¤µ®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIé×öI€é×öI‡÷Iü‡÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷IG»÷I,G»÷IøI˜øIã¢øI$ã¢øI»xùIè»xùIõdúIìõdúIeûIÌeûI5ØûI5ØûIÝêûI€} 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$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÓöIÐÓöIqDöILqDöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI¥möI(¥möIC®öI¤C®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIw×öI€w×öI÷Iü÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I­¸÷I,­¸÷IeøI˜eøI £øI$ £øIåxùIèåxùIeúIìeúICûIÌCûIÙûIÙûI»ëûI€} 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$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn/o/q¤q¤stvŽvŽxy{x{x}í}íb€b‚ׂׄL…L‡Á‡Á‰6Š6Œ«Œ«Ž � ‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI?öIÐ?öIÝDöILÝDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI¯®öI¤¯®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIã×öI€ã×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I³¸÷I,³¸÷I�øI˜�øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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$]%]'Ò'Ò)G*G,¼,¼.1/11¦1¦346�6�89;z;z=ï=ï?d@dBÙBÙDNENGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒyŒyŽîŽî�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÓöIÐÓöIqDöILqDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIŸmöI(ŸmöI=®öI¤=®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIq×öI€q×öI÷Iü÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I;»÷I,;»÷IõøI˜õøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIEöIÐEöIãDöILãDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI¯®öI¤¯®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIã×öI€ã×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷IA»÷I,A»÷IøI˜øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIËûIÌËûI›ØûI›ØûICëûI€} 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$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIÙöIÐÙöIwDöILwDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI«möI(«möII®öI¤I®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöI}×öI€}×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I­¸÷I,­¸÷IøI˜øI £øI$ £øIåxùIèåxùIeúIìeúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâI[J[LÒLÒNIOIQÀQÀS7T7V®V®X%Y%[œ[œ]^`Š`ŠbcexexgïgïifjflÝlÝnOoOqÁqÁs3t3v¥v¥xy{‰{‰}û}ûm€m‚߂߄Q…Q‡ÇÉ5Š5Œ§Œ§Ž�‘‹‘‹“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIÙöIÐÙöIwDöILwDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI¥möI(¥möIC®öI¤C®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIw×öI€w×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷IG»÷I,G»÷IµøI˜µøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI³ûI̳ûIƒØûIƒØûI+ëûI€} 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$V%V'Ê'Ê)>*>,²,².&/&1š1š346‚6‚8ö8ö:j;j=Þ=Þ?P@PBÂBÂD4E4G¦G¦IJLŠLŠNüNüPnQnSàSàURVRXÄXÄZ6[6]¨]¨_`bŒbŒdþdþfpgpiâiâkTlTnÆnÆp8q8sªsªuvxŽxŽz{}r}rää�V‚V„ȄȆ:‡:‰«‰«‹ŒŽ�Ž��þ�þ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI9öIÐ9öI×DöIL×DöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI nöI( nöI©®öI¤©®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIÝ×öI€Ý×öI{÷Iü{÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷IG»÷I,G»÷IsøI˜søI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIßöIÐßöI}DöIL}DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI«möI(«möII®öI¤I®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI}×öI€}×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I³¸÷I,³¸÷IÍøI˜ÍøI%£øI$%£øIýxùIèýxùI7eúIì7eúIûIÌûIOÙûIOÙûI÷ëûI€} 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�ïõI3öIÐ3öIÑDöILÑDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöIÿmöI(ÿmöI�®öI¤�®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIÑ×öI€Ñ×öIo÷Iüo÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I³¸÷I,³¸÷I‘øI˜‘øI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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�ïõIÙöIÐÙöIwDöILwDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI«möI(«möII®öI¤I®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI}×öI€}×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IY»÷I,Y»÷IuøI˜uøI £øI$ £øIåxùIèåxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI3öIÐ3öIÑDöILÑDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöInöI(nöI£®öI¤£®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI××öI€××öIu÷Iüu÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷IS»÷I,S»÷I3øI˜3øI£øI$£øIßxùIèßxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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$m%m'á'á)U*U,É,É.=/=1±1±3%4%6™6™8 9 ;�;�=õ=õ?i@iBÝBÝDQEQGÅGÅI9J9L«L«NOQ�Q�STVsVsXåXåZW[W]É]É_;`;b­b­deg‘g‘ijlulunçnçpYqYsËsËu=v=x¯x¯z!{!}“}“€‚w‚w„é„é†[‡[‰͉͋?Œ?ޱޱ�#‘#“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIåöIÐåöIƒDöILƒDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI±möI(±möIO®öI¤O®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIƒ×öI€ƒ×öI!÷Iü!÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I¹¸÷I,¹¸÷I�øI˜�øI+£øI$+£øIyùIèyùI=eúIì=eúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþIvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI'öIÐ'öIÅDöILÅDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIùmöI(ùmöI—®öI¤—®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIË×öI€Ë×öIi÷Iüi÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷I¹¸÷I,¹¸÷IEøI˜EøI£øI$£øIßxùIèßxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?Q@QBÄBÄD7E7GªGªIJL�L�NOQvQvSéSéU\V\XÏXÏZB[B]µ]µ_(`(b›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIßöIÐßöI}DöIL}DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI±möI(±möIO®öI¤O®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIƒ×öI€ƒ×öI!÷Iü!÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷Ie»÷I,e»÷I5øI˜5øI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIéØûIéØûI‘ëûI€} 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$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI-öIÐ-öIËDöILËDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIÿmöI(ÿmöI�®öI¤�®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIÑ×öI€Ñ×öIo÷Iüo÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷Ie»÷I,e»÷IóøI˜óøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIwûIÌwûIGØûIGØûIïêûI€} 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$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¾=¾?/@/B B DEG‚G‚IóIóKdLdNÕNÕPFQFS·S·U(V(X™X™Z [ ]{]{_ì_ìa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI­DöI­DöIýHöI$ýHöIÛmöI�ÛmöIy®öIy®öIϲöI$ϲöI­×öI�­×öIK÷IK÷I¡÷I$¡÷I½`÷I�½`÷I…÷Iü…÷Iy¶÷I y¶÷IEøI| EøI7£øIÜ 7£øIyùI  yùIIeúI¤ IeúImûI„ mûI=ÙûIÈ=ÙûIåëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöI©KöI©KöIómöI€ómöI‘®öIü‘®öI{µöI{µöIÅ×öIpÅ×öIc÷Iìc÷IM÷IôM÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI­DöI­DöIýHöI$ýHöIámöI�ámöI®öI®öIϲöI$ϲöI³×öI�³×öIQ÷IQ÷I¡÷I$¡÷IÉ`÷I�É`÷I)…÷Iü)…÷Iy¶÷I y¶÷IEøI| EøI£øIÜ £øI÷xùI  ÷xùI1eúI¤ 1eúIûI„ ûIÑØûIÈÑØûIyëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöI£KöI£KöIùmöI€ùmöI—®öIü—®öIuµöIuµöIË×öIpË×öIi÷Iìi÷IG÷IôG÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI÷öI�÷öI•DöI •DöIýHöIýHöIÃmöI€ÃmöIa®öIüa®öIϲöIϲöI•×öIp•×öI3÷Iì3÷I¡÷Iô¡÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI-öI�-öIËDöI ËDöI‘KöI‘KöIùmöI€ùmöI—®öIü—®öIcµöIcµöIË×öIpË×öIi÷Iìi÷I5÷Iô5÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIñöI�ñöI�DöI �DöIýHöIýHöIÃmöI€ÃmöIa®öIüa®öIϲöIϲöI•×öIp•×öI3÷Iì3÷I¡÷Iô¡÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI-öI�-öIËDöI ËDöI—KöI—KöIÿmöI€ÿmöI�®öIü�®öIiµöIiµöIÑ×öIpÑ×öIo÷Iìo÷I;÷Iô;÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI-öI�-öIËDöI ËDöI£KöI£KöIÿmöI€ÿmöI�®öIü�®öIuµöIuµöIÑ×öIpÑ×öIo÷Iìo÷IG÷IôG÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIëöI�ëöI‰DöI ‰DöIýHöIýHöI½möI€½möI[®öIü[®öIϲöIϲöI�×öIp�×öI-÷Iì-÷I¡÷Iô¡÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI3öI�3öIÑDöI ÑDöI�KöI�KöIÿmöI€ÿmöI�®öIü�®öIoµöIoµöIÑ×öIpÑ×öIo÷Iìo÷IA÷IôA÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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�ïõIåöI�åöIƒDöI ƒDöI IöI IöI·möI€·möIU®öIüU®öIÛ²öIÛ²öI‰×öIp‰×öI'÷Iì'÷I­÷Iô­÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Im¶÷Iìm¶÷I9øIL 9øI+£øI¬ +£øIyùIp yùI=eúIt =eúI‘ûIT ‘ûIaÙûI˜aÙûI ìûIœ€} 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�ïõI3öI�3öIÑDöI ÑDöIKöIKöInöI€nöI£®öIü£®öIQµöIQµöI××öIp××öIu÷Iìu÷I#÷Iô#÷IÉ`÷I`É`÷I)…÷IÌ)…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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�ïõIåöI�åöIƒDöI ƒDöIIöIIöI±möI€±möIO®öIüO®öIá²öIá²öIƒ×öIpƒ×öI!÷Iì!÷I³÷Iô³÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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�ïõI9öI�9öI×DöI ×DöIKöIKöInöI€nöI£®öIü£®öIQµöIQµöI××öIp××öIu÷Iìu÷I#÷Iô#÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Im¶÷Iìm¶÷I9øIL 9øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIqûIT qûIAØûI˜AØûIéêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIßöI�ßöI}DöI }DöIýHöIýHöI±möI€±möIO®öIüO®öIϲöIϲöIƒ×öIpƒ×öI!÷Iì!÷I¡÷Iô¡÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI9öI�9öI×DöI ×DöI—KöI—KöI nöI€ nöI©®öIü©®öIiµöIiµöIÝ×öIpÝ×öI{÷Iì{÷I;÷Iô;÷IÉ`÷I`É`÷I)…÷IÌ)…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIßöI�ßöI}DöI }DöIýHöIýHöI«möI€«möII®öIüI®öIϲöIϲöI}×öIp}×öI÷Iì÷I¡÷Iô¡÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI9öI�9öI×DöI ×DöI—KöI—KöI nöI€ nöI©®öIü©®öIiµöIiµöIÝ×öIpÝ×öI{÷Iì{÷I;÷Iô;÷Iá`÷I`á`÷IA…÷IÌA…÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÙöI�ÙöIwDöI wDöIIöIIöI«möI€«möII®öIüI®öIÕ²öIÕ²öI}×öIp}×öI÷Iì÷I§÷Iô§÷I™`÷I`™`÷Iù„÷IÌù„÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI?öI�?öIÝDöI ÝDöIsKöIsKöInöI€nöI¯®öIü¯®öIEµöIEµöIã×öIpã×öI�÷Iì�÷I÷Iô÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÓöI�ÓöIqDöI qDöIIöIIöI¥möI€¥möIC®öIüC®öIÕ²öIÕ²öIw×öIpw×öI÷Iì÷I§÷Iô§÷I«`÷I`«`÷I …÷IÌ …÷Ig¶÷Iìg¶÷I3øIL 3øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI­ûIT ­ûI}ØûI˜}ØûI%ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI?öI�?öIÝDöI ÝDöIsKöIsKöInöI€nöI¯®öIü¯®öIEµöIEµöIã×öIpã×öI�÷Iì�÷I÷Iô÷Iç`÷I`ç`÷IG…÷IÌG…÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÓöI�ÓöIqDöI qDöIñHöIñHöI¥möI€¥möIC®öIüC®öIòöIòöIw×öIpw×öI÷Iì÷I•÷Iô•÷I™`÷I`™`÷Iù„÷IÌù„÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIsûIT sûICÙûI˜CÙûIëëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIEöI�EöIãDöI ãDöI‹KöI‹KöInöI€nöI¯®öIü¯®öI]µöI]µöIã×öIpã×öI�÷Iì�÷I/÷Iô/÷I«`÷I`«`÷I …÷IÌ …÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øI;wùIp ;wùIGúIt GúIIûIT IûI?×ûI˜?×ûIçéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÍöI�ÍöIkDöI kDöIñHöIñHöIŸmöI€ŸmöI=®öIü=®öIòöIòöIq×öIpq×öI÷Iì÷I•÷Iô•÷I·`÷I`·`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIEöI�EöIãDöI ãDöI‹KöI‹KöInöI€nöIµ®öIüµ®öI]µöI]µöIé×öIpé×öI‡÷Iì‡÷I/÷Iô/÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Ia¶÷Iìa¶÷I-øIL -øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúIÑûIT ÑûIÇÖûI˜ÇÖûIoéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÍöI�ÍöIkDöI kDöI÷HöI÷HöI™möI€™möI7®öIü7®öIɲöIɲöIk×öIpk×öI ÷Iì ÷I›÷Iô›÷I_÷I`_÷Iaƒ÷IÌaƒ÷I[¶÷Iì[¶÷I'øIL 'øI£øI¬ £øIGwùIp GwùI%GúIt %GúI…ûIT …ûI{×ûI˜{×ûI#êûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIKöI�KöIéDöI éDöImKöImKöInöI€nöIµ®öIüµ®öI?µöI?µöIé×öIpé×öI‡÷Iì‡÷I÷Iô÷Io`÷I`o`÷IÏ„÷IÔÏ„÷I[¶÷Iô[¶÷I'øIT 'øIû¢øI´ û¢øI#wùIx #wùIGúI| GúIÑûI\ ÑûIÇÖûI ÇÖûIoéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIKöI�KöIéDöI éDöImKöImKöInöI€nöI»®öIü»®öI?µöI?µöIï×öIpï×öI�÷Iì�÷I÷Iô÷I]`÷I`]`÷I½„÷IÔ½„÷I[¶÷Iô[¶÷I'øIT 'øI颸I´ 颸IwùIx wùIïFúI| ïFúIkûI\ kûIaÖûI aÖûI éûI¤€} 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„oïõI—öI—öI5DöI|5DöI™IöI„™IöIÏLöIðÏLöIYVöI|YVöI�YöIH�YöIcmöIÔcmöI®öIP®öIk³öIXk³öI¡¶öIÄ¡¶öI+ÀöIP+ÀöIaÃöIaÃöI5×öI¨5×öIÓ÷I$ Ó÷I=÷I, =÷Is ÷I˜ s ÷Iý)÷I$ ý)÷I3-÷Ið 3-÷IuZ÷I| uZ÷IÉ`÷IXÉ`÷I)…÷IÔ)…÷Ik²÷Iôk²÷I{À÷IÐ{À÷I7õ÷I\7õ÷I»øI€»øI¹øI¬¹øI£øI8£øIëxùIüëxùI%eúI%eúIŸrúIàŸrúIIûIlIûIÙûI°ÙûIÁëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI]öIè]öIûDöIdûDöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI/nöI@/nöIÍ®öI¼Í®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIØöI˜ØöIŸ÷IŸ÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷Iu`÷IÌ u`÷IÕ„÷IH Õ„÷I²÷Ih²÷IŸÀ÷IDŸÀ÷I[õ÷IÐ[õ÷IWøIôWøIUøI UøI£øI¬£øIñxùIpñxùI+eúIt+eúI¥rúIT¥rúIsûIàsûICÙûI$CÙûIëëûI(€} 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„oïõI‘öI‘öI/DöI|/DöI™IöI„™IöIÏLöIðÏLöIYVöI|YVöI�YöIH�YöIcmöIÔcmöI®öIP®öIk³öIXk³öI¡¶öIÄ¡¶öI+ÀöIP+ÀöIaÃöIaÃöI5×öI¨5×öIÓ÷I$ Ó÷I=÷I, =÷Is ÷I˜ s ÷Iý)÷I$ ý)÷I3-÷Ið 3-÷IuZ÷I| uZ÷I™`÷IX™`÷Iù„÷IÔù„÷I;²÷Iô;²÷I“À÷IГÀ÷IOõ÷I\Oõ÷I3øI€3øI1øI¬1øIû¢øI8û¢øIÓxùIüÓxùI eúI eúI‡rúIà‡rúIÑûIlÑûI¡ØûI°¡ØûIIëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIcöIècöIEöIdEöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI5nöI@5nöIÓ®öI¼Ó®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIØöI˜ØöI¥÷I¥÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I«`÷IÌ «`÷I …÷IH  …÷IM²÷IhM²÷IáÀ÷IDáÀ÷I�õ÷IÐ�õ÷I­øIô­øI« øI « øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIûIàûIÑØûI$ÑØûIyëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIÇöIèÇöIeDöIdeDöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI™möI@™möI7®öI¼7®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIk×öI˜k×öI ÷I ÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷I)Á÷ID)Á÷Iåõ÷IÐåõ÷IøIôøI"øI "øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúIÅûIàÅûI•ØûI$•ØûI=ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIWöIèWöIõDöIdõDöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI)nöI@)nöIÇ®öI¼Ç®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIû×öI˜û×öI™÷I™÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I�`÷IÌ �`÷Iá„÷IH á„÷I#²÷Ih#²÷I�À÷ID�À÷I=õ÷IÐ=õ÷I»øIô»øI¹øI ¹øI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúIyûIàyûIIÙûI$IÙûIñëûI(€} 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$l%l'ä'ä)\*\,Ô,Ô.L/L1Ä1Ä3<4<6´6´8,9,;¤;¤=>@”@”B C E„E„GüGüItJtLêLêN`O`QÖQÖSLTLVÂVÂX8Y8[®[®]$^$`š`šbce†e†gügüirjrlèlèn`o`qÖqÖsLtLvÂvÂx8y8{„{„}ú}úp€p‚æ‚æ„\…\‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxoïõIÇöIèÇöIeDöIdeDöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI™möI@™möI7®öI¼7®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIk×öI˜k×öI ÷I ÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷IŸ`÷IÌ Ÿ`÷Iÿ„÷IH ÿ„÷IA²÷IhA²÷I#Á÷ID#Á÷Ißõ÷IÐßõ÷IñøIôñøIï!øI ï!øIÝ¢øI¬Ý¢øIµxùIpµxùIïdúItïdúIirúITirúI_ûIà_ûI/ØûI$/ØûI×êûI(€} 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$8%8'«'«)*,‘,‘./1w1w3ê3ê5]6]8Ð8Ð:C;C=¶=¶?'@'B˜B˜D E GzGzIÀIÀK1L1N¢N¢PQS„S„U V X‘X‘Z[]s]s_ä_äa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxoïõI]öIè]öIûDöIdûDöI¯KöIl¯KöI9UöIè9UöIoXöI´oXöI/nöI@/nöIÍ®öI¼Í®öI�µöIÄ�µöI ¿öI@ ¿öIAÂöI AÂöIØöI˜ØöIŸ÷IŸ÷IS÷I S÷IÝ(÷I˜ Ý(÷I,÷Id ,÷IUY÷Ið UY÷I±`÷IÌ ±`÷I…÷IH …÷IS²÷IhS²÷I�À÷ID�À÷I=õ÷IÐ=õ÷I»øIô»øI¹øI ¹øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIûIàûIÑØûI$ÑØûIyëûI(€} 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$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqzqzsîsîubvbxÖxÖz{}“}“€‚{‚{„ï„ï†އމŠŒvŒvŽêŽê�^‘^“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÍöIÐÍöIkDöILkDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIŸmöI(ŸmöI=®öI¤=®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIq×öI€q×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷Iɽ÷I,ɽ÷IIøI˜IøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIûIÌûIרûIרûIëûI€} 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$W%W'Ì'Ì)A*A,¶,¶.+/+1 1 346Š6Š8ÿ8ÿ:t;t=é=é?^@^BÓBÓDHEHG½G½I3J3L§L§NOQ�Q�STVwVwXëXëZ_[_]Ó]Ó_G`Gb»b»d/e/g£g£ijl‹l‹nÿnÿpsqssçsçu[v[xÏxÏzC{C}·}·+€+‚Ÿ‚Ÿ„…‡‰‡‰‰þ‰þ‹sŒsŽèŽè�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIQöIÐQöIïDöILïDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI#nöI(#nöIÁ®öI¤Á®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIõ×öI€õ×öI“÷Iü“÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷I¥½÷I,¥½÷IAøI˜AøI£øI$£øIßxùIèßxùIeúIìeúICûIÌCûIÙûIÙûI»ëûI€} 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$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbèbèd]e]gÒgÒiGjGl¼l¼n1o1q¦q¦stv�v�xy{z{z}ï}ïd€d‚قلN…N‡ÇÉ8Š8Œ­Œ­Ž"�"‘—‘—“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÍöIÐÍöIkDöILkDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIŸmöI(ŸmöI=®öI¤=®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIq×öI€q×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IWÀ÷I,WÀ÷IÓøI˜ÓøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIÅûIÌÅûI•ØûI•ØûI=ëûI€} 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$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?L@LBÁBÁD6E6G«G«I J L“L“NOQyQySìSìU_V_XÒXÒZE[E]¸]¸_+`+b�b�deg�g�ióiókelen×n×pIqIs»s»u-v-xŸxŸz{}ƒ}ƒõõ�g‚g„لنM‡M‰À‰À‹3Œ3ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIWöIÐWöIõDöILõDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI#nöI(#nöIÁ®öI¤Á®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIõ×öI€õ×öI“÷Iü“÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷I!À÷I,!À÷I¿øI˜¿øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÓöIÐÓöIqDöILqDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI¥möI(¥möIC®öI¤C®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIw×öI€w×öI÷Iü÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I½½÷I,½½÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜIUJULÌLÌNCOCQºQºS1T1V¨V¨XY[–[–] ^ `„`„bûbûdrergégéi`j`l×l×nIoIq»q»s-t-vŸvŸxy{ƒ{ƒ}õ}õg€g‚قلK…K‡½‡½‰/Š/Œ¡Œ¡Ž�‘…‘…“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIKöIÐKöIéDöILéDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöI»®öI¤»®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIï×öI€ï×öI�÷Iü�÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I±½÷I,±½÷I¿øI˜¿øI£øI$£øIßxùIèßxùIeúIìeúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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·l{ïõIÓöIÐÓöIqDöILqDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIŸmöI(ŸmöI=®öI¤=®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIq×öI€q×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IEÀ÷I,EÀ÷I£øI˜£øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l{ïõIQöIÐQöIïDöILïDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöI»®öI¤»®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIï×öI€ï×öI�÷Iü�÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I3À÷I,3À÷ICøI˜CøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI­ûIÌ­ûI}ØûI}ØûI%ëûI€} 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·l{ïõIÙöIÐÙöIwDöILwDöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI«möI(«möII®öI¤I®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI}×öI€}×öI÷Iü÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷Iɽ÷I,ɽ÷I…øI˜…øI £øI$ £øIåxùIèåxùIeúIìeúIIûIÌIûIÙûIÙûIÁëûI€} 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·l{ïõIEöIÐEöIãDöILãDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI¯®öI¤¯®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIã×öI€ã×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷IϽ÷I,Ͻ÷I‹øI˜‹øI£øI$£øI÷xùIè÷xùI1eúIì1eúIyûIÌyûIIÙûIIÙûIñëûI€} 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$j™j™lmo�o�qôqôsgtgvÚvÚxMyM{À{À}3~3€¦€¦‚ƒ…Œ…Œ‡ÿ‡ÿ‰qŠqŒãŒãŽU�U‘ǑǓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÙöIÐÙöIwDöILwDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI¥möI(¥möIC®öI¤C®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIw×öI€w×öI÷Iü÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷IWÀ÷I,WÀ÷IøI˜øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIEöIÐEöIãDöILãDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöIµ®öI¤µ®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIé×öI€é×öI‡÷Iü‡÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I]À÷I,]À÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIרûIרûIëûI€} 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$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIßöIÐßöI}DöIL}DöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI«möI(«möII®öI¤I®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöI}×öI€}×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷Iá½÷I,á½÷IQøI˜QøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b•b•deg{g{iîiîkalanÔnÔpGqGsºsºu-v-x x z{}†}†ùù�l‚l„߄߆R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI?öIÐ?öIÝDöILÝDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI nöI( nöI©®öI¤©®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIÝ×öI€Ý×öI{÷Iü{÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷Iç½÷I,ç½÷IQøI˜QøI%£øI$%£øIýxùIèýxùI7eúIì7eúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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�ïõI?öIÐ?öIÝDöILÝDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöInöI(nöI¯®öI¤¯®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIã×öI€ã×öI�÷Iü�÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I�À÷I,�À÷IíøI˜íøI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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$P%P'Æ'Æ)<*<,²,².(/(1ž1ž346Š6Š89;v;v=ì=ì?b@bBØBØDNENGÄGÄI:J:L®L®N"O"Q–Q–S T V~V~XòXòZf[f]Ú]Ú_N`NbÂbÂd6e6gªgªijl’l’noqzqzsîsîubvbxÖxÖzJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIåöIÐåöIƒDöILƒDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI±möI(±möIO®öI¤O®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIƒ×öI€ƒ×öI!÷Iü!÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷Iÿ½÷I,ÿ½÷IøI˜øI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIéØûIéØûI‘ëûI€} 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$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI3öIÐ3öIÑDöILÑDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöInöI(nöI£®öI¤£®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI××öI€××öIu÷Iüu÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷Iÿ½÷I,ÿ½÷IøI˜øI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãIZJZLÏLÏNDODQ¹Q¹S.T.V£V£XY[�[�]^`w`wbìbìdaeagÖgÖiKjKlÀlÀn5o5qªqªstv”v”x y {~{~}ó}óh€h‚݂݄R…R‡LJlj<Š<Œ±Œ±Ž&�&‘›‘›“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIßöIÐßöI}DöIL}DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI±möI(±möIO®öI¤O®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIƒ×öI€ƒ×öI!÷Iü!÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I¥À÷I,¥À÷I¿øI˜¿øI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI9öIÐ9öI×DöIL×DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI nöI( nöI©®öI¤©®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIÝ×öI€Ý×öI{÷Iü{÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷I¥À÷I,¥À÷I¿øI˜¿øI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunènèp[q[sÎsÎuAvAx´x´z'{'}š}š € ‚€‚€„ó„ó†f‡f‰ىًLŒLŽ¿Ž¿�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I¾÷I,¾÷IãøI˜ãøI£øI$£øIñxùIèñxùI+eúIì+eúIIûIÌIûIÙûIÙûIÁëûI€} 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$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI-öIÐ-öIËDöILËDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöIÿmöI(ÿmöI�®öI¤�®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIÑ×öI€Ñ×öIo÷Iüo÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I¾÷I,¾÷IéøI˜éøI£øI$£øI÷xùIè÷xùI1eúIì1eúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·l‡ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IÉÀ÷I,ÉÀ÷I‘øI˜‘øI £øI$ £øIåxùIèåxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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·l‡ïõI3öIÐ3öIÑDöILÑDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöInöI(nöI£®öI¤£®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI××öI€××öIu÷Iüu÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IÉÀ÷I,ÉÀ÷I—øI˜—øI£øI$£øIßxùIèßxùIeúIìeúI¿ûIÌ¿ûI�ØûI�ØûI7ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI öI� öI§DöI§DöIýHöI$ýHöIÕmöI�ÕmöIs®öIs®öIϲöI$ϲöI§×öI�§×öIE÷IE÷I¡÷I$¡÷IÉ`÷I�É`÷I)…÷Iü)…÷Iy¶÷I y¶÷IEøI| EøI7£øIÜ 7£øIyùI  yùIIeúI¤ IeúImûI„ mûI=ÙûIÈ=ÙûIåëûIÌ€} 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$…—…—‡ˆŠzŠzŒìŒìŽ^�^‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L“ïõI öI� öI§DöI§DöIýHöI$ýHöIÛmöI�ÛmöIy®öIy®öIϲöI$ϲöI­×öI�­×öIK÷IK÷I¡÷I$¡÷IÏ`÷I�Ï`÷I/…÷Iü/…÷Iy¶÷I y¶÷IEøI| EøI£øIÜ £øI÷xùI  ÷xùI1eúI¤ 1eúIõûI„ õûIÅØûIÈÅØûImëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI-öI�-öIËDöI ËDöI£KöI£KöIÿmöI€ÿmöI�®öIü�®öIuµöIuµöIÑ×öIpÑ×öIo÷Iìo÷IG÷IôG÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIýöI�ýöI›DöI ›DöIýHöIýHöIÉmöI€ÉmöIg®öIüg®öIϲöIϲöI›×öIp›×öI9÷Iì9÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI÷öI�÷öI•DöI •DöIýHöIýHöIÉmöI€ÉmöIg®öIüg®öIϲöIϲöI›×öIp›×öI9÷Iì9÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIÙxùIp ÙxùIeúIt eúIƒûIT ƒûISØûI˜SØûIûêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI3öI�3öIÑDöI ÑDöI‘KöI‘KöIÿmöI€ÿmöI�®öIü�®öIcµöIcµöIÑ×öIpÑ×öIo÷Iìo÷I5÷Iô5÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIñöI�ñöI�DöI �DöIýHöIýHöIÃmöI€ÃmöIa®öIüa®öIϲöIϲöI•×öIp•×öI3÷Iì3÷I¡÷Iô¡÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI3öI�3öIÑDöI ÑDöI�KöI�KöInöI€nöI£®öIü£®öIoµöIoµöI××öIp××öIu÷Iìu÷IA÷IôA÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIñöI�ñöI�DöI �DöIIöIIöI½möI€½möI[®öIü[®öIÕ²öIÕ²öI�×öIp�×öI-÷Iì-÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI9öI�9öI×DöI ×DöI�KöI�KöInöI€nöI£®öIü£®öIoµöIoµöI××öIp××öIu÷Iìu÷IA÷IôA÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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‡ïõIëöI�ëöI‰DöI ‰DöI IöI IöI½möI€½möI[®öIü[®öIÛ²öIÛ²öI�×öIp�×öI-÷Iì-÷I­÷Iô­÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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‡ïõIëöI�ëöI‰DöI ‰DöI IöI IöI·möI€·möIU®öIüU®öIÛ²öIÛ²öI‰×öIp‰×öI'÷Iì'÷I­÷Iô­÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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‡ïõI9öI�9öI×DöI ×DöI…KöI…KöI nöI€ nöI©®öIü©®öIWµöIWµöIÝ×öIpÝ×öI{÷Iì{÷I)÷Iô)÷Iç`÷I`ç`÷IG…÷IÌG…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI¹ûIT ¹ûI‰ØûI˜‰ØûI1ëûIœ€} 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�ïõIåöI�åöIƒDöI ƒDöIýHöIýHöI·möI€·möIU®öIüU®öIϲöIϲöI‰×öIp‰×öI'÷Iì'÷I¡÷Iô¡÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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�ïõI?öI�?öIÝDöI ÝDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIyûIT yûIIÙûI˜IÙûIñëûIœ€} 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�ïõIßöI�ßöI}DöI }DöIýHöIýHöI±möI€±möIO®öIüO®öIϲöIϲöIƒ×öIpƒ×öI!÷Iì!÷I¡÷Iô¡÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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�ïõI?öI�?öIÝDöI ÝDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷Ií`÷I`í`÷IM…÷IÌM…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIßöI�ßöI}DöI }DöIýHöIýHöI±möI€±möIO®öIüO®öIϲöIϲöIƒ×öIpƒ×öI!÷Iì!÷I¡÷Iô¡÷I“`÷I`“`÷Ió„÷IÌó„÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIEöI�EöIãDöI ãDöIyKöIyKöInöI€nöI¯®öIü¯®öIKµöIKµöIã×öIpã×öI�÷Iì�÷I÷Iô÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIñxùIp ñxùI+eúIt +eúI…ûIT …ûIUÙûI˜UÙûIýëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÙöI�ÙöIwDöI wDöIýHöIýHöI«möI€«möII®öIüI®öIϲöIϲöI}×öIp}×öI÷Iì÷I¡÷Iô¡÷I«`÷I`«`÷I …÷IÌ …÷Ig¶÷Iìg¶÷I3øIL 3øIã¢øI¬ ã¢øI»xùIp »xùIõdúIt õdúIkûIT kûI;ØûI˜;ØûIãêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÙöI�ÙöIwDöI wDöI÷HöI÷HöI¥möI€¥möIC®öIüC®öIɲöIɲöIw×öIpw×öI÷Iì÷I›÷Iô›÷I“`÷I`“`÷Ió„÷IÌó„÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIKöI�KöIéDöI éDöI…KöI…KöInöI€nöIµ®öIüµ®öIWµöIWµöIé×öIpé×öI‡÷Iì‡÷I)÷Iô)÷I«`÷I`«`÷I …÷IÌ …÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øI;wùIp ;wùIGúIt GúIIûIT IûI?×ûI˜?×ûIçéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIÓöI�ÓöIqDöI qDöIñHöIñHöI¥möI€¥möIC®öIüC®öIòöIòöIw×öIpw×öI÷Iì÷I•÷Iô•÷I±`÷I`±`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIKöI�KöIéDöI éDöI…KöI…KöInöI€nöI»®öIü»®öIWµöIWµöIï×öIpï×öI�÷Iì�÷I)÷Iô)÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Ia¶÷Iìa¶÷I-øIL -øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúIãûIT ãûIÙÖûI˜ÙÖûI�éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIKöI�KöIéDöI éDöImKöImKöInöI€nöI»®öIü»®öI?µöI?µöIï×öIpï×öI�÷Iì�÷I÷Iô÷Ii`÷I`i`÷IÉ„÷IÔÉ„÷Ia¶÷Iôa¶÷I-øIT -øI £øI´  £øI5wùIx 5wùIGúI| GúI+ûI\ +ûI!×ûI !×ûIÉéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÍöI�ÍöIkDöI kDöIñHöIñHöIŸmöI€ŸmöI=®öIü=®öIòöIòöIq×öIpq×öI÷Iì÷I•÷Iô•÷I_÷I`_÷Igƒ÷IÌgƒ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øI)wùIp )wùIGúIt GúIãûIT ãûIÙÖûI˜ÙÖûI�éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIQöI�QöIïDöI ïDöImKöImKöInöI€nöI»®öIü»®öI?µöI?µöIï×öIpï×öI�÷Iì�÷I÷Iô÷Ii`÷I`i`÷IÉ„÷IÔÉ„÷Ia¶÷Iôa¶÷I-øIT -øIõ¢øI´ õ¢øIwùIx wùIûFúI| ûFúI³ûI\ ³ûI©ÖûI ©ÖûIQéûI¤€} 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$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„oïõI—öI—öI5DöI|5DöI™IöI„™IöIÏLöIðÏLöIYVöI|YVöI�YöIH�YöIimöIÔimöI®öIP®öIk³öIXk³öI¡¶öIÄ¡¶öI+ÀöIP+ÀöIaÃöIaÃöI;×öI¨;×öIÙ÷I$ Ù÷I=÷I, =÷Is ÷I˜ s ÷Iý)÷I$ ý)÷I3-÷Ið 3-÷IuZ÷I| uZ÷IÃ`÷IXÃ`÷I#…÷IÔ#…÷Ie²÷Iôe²÷I‡À÷IЇÀ÷ICõ÷I\Cõ÷IåøI€åøIãøI¬ãøI£øI8£øI÷xùIü÷xùI1eúI1eúI«rúIà«rúI…ûIl…ûIUÙûI°UÙûIýëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIcöIècöIEöIdEöIµKöIlµKöI?UöIè?UöIuXöI´uXöI5nöI@5nöIÓ®öI¼Ó®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöIØöI˜ØöI¥÷I¥÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷IŸ`÷IÌ Ÿ`÷Iÿ„÷IH ÿ„÷IA²÷IhA²÷I‡À÷ID‡À÷ICõ÷IÐCõ÷IåøIôåøIãøI ãøI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúI7ûIà7ûIÙûI$ÙûI¯ëûI(€} 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$]%]'Ô'Ô)K*K,Â,Â.9/91„1„3û3û5r6r8é8é:`;`=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnXoXqÊqÊsŠ>Œ°Œ°Ž"�"‘”‘”“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1„oïõI—öI—öI5DöI|5DöI™IöI„™IöIÏLöIðÏLöIYVöI|YVöI�YöIH�YöIcmöIÔcmöI®öIP®öIk³öIXk³öI¡¶öIÄ¡¶öI+ÀöIP+ÀöIaÃöIaÃöI5×öI¨5×öIÓ÷I$ Ó÷I=÷I, =÷Is ÷I˜ s ÷Iý)÷I$ ý)÷I3-÷Ið 3-÷IuZ÷I| uZ÷I±`÷IX±`÷I…÷IÔ…÷IS²÷IôS²÷IÃÀ÷IÐÃÀ÷Iõ÷I\õ÷IøI€øI øI¬ øI£øI8£øIßxùIüßxùIeúIeúI“rúIà“rúI ûIl ûIÝØûI°ÝØûI…ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIiöIèiöIEöIdEöIµKöIlµKöI?UöIè?UöIuXöI´uXöI;nöI@;nöIÙ®öI¼Ù®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöI ØöI˜ ØöI«÷I«÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷I{`÷IÌ {`÷IÛ„÷IH Û„÷I²÷Ih²÷I·À÷ID·À÷Isõ÷IÐsõ÷IÕøIôÕøIÓøI ÓøIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIÝûIàÝûI­ØûI$­ØûIUëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIÍöIèÍöIkDöIdkDöIµKöIlµKöI?UöIè?UöIuXöI´uXöIŸmöI@ŸmöI=®öI¼=®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöIq×öI˜q×öI÷I÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷I�À÷ID�À÷IIõ÷IÐIõ÷IåøIôåøIãøI ãøI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúICûIàCûIÙûI$ÙûI»ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI]öIè]öIûDöIdûDöIµKöIlµKöI?UöIè?UöIuXöI´uXöI/nöI@/nöIÍ®öI¼Í®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöIØöI˜ØöIŸ÷IŸ÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷I«`÷IÌ «`÷I …÷IH  …÷IM²÷IhM²÷I�À÷ID�À÷IIõ÷IÐIõ÷IåøIôåøIãøI ãøI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúI=ûIà=ûI ÙûI$ ÙûIµëûI(€} 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¤$mí6¶ÿÈH‘ZÚ#£ìl µ 5 ~ þ G Ç À  � Ù  b ¢ ë +t´ bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)2)B)Z-b1reasonLocalCopyvalue10bytes10240patternContiguousinvocations1 €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ff à àZZÐÐ)F*F4¼4¼>2?2I¦I¦ST^�^�hØhØrxsx} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 bšWuoperation.6ipu_stack_copy_u64%:%R)2))2)*)B)Z-b1reasonLocalCopyvalue10bytes2048patternContiguousinvocations5 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðffààZZÔÔ N N È ÈBB¼¼66°°**¤¤ ˜ ˜"#%ˆ%ˆ'þ'þ)t*t,ê,ê.`/`1Ö1Ö3L4L6Â6Â8898;®;®=$>$@š@šBCE†E†GüGüIrJrLæLæNZOZQÎQÎSBTBV¶V¶X*Y*[ž[ž]^`†`†búbúdnengâgâiVjVlÊlÊn@o@q´q´s(t(vœvœxy{X{X}Ì}Ì@€@‚´‚´„(…(‡ȇȉ<Š<Œ°Œ°Ž$�$‘˜‘˜“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxoïõIÍöIèÍöIkDöIdkDöIµKöIlµKöI?UöIè?UöIuXöI´uXöIŸmöI@ŸmöI=®öI¼=®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöIq×öI˜q×öI÷I÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷I�À÷ID�À÷IIõ÷IÐIõ÷IåøIôåøIãøI ãøIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúI×ûIà×ûI§ØûI$§ØûIOëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIcöIècöIEöIdEöIµKöIlµKöI?UöIè?UöIuXöI´uXöI/nöI@/nöIÍ®öI¼Í®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöIØöI˜ØöIŸ÷IŸ÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷I�`÷IÌ �`÷Iá„÷IH á„÷I#²÷Ih#²÷I�À÷ID�À÷IIõ÷IÐIõ÷IåøIôåøIãøI ãøIï¢øI¬ï¢øIÇxùIpÇxùIeúIteúI{rúIT{rúIËûIàËûI›ØûI$›ØûICëûI(€} 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$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnænæpXqXsÊsÊuŒ>ްް�"‘"“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÓöIÐÓöIqDöILqDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöI¥möI(¥möIC®öI¤C®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIw×öI€w×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷I§¸÷I,§¸÷I?øI˜?øI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIéØûIéØûI‘ëûI€} 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$q%q'ç'ç)]*],Ó,Ó.I/I1¿1¿35456«6«8!9!;—;—= > @�@�BõBõDiEiGÝGÝIQJQLÅLÅN9O9Q­Q­S!T!V•V•X Y [}[}]ñ]ñ_e`ebÙbÙdMeMgÁgÁi5j5l©l©noq‘q‘stvyvyxíxíza{a}Õ}ÕI€I‚½‚½„1…1‡¥‡¥‰ŠŒ‹Œ‹ŽþŽþ�q‘q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIWöIÐWöIõDöILõDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI)nöI()nöIÇ®öI¤Ç®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIû×öI€û×öI™÷Iü™÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I§¸÷I,§¸÷IEøI˜EøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIÅûIÌÅûI•ØûI•ØûI=ëûI€} 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$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIÓöIÐÓöIqDöILqDöI»KöIT»KöIEUöIÐEUöI{XöIœ{XöIŸmöI(ŸmöI=®öI¤=®öI�µöI¬�µöI¿öI(¿öIMÂöIôMÂöIq×öI€q×öI÷Iü÷I_÷I _÷Ié(÷I€ é(÷I,÷IL ,÷IaY÷IØ aY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷I#»÷I,#»÷I½øI˜½øI颸I$颸IÁxùIèÁxùIûdúIìûdúI§ûI̧ûIwØûIwØûIëûI€} 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·luïõI]öIÐ]öIûDöILûDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI)nöI()nöIÇ®öI¤Ç®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIû×öI€û×öI™÷Iü™÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷I#»÷I,#»÷IÃøI˜ÃøIÝ¢øI$Ý¢øIµxùIèµxùIïdúIìïdúI_ûIÌ_ûI/ØûI/ØûI×êûI€} 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·l{ïõIÙöIÐÙöIwDöILwDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI«möI(«möII®öI¤I®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI}×öI€}×öI÷Iü÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷I§¸÷I,§¸÷IóøI˜óøI£øI$£øIëxùIèëxùI%eúIì%eúIsûIÌsûICÙûICÙûIëëûI€} 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·l{ïõIQöIÐQöIïDöILïDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöI»®öI¤»®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIï×öI€ï×öI�÷Iü�÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I§¸÷I,§¸÷I·øI˜·øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÙöIÐÙöIwDöILwDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI¥möI(¥möIC®öI¤C®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIw×öI€w×öI÷Iü÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I/»÷I,/»÷I}øI˜}øI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIÑØûIÑØûIyëûI€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n"o"q“q“stvuvuxæxæzW{W}È}È9€9‚ª‚ª„…‡�‡�‰þ‰þ‹oŒoŽàŽà�Q‘Q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIQöIÐQöIïDöILïDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI#nöI(#nöIÁ®öI¤Á®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIõ×öI€õ×öI“÷Iü“÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I/»÷I,/»÷I;øI˜;øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b b deg„g„iöiökhlhnÚnÚpLqLs¾s¾u0v0x¢x¢z{}†}†øø�j‚j„܄܆P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIßöIÐßöI}DöIL}DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI«möI(«möII®öI¤I®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI}×öI€}×öI÷Iü÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I­¸÷I,­¸÷I§øI˜§øI£øI$£øI÷xùIè÷xùI1eúIì1eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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$k%k'ã'ã)[*[,Ó,Ó.K/K1Ã1Ã3;4;6³6³8+9+;£;£=>@“@“B C EƒEƒGûGûIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçn]o]qÓqÓsItIv¿v¿x5y5{«{«}!~!€—€—‚ ƒ …ƒ…ƒ‡ù‡ù‰oŠoŒåŒåŽ[�[‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIKöIÐKöIéDöILéDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöIµ®öI¤µ®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIé×öI€é×öI‡÷Iü‡÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷I­¸÷I,­¸÷IkøI˜køI£øI$£øIëxùIèëxùI%eúIì%eúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnNoNqÀqÀs2t2v¤v¤xy{ˆ{ˆ}ú}úl€l‚Þ‚Þ„P…P‡‡‰4Š4Œ¦Œ¦Ž�‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÙöIÐÙöIwDöILwDöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI«möI(«möII®öI¤I®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI}×öI€}×öI÷Iü÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IA»÷I,A»÷I=øI˜=øI£øI$£øIßxùIèßxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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"×"×$M%M'Ã'Ã)9*9,¯,¯.%/%1›1›346‡6‡8ý8ý:s;s=é=é?_@_BÕBÕDKEKGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIKöIÐKöIéDöILéDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI»®öI¤»®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIï×öI€ï×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I;»÷I,;»÷IûøI˜ûøI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIåöIÐåöIƒDöILƒDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI±möI(±möIO®öI¤O®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIƒ×öI€ƒ×öI!÷Iü!÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I³¸÷I,³¸÷I[øI˜[øI£øI$£øIëxùIèëxùI%eúIì%eúIOûIÌOûIÙûIÙûIÇëûI€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n0o0q¥q¥stv�v�xy{y{y}î}îc€c‚؂؄M…M‡‡‰7Š7Œ¬Œ¬Ž!�!‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI?öIÐ?öIÝDöILÝDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöInöI(nöI¯®öI¤¯®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIã×öI€ã×öI�÷Iü�÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷I³¸÷I,³¸÷IøI˜øI£øI$£øIñxùIèñxùI+eúIì+eúIOûIÌOûIÙûIÙûIÇëûI€} 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�ïõIßöIÐßöI}DöIL}DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI±möI(±möIO®öI¤O®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIƒ×öI€ƒ×öI!÷Iü!÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷IM»÷I,M»÷IýøI˜ýøI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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�ïõIEöIÐEöIãDöILãDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöInöI(nöIµ®öI¤µ®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIé×öI€é×öI‡÷Iü‡÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷IM»÷I,M»÷I»øI˜»øI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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‡ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷I³¸÷I,³¸÷I£øI˜£øI£øI$£øIëxùIèëxùI%eúIì%eúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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‡ïõI9öIÐ9öI×DöIL×DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI nöI( nöI©®öI¤©®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIÝ×öI€Ý×öI{÷Iü{÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷I³¸÷I,³¸÷IÓøI˜ÓøI£øI$£øIßxùIèßxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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‡ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IY»÷I,Y»÷IEøI˜EøI£øI$£øIÙxùIèÙxùIeúIìeúI¹ûI̹ûI‰ØûI‰ØûI1ëûI€} 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‡ïõI?öIÐ?öIÝDöILÝDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI¯®öI¤¯®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIã×öI€ã×öI�÷Iü�÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IY»÷I,Y»÷IuøI˜uøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIwûIÌwûIGØûIGØûIïêûI€} 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·l‡ïõIëöIÐëöI‰DöIL‰DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI½möI(½möI[®öI¤[®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI�×öI€�×öI-÷Iü-÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷I¹¸÷I,¹¸÷IWøI˜WøI+£øI$+£øIyùIèyùI=eúIì=eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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·l‡ïõI3öIÐ3öIÑDöILÑDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI£®öI¤£®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI××öI€××öIu÷Iüu÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I¹¸÷I,¹¸÷I‡øI˜‡øI£øI$£øIñxùIèñxùI+eúIì+eúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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·l‡ïõI9öIÐ9öI×DöIL×DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI£®öI¤£®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI××öI€××öIu÷Iüu÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷Ik»÷I,k»÷I5øI˜5øI £øI$ £øIåxùIèåxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¤b¤degˆgˆiúiúklllnÞnÞpPqPsÂsÂu4v4x¦x¦z{}Š}Šüü�n‚n„à„à†T‡T‰ljNj:Œ:Ž­Ž­� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L“ïõI öI� öI§DöI§DöIýHöI$ýHöIÕmöI�ÕmöIs®öIs®öIϲöI$ϲöI§×öI�§×öIE÷IE÷I¡÷I$¡÷IÏ`÷I�Ï`÷I/…÷Iü/…÷Iy¶÷I y¶÷IEøI| EøIC£øIÜ C£øIyùI  yùIUeúI¤ UeúI©ûI„ ©ûIyÙûIÈyÙûI!ìûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI-öI�-öIËDöI ËDöI£KöI£KöIÿmöI€ÿmöI�®öIü�®öIuµöIuµöIÑ×öIpÑ×öIo÷Iìo÷IG÷IôG÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI¡DöI¡DöIýHöI$ýHöIÕmöI�ÕmöIs®öIs®öIϲöI$ϲöI§×öI�§×öIE÷IE÷I¡÷I$¡÷IÕ`÷I�Õ`÷I5…÷Iü5…÷Iy¶÷I y¶÷IEøI| EøI+£øIÜ +£øIyùI  yùI=eúI¤ =eúI1ûI„ 1ûIÙûIÈÙûI©ëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI3öI�3öIÑDöI ÑDöI£KöI£KöIÿmöI€ÿmöI�®öIü�®öIuµöIuµöIÑ×öIpÑ×öIo÷Iìo÷IG÷IôG÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIýöI�ýöI›DöI ›DöIýHöIýHöIÏmöI€ÏmöIm®öIüm®öIϲöIϲöI¡×öIp¡×öI?÷Iì?÷I¡÷Iô¡÷IÃ`÷I`Ã`÷I#…÷IÌ#…÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúIgûIT gûI7ÙûI˜7ÙûIßëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI3öI�3öIÑDöI ÑDöI‘KöI‘KöInöI€nöI£®öIü£®öIcµöIcµöI××öIp××öIu÷Iìu÷I5÷Iô5÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIýöI�ýöI›DöI ›DöIýHöIýHöIÉmöI€ÉmöIg®öIüg®öIϲöIϲöI›×öIp›×öI9÷Iì9÷I¡÷Iô¡÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI9öI�9öI×DöI ×DöI‘KöI‘KöInöI€nöI£®öIü£®öIcµöIcµöI××öIp××öIu÷Iìu÷I5÷Iô5÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI÷öI�÷öI•DöI •DöIIöIIöIÉmöI€ÉmöIg®öIüg®öIÕ²öIÕ²öI›×öIp›×öI9÷Iì9÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI9öI�9öI×DöI ×DöI—KöI—KöI nöI€ nöI©®öIü©®öIiµöIiµöIÝ×öIpÝ×öI{÷Iì{÷I;÷Iô;÷Iá`÷I`á`÷IA…÷IÌA…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI÷öI�÷öI•DöI •DöIIöIIöIÃmöI€ÃmöIa®öIüa®öIÕ²öIÕ²öI•×öIp•×öI3÷Iì3÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI9öI�9öI×DöI ×DöI—KöI—KöI nöI€ nöI©®öIü©®öIiµöIiµöIÝ×öIpÝ×öI{÷Iì{÷I;÷Iô;÷Ií`÷I`í`÷IM…÷IÌM…÷Is¶÷Iìs¶÷I?øIL ?øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI}ûIT }ûIMØûI˜MØûIõêûIœ€} 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‡ïõIñöI�ñöI�DöI �DöIIöIIöIÃmöI€ÃmöIa®öIüa®öIÕ²öIÕ²öI•×öIp•×öI3÷Iì3÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI+£øI¬ +£øIyùIp yùI=eúIt =eúI…ûIT …ûIUÙûI˜UÙûIýëûIœ€} 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‡ïõI?öI�?öIÝDöI ÝDöI…KöI…KöInöI€nöI¯®öIü¯®öIWµöIWµöIã×öIpã×öI�÷Iì�÷I)÷Iô)÷Iç`÷I`ç`÷IG…÷IÌG…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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‡ïõIëöI�ëöI‰DöI ‰DöI IöI IöI½möI€½möI[®öIü[®öIÛ²öIÛ²öI�×öIp�×öI-÷Iì-÷I­÷Iô­÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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‡ïõI?öI�?öIÝDöI ÝDöI…KöI…KöInöI€nöI¯®öIü¯®öIWµöIWµöIã×öIpã×öI�÷Iì�÷I)÷Iô)÷Ió`÷I`ó`÷IS…÷IÌS…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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�ïõIëöI�ëöI‰DöI ‰DöIIöIIöI½möI€½möI[®öIü[®öIÕ²öIÕ²öI�×öIp�×öI-÷Iì-÷I§÷Iô§÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI‘ûIT ‘ûIaÙûI˜aÙûI ìûIœ€} 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�ïõIåöI�åöIƒDöI ƒDöIýHöIýHöI·möI€·möIU®öIüU®öIϲöIϲöI‰×öIp‰×öI'÷Iì'÷I¡÷Iô¡÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIåöI�åöIƒDöI ƒDöIýHöIýHöI±möI€±möIO®öIüO®öIϲöIϲöIƒ×öIpƒ×öI!÷Iì!÷I¡÷Iô¡÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIKöI�KöIéDöI éDöIyKöIyKöInöI€nöIµ®öIüµ®öIKµöIKµöIé×öIpé×öI‡÷Iì‡÷I÷Iô÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIßöI�ßöI}DöI }DöIýHöIýHöI±möI€±möIO®öIüO®öIϲöIϲöIƒ×öIpƒ×öI!÷Iì!÷I¡÷Iô¡÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIKöI�KöIéDöI éDöIyKöIyKöInöI€nöI»®öIü»®öIKµöIKµöIï×öIpï×öI�÷Iì�÷I÷Iô÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIßöI�ßöI}DöI }DöI÷HöI÷HöI«möI€«möII®öIüI®öIɲöIɲöI}×öIp}×öI÷Iì÷I›÷Iô›÷I“`÷I`“`÷Ió„÷IÌó„÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIßxùIp ßxùIeúIt eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIÙöI�ÙöIwDöI wDöI÷HöI÷HöI«möI€«möII®öIüI®öIɲöIɲöI}×öIp}×öI÷Iì÷I›÷Iô›÷I±`÷I`±`÷I…÷IÌ…÷Ia¶÷Iìa¶÷I-øIL -øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúI­ûIT ­ûI}ØûI˜}ØûI%ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIQöI�QöIïDöI ïDöIKöIKöInöI€nöI»®öIü»®öIQµöIQµöIï×öIpï×öI�÷Iì�÷I#÷Iô#÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Ia¶÷Iìa¶÷I-øIL -øIã¢øI¬ ã¢øI»xùIp »xùIõdúIt õdúIkûIT kûI;ØûI˜;ØûIãêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIÓöI�ÓöIqDöI qDöIñHöIñHöI¥möI€¥möIC®öIüC®öIòöIòöIw×öIpw×öI÷Iì÷I•÷Iô•÷I_÷I`_÷Igƒ÷IÌgƒ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øIGwùIp GwùI%GúIt %GúIyûIT yûIo×ûI˜o×ûIêûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIQöI�QöIïDöI ïDöImKöImKöI#nöI€#nöIÁ®öIüÁ®öI?µöI?µöIõ×öIpõ×öI“÷Iì“÷I÷Iô÷Ic`÷I`c`÷IÄ÷IÔÄ÷Ia¶÷Iôa¶÷I-øIT -øI £øI´  £øI5wùIx 5wùIGúI| GúIûI\ ûI×ûI ×ûI·éûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÓöI�ÓöIqDöI qDöIñHöIñHöIŸmöI€ŸmöI=®öIü=®öIòöIòöIq×öIpq×öI÷Iì÷I•÷Iô•÷I_÷I`_÷Igƒ÷IÌgƒ÷Ia¶÷Iìa¶÷I-øIL -øI£øI¬ £øI/wùIp /wùI GúIt  GúIûIT ûIýÖûI˜ýÖûI¥éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIWöI�WöIõDöI õDöImKöImKöI#nöI€#nöIÁ®öIüÁ®öI?µöI?µöIõ×öIpõ×öI“÷Iì“÷I÷Iô÷Io`÷I`o`÷IÏ„÷IÔÏ„÷Ia¶÷Iôa¶÷I-øIT -øI£øI´ £øI)wùIx )wùIGúI| GúIÑûI\ ÑûIÇÖûI ÇÖûIoéûI¤€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIiöIèiöIEöIdEöIµKöIlµKöI?UöIè?UöIuXöI´uXöI;nöI@;nöIÙ®öI¼Ù®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöI ØöI˜ ØöI«÷I«÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷I�À÷ID�À÷IIõ÷IÐIõ÷I øIô øIøI øI£øI¬£øIñxùIpñxùI+eúIt+eúI¥rúIT¥rúI…ûIà…ûIUÙûI$UÙûIýëûI(€} 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„oïõI—öI—öI5DöI|5DöIŸIöI„ŸIöIÕLöIðÕLöI_VöI|_VöI•YöIH•YöIimöIÔimöI®öIP®öIq³öIXq³öI§¶öIħ¶öI1ÀöIP1ÀöIgÃöIgÃöI;×öI¨;×öIÙ÷I$ Ù÷IC÷I, C÷Iy ÷I˜ y ÷I*÷I$ *÷I9-÷Ið 9-÷I{Z÷I| {Z÷I‡`÷IX‡`÷Iç„÷IÔç„÷I)²÷Iô)²÷IíÀ÷IÐíÀ÷I©õ÷I\©õ÷IéøI€éøIç øI¬ç øI£øI8£øIÙxùIüÙxùIeúIeúI�rúIà�rúIãûIlãûI³ØûI°³ØûI[ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIoöIèoöI EöId EöIµKöIlµKöI?UöIè?UöIuXöI´uXöI;nöI@;nöIÙ®öI¼Ù®öI‡µöIćµöI¿öI@¿öIGÂöI GÂöI ØöI˜ ØöI«÷I«÷IY÷I Y÷Iã(÷I˜ ã(÷I,÷Id ,÷I[Y÷Ið [Y÷Ic`÷IÌ c`÷IÄ÷IH Ä÷I²÷Ih²÷I�À÷ID�À÷IIõ÷IÐIõ÷I øIô øIøI øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúI ûIà ûIÝØûI$ÝØûI…ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIÓöIèÓöIqDöIdqDöI»KöIl»KöIEUöIèEUöI{XöI´{XöI¥möI@¥möIC®öI¼C®öI�µöIÄ�µöI¿öI@¿öIMÂöI MÂöIw×öI˜w×öI÷I÷I_÷I _÷Ié(÷I˜ é(÷I,÷Id ,÷IaY÷Ið aY÷IÛ`÷IÌ Û`÷I;…÷IH ;…÷I}²÷Ih}²÷I“À÷ID“À÷IOõ÷IÐOõ÷IøIôøI øI  øI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúICûIàCûIÙûI$ÙûI»ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIcöIècöIEöIdEöI»KöIl»KöIEUöIèEUöI{XöI´{XöI5nöI@5nöIÓ®öI¼Ó®öI�µöIÄ�µöI¿öI@¿öIMÂöI MÂöIØöI˜ØöI¥÷I¥÷I_÷I _÷Ié(÷I˜ é(÷I,÷Id ,÷IaY÷Ið aY÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷I“À÷ID“À÷IOõ÷IÐOõ÷IøIôøI øI  øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúI=ûIà=ûI ÙûI$ ÙûIµëûI(€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n2o2q§q§stv‘v‘xy{M{M}Â}Â7€7‚¬‚¬„!…!‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxuïõIÓöIèÓöIqDöIdqDöI»KöIl»KöIEUöIèEUöI{XöI´{XöIŸmöI@ŸmöI=®öI¼=®öI�µöIÄ�µöI¿öI@¿öIMÂöI MÂöIq×öI˜q×öI÷I÷I_÷I _÷Ié(÷I˜ é(÷I,÷Id ,÷IaY÷Ið aY÷I±`÷IÌ ±`÷I…÷IH …÷IS²÷IhS²÷I“À÷ID“À÷IOõ÷IÐOõ÷IøIôøIøI øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúIËûIàËûI›ØûI$›ØûICëûI(€} 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$f%f'Ú'Ú)N*N,Â,Â.6/61ª1ª346’6’89;z;z=î=î?b@bBÖBÖDJEJG¾G¾I2J2L¤L¤NOQˆQˆSúSúU=V=X¯X¯Z![!]“]“_`b¦b¦degŠgŠiüiüknlnnènèpZqZsÌsÌu>v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxuïõIiöIèiöIEöIdEöI»KöIl»KöIEUöIèEUöI{XöI´{XöI5nöI@5nöIÓ®öI¼Ó®öI�µöIÄ�µöI¿öI@¿öIMÂöI MÂöIØöI˜ØöI¥÷I¥÷I_÷I _÷Ié(÷I˜ é(÷I,÷Id ,÷IaY÷Ið aY÷Io`÷IÌ o`÷IÏ„÷IH Ï„÷I²÷Ih²÷I“À÷ID“À÷IOõ÷IÐOõ÷IøIôøI øI  øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúI×ûIà×ûI§ØûI$§ØûIOëûI(€} 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·luïõI]öIÐ]öIûDöILûDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI)nöI()nöIÇ®öI¤Ç®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIû×öI€û×öI™÷Iü™÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I·½÷I,·½÷IãøI˜ãøI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIïØûIïØûI—ëûI€} 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·luïõIÙöIÐÙöIwDöILwDöIÁKöITÁKöIKUöIÐKUöI�XöIœ�XöI¥möI(¥möIC®öI¤C®öI“µöI¬“µöI¿öI(¿öISÂöIôSÂöIw×öI€w×öI÷Iü÷Ie÷I e÷Iï(÷I€ ï(÷I%,÷IL %,÷IgY÷IØ gY÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷IÀ÷I,À÷I¹øI˜¹øIÝ¢øI$Ý¢øIµxùIèµxùIïdúIìïdúI_ûIÌ_ûI/ØûI/ØûI×êûI€} 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$9%9'¬'¬)*,’,’./1x1x3ë3ë5^6^8Ñ8Ñ:D;D=·=·?(@(B™B™D E G{G{IìIìK]L]NÎNÎP?Q?S°S°U!V!X’X’Z[]t]t_å_åaVbVdÇdÇf8g8i©i©kln‹n‹püpürmsmuÞuÞwOxOzÀzÀ|1}1¢¢�‚„„„„†õ†õˆe‰e‹Õ‹Õ�EŽE�µ�µ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõI]öIÐ]öIûDöILûDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI/nöI(/nöIÍ®öI¤Í®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIØöI€ØöIŸ÷IüŸ÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I9À÷I,9À÷IgøI˜gøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI§ûI̧ûIwØûIwØûIëûI€} 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$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIßöIÐßöI}DöIL}DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI±möI(±möIO®öI¤O®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIƒ×öI€ƒ×öI!÷Iü!÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷I·½÷I,·½÷IøI˜øI£øI$£øIÙxùIèÙxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoqtqtsèsèu\v\xÐxÐzD{D}¸}¸,€,‚ ‚ „…‡ЇЉÿ‰ÿ‹tŒtŽéŽé�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIWöIÐWöIõDöILõDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI#nöI(#nöIÁ®öI¤Á®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIõ×öI€õ×öI“÷Iü“÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷IϽ÷I,Ͻ÷I©øI˜©øI£øI$£øIñxùIèñxùI+eúIì+eúIsûIÌsûICÙûICÙûIëëûI€} 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$d€� bJ7MWoperation.5exchange :z2breasonOperatorInputsvalue9invocations1�`à0° € 4t €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10gg à àYYÒÒ)K*K4Ä4Ä>=?=I´I´S)T)^œ^œhis‚s‚} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðggààYYÒÒ K K Ä Ä==¶¶//¨¨!!šš Œ Œ"#%~%~'÷'÷)p*p,é,é.b/b1Û1Û3T4T6Í6Í8F9F;¿;¿=8>8@±@±B*C*E£E£GHJ•J•L M OƒOƒQúQúSqTqVèVèX_Y_[Ö[Ö]M^M`Ä`Äb9c9e®e®g#h#j˜j˜l m o€o€qóqósftfvÙvÙxLyL{¿{¿}2~2€¥€¥‚ƒ…‹…‹‡þ‡þ‰pŠpŒâŒâŽT�T‘ƑƓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIÙöIÐÙöIwDöILwDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI«möI(«möII®öI¤I®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI}×öI€}×öI÷Iü÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷I?À÷I,?À÷I‹øI˜‹øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?M@MBÂBÂD7E7G¬G¬I!J!L”L”NOQzQzSíSíU`V`XÓXÓZF[F]¹]¹_,`,bžbždeg‚g‚iôiôkflfnØnØpJqJs¼s¼u.v.x x z{}„}„öö�h‚h„Ú„Ú†N‡N‰Á‰Á‹4Œ4ާާ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIWöIÐWöIõDöILõDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI)nöI()nöIÇ®öI¤Ç®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIû×öI€û×öI™÷Iü™÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I]À÷I,]À÷I9øI˜9øI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIÑØûIÑØûIyëûI€} 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$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIåöIÐåöIƒDöILƒDöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI·möI(·möIU®öI¤U®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI‰×öI€‰×öI'÷Iü'÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷IÕ½÷I,Õ½÷IÍøI˜ÍøI £øI$ £øIåxùIèåxùIeúIìeúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·l�ïõIKöIÐKöIéDöILéDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI»®öI¤»®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIï×öI€ï×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷Iç½÷I,ç½÷IuøI˜uøI£øI$£øI÷xùIè÷xùI1eúIì1eúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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·l�ïõIßöIÐßöI}DöIL}DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI±möI(±möIO®öI¤O®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIƒ×öI€ƒ×öI!÷Iü!÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷IiÀ÷I,iÀ÷IcøI˜cøI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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·l�ïõIQöIÐQöIïDöILïDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI#nöI(#nöIÁ®öI¤Á®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIõ×öI€õ×öI“÷Iü“÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷I�À÷I,�À÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} bJ7U'operation.0exchange :Ú!2!!b%reasonOperatorInput { input: 0 }value0invocations1z®T  b-wñoperation.0ipu_stack_gemm_f16_init_small_rows_interleaved_k576_c16_r72_r725:5z9299Z=úyz}*}z�2�z…*…b‰reasonOperatorKernelvalue4kernelSpecPlanned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 576, output_columns: 16 })outputElements1152input0Elements41472input1Elements9216invocations1€_ bJ7M7operation.0exchange :z2breasonOperatorInputsvalue4invocations1 U•�Ïä$ 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$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIëöIÐëöI‰DöIL‰DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷Ií½÷I,í½÷I™øI˜™øI£øI$£øIñxùIèñxùI+eúIì+eúIOûIÌOûIÙûIÙûIÇëûI€} 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�ïõIEöIÐEöIãDöILãDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöIµ®öI¤µ®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIé×öI€é×öI‡÷Iü‡÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷I¾÷I,¾÷IAøI˜AøI£øI$£øIñxùIèñxùI+eúIì+eúIOûIÌOûIÙûIÙûIÇëûI€} 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�ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I�À÷I,�À÷I5øI˜5øI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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�ïõIKöIÐKöIéDöILéDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI»®öI¤»®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIï×öI€ï×öI�÷Iü�÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷I«À÷I,«À÷IãøI˜ãøI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIñöIÐñöI�DöIL�DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷Iù½÷I,ù½÷IçøI˜çøI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI?öIÐ?öIÝDöILÝDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöInöI(nöI¯®öI¤¯®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIã×öI€ã×öI�÷Iü�÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I ¾÷I, ¾÷IeøI˜eøI£øI$£øIñxùIèñxùI+eúIì+eúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·l‡ïõIëöIÐëöI‰DöIL‰DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I™À÷I,™À÷I‰øI˜‰øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIwûIÌwûIGØûIGØûIïêûI€} 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$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=á=á?S@SBÅBÅD7E7G©G©IJL�L�NÿNÿPqQqSãSãUUVUXÇXÇZ9[9]«]«_`b�b�degsgsiåiåkWlWnÉnÉp;q;s­s­uvx‘x‘z{}u}uçç�Y‚Y„˄ˆ=‡=‰®‰®‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIñöIÐñöI�DöIL�DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöIÃmöI(ÃmöIa®öI¤a®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI•×öI€•×öI3÷Iü3÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I¾÷I,¾÷I­øI˜­øI£øI$£øIñxùIèñxùI+eúIì+eúIûIÌûIïØûIïØûI—ëûI€} 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·l�ïõI9öIÐ9öI×DöIL×DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI nöI( nöI©®öI¤©®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIÝ×öI€Ý×öI{÷Iü{÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷I#¾÷I,#¾÷I1øI˜1øI1£øI$1£øI yùIè yùICeúIìCeúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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·l�ïõIñöIÐñöI�DöIL�DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷IÃÀ÷I,ÃÀ÷I[øI˜[øI £øI$ £øIåxùIèåxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI?öIÐ?öIÝDöILÝDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI nöI( nöI©®öI¤©®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIÝ×öI€Ý×öI{÷Iü{÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IÕÀ÷I,ÕÀ÷IåøI˜åøI£øI$£øIñxùIèñxùI+eúIì+eúIûIÌûIéØûIéØûI‘ëûI€} 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$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡“‡“‰ŠŒuŒu޿޿�W‘W“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI­DöI­DöIIöI$IöIÛmöI�ÛmöIy®öIy®öIÕ²öI$Õ²öI­×öI�­×öIK÷IK÷I§÷I$§÷IÕ`÷I�Õ`÷I5…÷Iü5…÷I¶÷I ¶÷IKøI| KøI7£øIÜ 7£øIyùI  yùIIeúI¤ IeúIaûI„ aûI1ÙûIÈ1ÙûIÙëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI3öI�3öIÑDöI ÑDöI£KöI£KöInöI€nöI£®öIü£®öIuµöIuµöI××öIp××öIu÷Iìu÷IG÷IôG÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷I¶÷Iì¶÷IKøIL KøIC£øI¬ C£øIyùIp yùIUeúIt UeúI©ûIT ©ûIyÙûI˜yÙûI!ìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI öI� öI§DöI§DöIIöI$IöIÕmöI�ÕmöIs®öIs®öIÕ²öI$Õ²öI§×öI�§×öIE÷IE÷I§÷I$§÷Iá`÷I�á`÷IA…÷IüA…÷I¶÷I ¶÷IKøI| KøI%£øIÜ %£øIýxùI  ýxùI7eúI¤ 7eúIûI„ ûIרûIÈרûIëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI9öI�9öI×DöI ×DöI£KöI£KöInöI€nöI£®öIü£®öIuµöIuµöI××öIp××öIu÷Iìu÷IG÷IôG÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷I¶÷Iì¶÷IKøIL KøI+£øI¬ +£øIyùIp yùI=eúIt =eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI öI� öI§DöI §DöIýHöIýHöIÕmöI€ÕmöIs®öIüs®öIϲöIϲöI§×öIp§×öIE÷IìE÷I¡÷Iô¡÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúIgûIT gûI7ÙûI˜7ÙûIßëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI9öI�9öI×DöI ×DöI‘KöI‘KöI nöI€ nöI©®öIü©®öIcµöIcµöIÝ×öIpÝ×öI{÷Iì{÷I5÷Iô5÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúIgûIT gûI7ÙûI˜7ÙûIßëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¡DöI ¡DöIýHöIýHöIÏmöI€ÏmöIm®öIüm®öIϲöIϲöI¡×öIp¡×öI?÷Iì?÷I¡÷Iô¡÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI9öI�9öI×DöI ×DöI‘KöI‘KöI nöI€ nöI©®öIü©®öIcµöIcµöIÝ×öIpÝ×öI{÷Iì{÷I5÷Iô5÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIýöI�ýöI›DöI ›DöIIöIIöIÏmöI€ÏmöIm®öIüm®öIÕ²öIÕ²öI¡×öIp¡×öI?÷Iì?÷I§÷Iô§÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI?öI�?öIÝDöI ÝDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷Ió`÷I`ó`÷IS…÷IÌS…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI÷öI�÷öI•DöI •DöIIöIIöIÉmöI€ÉmöIg®öIüg®öIÕ²öIÕ²öI›×öIp›×öI9÷Iì9÷I§÷Iô§÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Is¶÷Iìs¶÷I?øIL ?øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI}ûIT }ûIMØûI˜MØûIõêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI?öI�?öIÝDöI ÝDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷Iù`÷I`ù`÷IY…÷IÌY…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI÷öI�÷öI•DöI •DöIIöIIöIÉmöI€ÉmöIg®öIüg®öIÕ²öIÕ²öI›×öIp›×öI9÷Iì9÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIEöI�EöIãDöI ãDöI‹KöI‹KöInöI€nöI¯®öIü¯®öI]µöI]µöIã×öIpã×öI�÷Iì�÷I/÷Iô/÷Ió`÷I`ó`÷IS…÷IÌS…÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI…ûIT …ûIUÙûI˜UÙûIýëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIEöI�EöIãDöI ãDöI‹KöI‹KöInöI€nöIµ®öIüµ®öI]µöI]µöIé×öIpé×öI‡÷Iì‡÷I/÷Iô/÷Ia÷I`a÷Ie…÷IÌe…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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�ïõIñöI�ñöI�DöI �DöIIöIIöIÃmöI€ÃmöIa®öIüa®öIÕ²öIÕ²öI•×öIp•×öI3÷Iì3÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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�ïõIKöI�KöIéDöI éDöI…KöI…KöInöI€nöIµ®öIüµ®öIWµöIWµöIé×öIpé×öI‡÷Iì‡÷I)÷Iô)÷Iç`÷I`ç`÷IG…÷IÌG…÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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�ïõIëöI�ëöI‰DöI ‰DöIIöIIöI½möI€½möI[®öIü[®öIÕ²öIÕ²öI�×öIp�×öI-÷Iì-÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIëöI�ëöI‰DöI ‰DöI÷HöI÷HöI½möI€½möI[®öIü[®öIɲöIɲöI�×öIp�×öI-÷Iì-÷I›÷Iô›÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIKöI�KöIéDöI éDöIKöIKöInöI€nöI»®öIü»®öIQµöIQµöIï×öIpï×öI�÷Iì�÷I#÷Iô#÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIåöI�åöIƒDöI ƒDöI÷HöI÷HöI·möI€·möIU®öIüU®öIɲöIɲöI‰×öIp‰×öI'÷Iì'÷I›÷Iô›÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIQöI�QöIïDöI ïDöIKöIKöInöI€nöI»®öIü»®öIQµöIQµöIï×öIpï×öI�÷Iì�÷I#÷Iô#÷Ia÷I`a÷Ie…÷IÌe…÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIåöI�åöIƒDöI ƒDöI÷HöI÷HöI±möI€±möIO®öIüO®öIɲöIɲöIƒ×öIpƒ×öI!÷Iì!÷I›÷Iô›÷I�`÷I`�`÷Ií„÷IÌí„÷Ig¶÷Iìg¶÷I3øIL 3øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIQöI�QöIïDöI ïDöIyKöIyKöI#nöI€#nöIÁ®öIüÁ®öIKµöIKµöIõ×öIpõ×öI“÷Iì“÷I÷Iô÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIßöI�ßöI}DöI }DöI÷HöI÷HöI«möI€«möII®öIüI®öIɲöIɲöI}×öIp}×öI÷Iì÷I›÷Iô›÷I«`÷I`«`÷I …÷IÌ …÷Ig¶÷Iìg¶÷I3øIL 3øIã¢øI¬ ã¢øI»xùIp »xùIõdúIt õdúIeûIT eûI5ØûI˜5ØûIÝêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIWöI�WöIõDöI õDöIyKöIyKöI#nöI€#nöIÁ®öIüÁ®öIKµöIKµöIõ×öIpõ×öI“÷Iì“÷I÷Iô÷I1_÷I`1_÷I‘ƒ÷IÌ‘ƒ÷Ig¶÷Iìg¶÷I3øIL 3øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI­ûIT ­ûI}ØûI˜}ØûI%ëûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIWöI�WöIõDöI õDöIsKöIsKöI)nöI€)nöIÇ®öIüÇ®öIEµöIEµöIû×öIpû×öI™÷Iì™÷I÷Iô÷I]`÷I`]`÷I½„÷IÔ½„÷Ia¶÷Iôa¶÷I-øIT -øI£øI´ £øIGwùIx GwùI%GúI| %GúIûI\ ûIu×ûI u×ûIêûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIÙöI�ÙöIwDöI wDöIëHöIëHöI¥möI€¥möIC®öIüC®öI½²öI½²öIw×öIpw×öI÷Iì÷I�÷Iô�÷I _÷I` _÷Imƒ÷IÌmƒ÷Ia¶÷Iìa¶÷I-øIL -øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúIÑûIT ÑûIÇÖûI˜ÇÖûIoéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIWöI�WöIõDöI õDöIsKöIsKöI)nöI€)nöIÇ®öIüÇ®öIEµöIEµöIû×öIpû×öI™÷Iì™÷I÷Iô÷I‡`÷I`‡`÷Iç„÷IÔç„÷Ia¶÷Iôa¶÷I-øIT -øI £øI´  £øI5wùIx 5wùIGúI| GúIûI\ ûIýÖûI ýÖûI¥éûI¤€} 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„oïõI�öI�öI;DöI|;DöIŸIöI„ŸIöIÕLöIðÕLöI_VöI|_VöI•YöIH•YöIomöIÔomöI ®öIP ®öIq³öIXq³öI§¶öIħ¶öI1ÀöIP1ÀöIgÃöIgÃöIA×öI¨A×öIß÷I$ ß÷IC÷I, C÷Iy ÷I˜ y ÷I*÷I$ *÷I9-÷Ið 9-÷I{Z÷I| {Z÷I·`÷IX·`÷I…÷IÔ…÷IY²÷IôY²÷I™À÷IЙÀ÷IUõ÷I\Uõ÷I3øI€3øI1øI¬1øI£øI8£øI÷xùIü÷xùI1eúI1eúI«rúIà«rúIyûIlyûIIÙûI°IÙûIñëûI´€} 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„oïõI�öI�öI;DöI|;DöIŸIöI„ŸIöIÕLöIðÕLöI_VöI|_VöI•YöIH•YöIomöIÔomöI ®öIP ®öIq³öIXq³öI§¶öIħ¶öI1ÀöIP1ÀöIgÃöIgÃöIA×öI¨A×öIß÷I$ ß÷IC÷I, C÷Iy ÷I˜ y ÷I*÷I$ *÷I9-÷Ið 9-÷I{Z÷I| {Z÷I�`÷IX�`÷Iá„÷IÔá„÷I#²÷Iô#²÷IÁ÷IÐÁ÷IÙõ÷I\Ùõ÷IÇøI€ÇøIÅ!øI¬Å!øI£øI8£øIßxùIüßxùIeúIeúI“rúIà“rúIûIlûIרûI°רûIëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIuöIèuöIEöIdEöI»KöIl»KöIEUöIèEUöI{XöI´{XöIAnöI@AnöIß®öI¼ß®öI�µöIÄ�µöI¿öI@¿öIMÂöI MÂöIØöI˜ØöI±÷I±÷I_÷I _÷Ié(÷I˜ é(÷I,÷Id ,÷IaY÷Ið aY÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷I™À÷ID™À÷IUõ÷IÐUõ÷I3øIô3øI1øI 1øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIÑûIàÑûI¡ØûI$¡ØûIIëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIÙöIèÙöIwDöIdwDöIÁKöIlÁKöIKUöIèKUöI�XöI´�XöI«möI@«möII®öI¼I®öI“µöIÄ“µöI¿öI@¿öISÂöI SÂöI}×öI˜}×öI÷I÷Ie÷I e÷Iï(÷I˜ ï(÷I%,÷Id %,÷IgY÷Ið gY÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷IŸÀ÷IDŸÀ÷I[õ÷IÐ[õ÷I9øIô9øI7øI 7øI£øI¬£øIñxùIpñxùI+eúIt+eúI¥rúIT¥rúIûIàûIOÙûI$OÙûI÷ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIÙöIèÙöIwDöIdwDöIÁKöIlÁKöIKUöIèKUöI�XöI´�XöI¥möI@¥möIC®öI¼C®öI“µöIÄ“µöI¿öI@¿öISÂöI SÂöIw×öI˜w×öI÷I÷Ie÷I e÷Iï(÷I˜ ï(÷I%,÷Id %,÷IgY÷Ið gY÷I«`÷IÌ «`÷I …÷IH  …÷IM²÷IhM²÷IŸÀ÷IDŸÀ÷I[õ÷IÐ[õ÷I9øIô9øI7øI 7øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIûIàûIרûI$רûIëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIiöIèiöIEöIdEöIÁKöIlÁKöIKUöIèKUöI�XöI´�XöI;nöI@;nöIÙ®öI¼Ù®öI“µöIÄ“µöI¿öI@¿öISÂöI SÂöI ØöI˜ ØöI«÷I«÷Ie÷I e÷Iï(÷I˜ ï(÷I%,÷Id %,÷IgY÷Ið gY÷I½`÷IÌ ½`÷I…÷IH …÷I_²÷Ih_²÷IóÀ÷IDóÀ÷I¯õ÷IЯõ÷IÝøIôÝøIÛ øI Û øIÝ¢øI¬Ý¢øIµxùIpµxùIïdúItïdúIirúITirúIeûIàeûI5ØûI$5ØûIÝêûI(€} 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·luïõIßöIÐßöI}DöIL}DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI«möI(«möII®öI¤I®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI}×öI€}×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I­¸÷I,­¸÷IEøI˜EøI£øI$£øIßxùIèßxùIeúIìeúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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$[%['Ð'Ð)E*E,º,º.///1¤1¤346Ž6Ž89;x;x=í=í?b@bB×B×DLELGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡�‡�‰ŠŒwŒwŽìŽì�a‘a“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIßöIÐßöI}DöIL}DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI«möI(«möII®öI¤I®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöI}×öI€}×öI÷Iü÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I)»÷I,)»÷IÃøI˜ÃøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†`‡`‰Ó‰Ó‹FŒF޹޹�,‘,“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·luïõIcöIÐcöIEöILEöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI5nöI(5nöIÓ®öI¤Ó®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIØöI€ØöI¥÷Iü¥÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I/»÷I,/»÷IYøI˜YøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIÅûIÌÅûI•ØûI•ØûI=ëûI€} 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$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†O‡O‰‰‹5Œ5ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI±möI(±möIO®öI¤O®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIƒ×öI€ƒ×öI!÷Iü!÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷I­¸÷I,­¸÷IÉøI˜ÉøI£øI$£øIßxùIèßxùIeúIìeúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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·l{ïõIWöIÐWöIõDöILõDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI)nöI()nöIÇ®öI¤Ç®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIû×öI€û×öI™÷Iü™÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Iÿ`÷I´ ÿ`÷I_…÷I0 _…÷I¡²÷IP¡²÷I­¸÷I,­¸÷I‰øI˜‰øI £øI$ £øIåxùIèåxùIeúIìeúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·l{ïõIßöIÐßöI}DöIL}DöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI±möI(±möIO®öI¤O®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIƒ×öI€ƒ×öI!÷Iü!÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷I/»÷I,/»÷IMøI˜MøIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúI­ûIÌ­ûI}ØûI}ØûI%ëûI€} 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$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŠbŠdüdüfngniàiàkRlRnÄnÄp6q6s¨s¨uvxŒxŒzþzþ|p}pââ�T‚T„ƄƆ8‡8‰©‰©‹ŒŽ‹Ž‹�ü�ü’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI]öIÐ]öIûDöILûDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI/nöI(/nöIÍ®öI¤Í®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöIŸ÷IüŸ÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I;»÷I,;»÷IøI˜øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIåöIÐåöIƒDöILƒDöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI·möI(·möIU®öI¤U®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI‰×öI€‰×öI'÷Iü'÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I­¸÷I,­¸÷I}øI˜}øI£øI$£øI÷xùIè÷xùI1eúIì1eúIyûIÌyûIIÙûIIÙûIñëûI€} 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·l�ïõIQöIÐQöIïDöILïDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI#nöI(#nöIÁ®öI¤Á®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIõ×öI€õ×öI“÷Iü“÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷I­¸÷I,­¸÷I=øI˜=øI£øI$£øIñxùIèñxùI+eúIì+eúIOûIÌOûIÙûIÙûIÇëûI€} 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·l�ïõIåöIÐåöIƒDöILƒDöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI·möI(·möIU®öI¤U®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI‰×öI€‰×öI'÷Iü'÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷IA»÷I,A»÷I øI˜ øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIרûIרûIëûI€} 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$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIWöIÐWöIõDöILõDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI)nöI()nöIÇ®öI¤Ç®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIû×öI€û×öI™÷Iü™÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷IM»÷I,M»÷IÙøI˜ÙøI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIëöIÐëöI‰DöIL‰DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷I³¸÷I,³¸÷I1øI˜1øI%£øI$%£øIýxùIèýxùI7eúIì7eúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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�ïõIKöIÐKöIéDöILéDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI»®öI¤»®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIï×öI€ï×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I³¸÷I,³¸÷IñøI˜ñøI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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�ïõIëöIÐëöI‰DöIL‰DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷IM»÷I,M»÷IÇøI˜ÇøI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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�ïõIQöIÐQöIïDöILïDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöI»®öI¤»®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIï×öI€ï×öI�÷Iü�÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IY»÷I,Y»÷I™øI˜™øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIqûIÌqûIAØûIAØûIéêûI€} 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·l‡ïõIñöIÐñöI�DöIL�DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöIÃmöI(ÃmöIa®öI¤a®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI•×öI€•×öI3÷Iü3÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I¹¸÷I,¹¸÷IåøI˜åøI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l‡ïõIEöIÐEöIãDöILãDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöIµ®öI¤µ®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIé×öI€é×öI‡÷Iü‡÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I¹¸÷I,¹¸÷IøI˜øI£øI$£øIñxùIèñxùI+eúIì+eúIûIÌûIïØûIïØûI—ëûI€} 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·l‡ïõIñöIÐñöI�DöIL�DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöIÃmöI(ÃmöIa®öI¤a®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI•×öI€•×öI3÷Iü3÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷I_»÷I,_»÷I‡øI˜‡øI£øI$£øIßxùIèßxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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·l‡ïõIKöIÐKöIéDöILéDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöInöI(nöIµ®öI¤µ®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIé×öI€é×öI‡÷Iü‡÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷I_»÷I,_»÷I½øI˜½øI£øI$£øIßxùIèßxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l�ïõI?öIÐ?öIÝDöILÝDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI nöI( nöI©®öI¤©®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIÝ×öI€Ý×öI{÷Iü{÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷I¹¸÷I,¹¸÷IÉøI˜ÉøI%£øI$%£øIýxùIèýxùI7eúIì7eúIOûIÌOûIÙûIÙûIÇëûI€} 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·l�ïõI÷öIÐ÷öI•DöIL•DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöIÃmöI(ÃmöIa®öI¤a®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöI•×öI€•×öI3÷Iü3÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷Ik»÷I,k»÷IGøI˜GøI£øI$£øIßxùIèßxùIeúIìeúI¿ûIÌ¿ûI�ØûI�ØûI7ëûI€} 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·l�ïõI?öIÐ?öIÝDöILÝDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöI¯®öI¤¯®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIã×öI€ã×öI�÷Iü�÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷Iq»÷I,q»÷I}øI˜}øI£øI$£øIëxùIèëxùI%eúIì%eúIõûIÌõûIÅØûIÅØûImëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI­DöI­DöIIöI$IöIÛmöI�ÛmöIy®öIy®öIÕ²öI$Õ²öI­×öI�­×öIK÷IK÷I§÷I$§÷Iç`÷I�ç`÷IG…÷IüG…÷I¶÷I ¶÷IKøI| KøI+£øIÜ +£øIyùI  yùI=eúI¤ =eúI+ûI„ +ûIûØûIÈûØûI£ëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI9öI�9öI×DöI ×DöI£KöI£KöI nöI€ nöI©®öIü©®öIuµöIuµöIÝ×öIpÝ×öI{÷Iì{÷IG÷IôG÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI öI� öI§DöI §DöIýHöIýHöIÛmöI€ÛmöIy®öIüy®öIϲöIϲöI­×öIp­×öIK÷IìK÷I¡÷Iô¡÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Iy¶÷Iìy¶÷IEøIL EøI=£øI¬ =£øIyùIp yùIOeúIt OeúI£ûIT £ûIsÙûI˜sÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI?öI�?öIÝDöI ÝDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI?öI�?öIÝDöI ÝDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIÙxùIp ÙxùIeúIt eúIƒûIT ƒûISØûI˜SØûIûêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¡DöI ¡DöIIöIIöIÕmöI€ÕmöIs®öIüs®öIÕ²öIÕ²öI§×öIp§×öIE÷IìE÷I§÷Iô§÷IÉ`÷I`É`÷I)…÷IÌ)…÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIýöI�ýöI›DöI ›DöIIöIIöIÏmöI€ÏmöIm®öIüm®öIÕ²öIÕ²öI¡×öIp¡×öI?÷Iì?÷I§÷Iô§÷Iá`÷I`á`÷IA…÷IÌA…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIEöI�EöIãDöI ãDöI‹KöI‹KöInöI€nöIµ®öIüµ®öI]µöI]µöIé×öIpé×öI‡÷Iì‡÷I/÷Iô/÷I a÷I` a÷Ik…÷IÌk…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIýöI�ýöI›DöI ›DöIýHöIýHöIÉmöI€ÉmöIg®öIüg®öIϲöIϲöI›×öIp›×öI9÷Iì9÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIKöI�KöIéDöI éDöI‹KöI‹KöInöI€nöIµ®öIüµ®öI]µöI]µöIé×öIpé×öI‡÷Iì‡÷I/÷Iô/÷Ió`÷I`ó`÷IS…÷IÌS…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI÷öI�÷öI•DöI •DöIIöIIöIÉmöI€ÉmöIg®öIüg®öIÕ²öIÕ²öI›×öIp›×öI9÷Iì9÷I§÷Iô§÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúI¿ûIT ¿ûI�ØûI˜�ØûI7ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIKöI�KöIéDöI éDöI‹KöI‹KöInöI€nöI»®öIü»®öI]µöI]µöIï×öIpï×öI�÷Iì�÷I/÷Iô/÷I a÷I` a÷Ik…÷IÌk…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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�ïõI÷öI�÷öI•DöI •DöIIöIIöIÃmöI€ÃmöIa®öIüa®öIÕ²öIÕ²öI•×öIp•×öI3÷Iì3÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIOÙûI˜OÙûI÷ëûIœ€} 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�ïõIKöI�KöIéDöI éDöIKöIKöInöI€nöI»®öIü»®öIQµöIQµöIï×öIpï×öI�÷Iì�÷I#÷Iô#÷Iç`÷I`ç`÷IG…÷IÌG…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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�ïõIñöI�ñöI�DöI �DöIIöIIöIÃmöI€ÃmöIa®öIüa®öIÕ²öIÕ²öI•×öIp•×öI3÷Iì3÷I§÷Iô§÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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�ïõIQöI�QöIïDöI ïDöIKöIKöInöI€nöI»®öIü»®öIQµöIQµöIï×öIpï×öI�÷Iì�÷I#÷Iô#÷I a÷I` a÷Ik…÷IÌk…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIëöI�ëöI‰DöI ‰DöI÷HöI÷HöI½möI€½möI[®öIü[®öIɲöIɲöI�×öIp�×öI-÷Iì-÷I›÷Iô›÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIQöI�QöIïDöI ïDöIKöIKöI#nöI€#nöIÁ®öIüÁ®öIQµöIQµöIõ×öIpõ×öI“÷Iì“÷I#÷Iô#÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIëöI�ëöI‰DöI ‰DöI÷HöI÷HöI½möI€½möI[®öIü[®öIɲöIɲöI�×öIp�×öI-÷Iì-÷I›÷Iô›÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIWöI�WöIõDöI õDöIKöIKöI#nöI€#nöIÁ®öIüÁ®öIQµöIQµöIõ×öIpõ×öI“÷Iì“÷I#÷Iô#÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI颸I¬ 颸IÁxùIp ÁxùIûdúIt ûdúIkûIT kûI;ØûI˜;ØûIãêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIåöI�åöIƒDöI ƒDöI÷HöI÷HöI·möI€·möIU®öIüU®öIɲöIɲöI‰×öIp‰×öI'÷Iì'÷I›÷Iô›÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIWöI�WöIõDöI õDöIyKöIyKöI)nöI€)nöIÇ®öIüÇ®öIKµöIKµöIû×öIpû×öI™÷Iì™÷I÷Iô÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIåöI�åöIƒDöI ƒDöI÷HöI÷HöI±möI€±möIO®öIüO®öIɲöIɲöIƒ×öIpƒ×öI!÷Iì!÷I›÷Iô›÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIWöI�WöIõDöI õDöIsKöIsKöI)nöI€)nöIÇ®öIüÇ®öIEµöIEµöIû×öIpû×öI™÷Iì™÷I÷Iô÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LuïõIßöI�ßöI}DöI }DöIëHöIëHöI±möI€±möIO®öIüO®öI½²öI½²öIƒ×öIpƒ×öI!÷Iì!÷I�÷Iô�÷I _÷I` _÷Imƒ÷IÌmƒ÷Ia¶÷Iìa¶÷I-øIL -øI £øI¬  £øI5wùIp 5wùIGúIt GúI+ûIT +ûI!×ûI˜!×ûIÉéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõI]öI�]öIûDöI ûDöIsKöIsKöI/nöI€/nöIÍ®öIüÍ®öIEµöIEµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷Ic`÷I`c`÷IÄ÷IÔÄ÷Ia¶÷Iôa¶÷I-øIT -øI£øI´ £øIAwùIx AwùIGúI| GúICûI\ CûI9×ûI 9×ûIáéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõIßöI�ßöI}DöI }DöIëHöIëHöI«möI€«möII®öIüI®öI½²öI½²öI}×öIp}×öI÷Iì÷I�÷Iô�÷I _÷I` _÷Imƒ÷IÌmƒ÷Ia¶÷Iìa¶÷I-øIL -øIû¢øI¬ û¢øI#wùIp #wùIGúIt GúI³ûIT ³ûI©ÖûI˜©ÖûIQéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17LuïõI]öI�]öIûDöI ûDöIsKöIsKöI/nöI€/nöIÍ®öIüÍ®öIEµöIEµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷I™`÷I`™`÷Iù„÷IÔù„÷Ia¶÷Iôa¶÷I-øIT -øI£øI´ £øI/wùIx /wùI GúI|  GúIéûI\ éûIßÖûI ßÖûI‡éûI¤€} 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„oïõI£öI£öIADöI|ADöI¥IöI„¥IöIÛLöIðÛLöIeVöI|eVöI›YöIH›YöIomöIÔomöI ®öIP ®öIw³öIXw³öI­¶öIÄ­¶öI7ÀöIP7ÀöImÃöImÃöIA×öI¨A×öIß÷I$ ß÷II÷I, I÷I ÷I˜  ÷I *÷I$  *÷I?-÷Ið ?-÷I�Z÷I| �Z÷I±`÷IX±`÷I…÷IÔ…÷IS²÷IôS²÷I«À÷IЫÀ÷Igõ÷I\gõ÷I™øI€™øI—øI¬—øI £øI8 £øIåxùIüåxùIeúIeúI™rúIà™rúIûIlûIéØûI°éØûI‘ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõIuöIèuöIEöIdEöIÁKöIlÁKöIKUöIèKUöI�XöI´�XöIAnöI@AnöIß®öI¼ß®öI“µöIÄ“µöI¿öI@¿öISÂöI SÂöIØöI˜ØöI±÷I±÷Ie÷I e÷Iï(÷I˜ ï(÷I%,÷Id %,÷IgY÷Ið gY÷I«`÷IÌ «`÷I …÷IH  …÷IM²÷IhM²÷IŸÀ÷IDŸÀ÷I[õ÷IÐ[õ÷I]øIô]øI[øI [øI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúIIûIàIûIÙûI$ÙûIÁëûI(€} 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„oïõI�öI�öI;DöI|;DöI¥IöI„¥IöIÛLöIðÛLöIeVöI|eVöI›YöIH›YöIomöIÔomöI ®öIP ®öIw³öIXw³öI­¶öIÄ­¶öI7ÀöIP7ÀöImÃöImÃöIA×öI¨A×öIß÷I$ ß÷II÷I, I÷I ÷I˜  ÷I *÷I$  *÷I?-÷Ið ?-÷I�Z÷I| �Z÷I“`÷IX“`÷Ió„÷IÔó„÷I5²÷Iô5²÷IGÁ÷IÐGÁ÷Iö÷I\ö÷I¥øI€¥øI£"øI¬£"øIû¢øI8û¢øIÓxùIüÓxùI eúI eúI‡rúIà‡rúIÑûIlÑûI¡ØûI°¡ØûIIëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIßöIèßöI}DöId}DöIÇKöIlÇKöIQUöIèQUöI‡XöI´‡XöI«möI@«möII®öI¼I®öI™µöIÄ™µöI#¿öI@#¿öIYÂöI YÂöI}×öI˜}×öI÷I÷Ik÷I k÷Iõ(÷I˜ õ(÷I+,÷Id +,÷ImY÷Ið mY÷IÏ`÷IÌ Ï`÷I/…÷IH /…÷Iq²÷Ihq²÷I¥À÷ID¥À÷Iaõ÷IÐaõ÷IcøIôcøIaøI aøI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúI7ûIà7ûIÙûI$ÙûI¯ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIoöIèoöI EöId EöIÇKöIlÇKöIQUöIèQUöI‡XöI´‡XöI;nöI@;nöIÙ®öI¼Ù®öI™µöIÄ™µöI#¿öI@#¿öIYÂöI YÂöI ØöI˜ ØöI«÷I«÷Ik÷I k÷Iõ(÷I˜ õ(÷I+,÷Id +,÷ImY÷Ið mY÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷I#Á÷ID#Á÷Ißõ÷IÐßõ÷IÍøIôÍøIË!øI Ë!øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúIûIàûIïØûI$ïØûI—ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIoöIèoöI EöId EöIÇKöIlÇKöIQUöIèQUöI‡XöI´‡XöIAnöI@AnöIß®öI¼ß®öI™µöIÄ™µöI#¿öI@#¿öIYÂöI YÂöIØöI˜ØöI±÷I±÷Ik÷I k÷Iõ(÷I˜ õ(÷I+,÷Id +,÷ImY÷Ið mY÷I¥`÷IÌ ¥`÷I…÷IH …÷IG²÷IhG²÷IÃÀ÷IDÃÀ÷Iõ÷IÐõ÷IÿøIôÿøIýøI ýøIï¢øI¬ï¢øIÇxùIpÇxùIeúIteúI{rúIT{rúI§ûIà§ûIwØûI$wØûIëûI(€} 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$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŽbŽdegrgriäiäkVlVn•n•pqsysyuëuëw]x]z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIåöIÐåöIƒDöILƒDöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI±möI(±möIO®öI¤O®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIƒ×öI€ƒ×öI!÷Iü!÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I¥½÷I,¥½÷IAøI˜AøI£øI$£øIßxùIèßxùIeúIìeúICûIÌCûIÙûIÙûI»ëûI€} 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·l{ïõIcöIÐcöIEöILEöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI5nöI(5nöIÓ®öI¤Ó®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöIØöI€ØöI¥÷Iü¥÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷Iý÷I,ý÷I=øI˜=øI£øI$£øIñxùIèñxùI+eúIì+eúIsûIÌsûICÙûICÙûIëëûI€} 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·l{ïõIßöIÐßöI}DöIL}DöIÇKöITÇKöIQUöIÐQUöI‡XöIœ‡XöI±möI(±möIO®öI¤O®öI™µöI¬™µöI#¿öI(#¿öIYÂöIôYÂöIƒ×öI€ƒ×öI!÷Iü!÷Ik÷I k÷Iõ(÷I€ õ(÷I+,÷IL +,÷ImY÷IØ mY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷I!À÷I,!À÷I¿øI˜¿øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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$g%g'Û'Û)O*O,Ã,Ã.7/71«1«346“6“89;{;{=ï=ï?c@cB×B×DKEKG¿G¿I3J3L¥L¥NOQ‰Q‰SûSûUmVmXßXßZQ[Q]Ã]Ã_5`5b§b§deg‹g‹iýiýkolonánápSqSsÅsÅu7v7x©x©z{}�}�ÿÿ�q‚q„ã„ã†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIiöIÐiöIEöILEöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI;nöI(;nöIÙ®öI¤Ù®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI ØöI€ ØöI«÷Iü«÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷IKÀ÷I,KÀ÷IÇøI˜ÇøI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIÑØûIÑØûIyëûI€} 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$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷I·½÷I,·½÷IÑøI˜ÑøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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$D%D'¹'¹).*.,£,£./1�1�346w6w8ì8ì:a;a=Ö=Ö?I@IB¼B¼D/E/G¢G¢IJLˆLˆNûNûPnQnSáSáUTVTXÇXÇZ:[:]­]­_ ` b“b“degygyiìiìk_l_nÒnÒpEqEs¸s¸u+v+xžxžz{}„}„÷÷�j‚j„݄݆P‡P‰‰‹4Œ4ަަ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI]öIÐ]öIûDöILûDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI/nöI(/nöIÍ®öI¤Í®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöIŸ÷IüŸ÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Iÿ`÷I´ ÿ`÷I_…÷I0 _…÷I¡²÷IP¡²÷IÛ½÷I,Û½÷I÷øI˜÷øI£øI$£øI÷xùIè÷xùI1eúIì1eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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$n%n'æ'æ)^*^,Ö,Ö.N/N1Æ1Æ3>4>6¶6¶8.9.;¦;¦=>@–@–BCE†E†GþGþIvJvLìLìNbObQØQØSNTNVÄVÄX:Y:[°[°]&^&`œ`œbceˆeˆgþgþitjtlêlên`o`qÖqÖsLtLvÂvÂx8y8{®{®}$~$€š€š‚ƒ…†…†‡ü‡ü‰rŠrŒèŒèŽ^�^‘Ô‘Ô“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I9À÷I,9À÷IUøI˜UøIã¢øI$ã¢øI»xùIè»xùIõdúIìõdúIeûIÌeûI5ØûI5ØûIÝêûI€} 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$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æaWbWdÈdÈf9g9iªiªklnŒnŒpýpýrnsnußußwPxPzÁzÁ|2}2££�‚„…„…†ö†öˆf‰f‹Ö‹Ö�FŽF�¶�¶’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIcöIÐcöIEöILEöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI5nöI(5nöIÓ®öI¤Ó®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöI¥÷Iü¥÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷IoÀ÷I,oÀ÷I‡øI˜‡øI£øI$£øIßxùIèßxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$N%N'Ä'Ä):*:,°,°.&/&1œ1œ346ˆ6ˆ8þ8þ:t;t=ê=ê?`@`BÖBÖDLELGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡Œ‡Œ‰ŠŒtŒtŽèŽè�\‘\“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIëöIÐëöI‰DöIL‰DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI½möI(½möI[®öI¤[®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI�×öI€�×öI-÷Iü-÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷IϽ÷I,Ͻ÷I�øI˜�øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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$Y%Y'Î'Î)C*C,¸,¸.-/-1¢1¢346Œ6Œ89;v;v=ë=ë?`@`BÕBÕDJEJG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‹‡‹‰ŠŒuŒuŽêŽê�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIWöIÐWöIõDöILõDöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI)nöI()nöIÇ®öI¤Ç®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIû×öI€û×öI™÷Iü™÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iÿ`÷I´ ÿ`÷I_…÷I0 _…÷I¡²÷IP¡²÷Ió½÷I,ó½÷I½øI˜½øI£øI$£øIñxùIèñxùI+eúIì+eúIOûIÌOûIÙûIÙûIÇëûI€} 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$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubêbêd_e_gÔgÔiIjIl¾l¾n3o3q¨q¨stv’v’xy{|{|}ñ}ñf€f‚Û‚Û„P…P‡Ňʼn:Š:Œ¯Œ¯Ž$�$‘™‘™“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIëöIÐëöI‰DöIL‰DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI½möI(½möI[®öI¤[®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI�×öI€�×öI-÷Iü-÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I]À÷I,]À÷I-øI˜-øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?N@NBÃBÃD8E8G­G­I"J"L•L•NOQ{Q{SîSîUaVaXÔXÔZG[G]º]º_-`-bŸbŸdegƒgƒiõiõkglgnÙnÙpKqKs½s½u/v/x¡x¡z{}…}…÷÷�i‚i„Û„Û†O‡O‰‰‹5Œ5ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI]öIÐ]öIûDöILûDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI)nöI()nöIÇ®öI¤Ç®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIû×öI€û×öI™÷Iü™÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷I“À÷I,“À÷I_øI˜_øI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$N%N'Ã'Ã)8*8,­,­."/"1—1—3 4 6�6�8ö8ö:k;k=à=à?U@UBÊBÊD?E?G´G´I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnænæpYqYsÌsÌu?v?x²x²z%{%}˜}˜ € ‚~‚~„ñ„ñ†d‡d‰׉׋JŒJ޽޽�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIñöIÐñöI�DöIL�DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöIÃmöI(ÃmöIa®öI¤a®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöI•×öI€•×öI3÷Iü3÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷Iç½÷I,ç½÷IcøI˜cøI£øI$£øIëxùIèëxùI%eúIì%eúICûIÌCûIÙûIÙûI»ëûI€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIWJWLÎLÎNEOEQ¼Q¼S3T3VªVªX!Y![˜[˜]^`†`†býbýdtetgëgëibjblÙlÙnKoKq½q½s/t/v¡v¡xy{…{…}÷}÷i€i‚Û‚Û„M…M‡¿‡¿‰1Š1Œ£Œ£Ž�‘‡‘‡“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIQöIÐQöIïDöILïDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI#nöI(#nöIÁ®öI¤Á®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIõ×öI€õ×öI“÷Iü“÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷I ¾÷I, ¾÷I‰øI˜‰øI£øI$£øIñxùIèñxùI+eúIì+eúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIWöIÐWöIõDöILõDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI#nöI(#nöIÁ®öI¤Á®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIõ×öI€õ×öI“÷Iü“÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I·À÷I,·À÷I1øI˜1øI£øI$£øIÙxùIèÙxùIeúIìeúI¹ûI̹ûI‰ØûI‰ØûI1ëûI€} 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$X%X'Ì'Ì)@*@,´,´.(/(1œ1œ346„6„8ø8ø:l;l=à=à?R@RBÄBÄD6E6G¨G¨IJLŒLŒNþNþPpQpSâSâUTVTXÆXÆZ8[8]ª]ª_`bŽbŽdegrgriäiäkVlVnÈnÈp:q:s¬s¬uvx�x�z{}t}tææ�X‚X„ʄʆ<‡<‰­‰­‹ŒŽ�Ž��‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI÷öIÐ÷öI•DöIL•DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöIÉmöI(ÉmöIg®öI¤g®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöI›×öI€›×öI9÷Iü9÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I¾÷I,¾÷I/øI˜/øI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l‡ïõIKöIÐKöIéDöILéDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöIµ®öI¤µ®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIé×öI€é×öI‡÷Iü‡÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I)¾÷I,)¾÷IUøI˜UøI+£øI$+£øIyùIèyùI=eúIì=eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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·l‡ïõIKöIÐKöIéDöILéDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöInöI(nöI»®öI¤»®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIï×öI€ï×öI�÷Iü�÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IÛÀ÷I,ÛÀ÷I øI˜ øI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIãØûIãØûI‹ëûI€} 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·l�ïõIýöIÐýöI›DöIL›DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÉmöI(ÉmöIg®öI¤g®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI›×öI€›×öI9÷Iü9÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷I¾÷I,¾÷IûøI˜ûøI £øI$ £øIåxùIèåxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l�ïõIEöIÐEöIãDöILãDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöInöI(nöI¯®öI¤¯®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIã×öI€ã×öI�÷Iü�÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷IA¾÷I,A¾÷IøI˜øI1£øI$1£øI yùIè yùICeúIìCeúI—ûIÌ—ûIgÙûIgÙûIìûI€} 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·l�ïõIEöIÐEöIãDöILãDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöInöI(nöIµ®öI¤µ®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIé×öI€é×öI‡÷Iü‡÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷IÿÀ÷I,ÿÀ÷IÛøI˜ÛøI£øI$£øIñxùIèñxùI+eúIì+eúIûIÌûIïØûIïØûI—ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI?öI�?öIÝDöI ÝDöI£KöI£KöInöI€nöI¯®öIü¯®öIuµöIuµöIã×öIpã×öI�÷Iì�÷IG÷IôG÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷I¶÷Iì¶÷IKøIL KøI7£øI¬ 7£øIyùIp yùIIeúIt IeúImûIT mûI=ÙûI˜=ÙûIåëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI­DöI­DöIIöI$IöIámöI�ámöI®öI®öIÕ²öI$Õ²öI³×öI�³×öIQ÷IQ÷I§÷I$§÷Ií`÷I�í`÷IM…÷IüM…÷I¶÷I ¶÷IKøI| KøI£øIÜ £øI÷xùI  ÷xùI1eúI¤ 1eúIõûI„ õûIÅØûIÈÅØûImëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI?öI�?öIÝDöI ÝDöI£KöI£KöInöI€nöI¯®öIü¯®öIuµöIuµöIã×öIpã×öI�÷Iì�÷IG÷IôG÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷I¶÷Iì¶÷IKøIL KøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI­DöI ­DöIýHöIýHöIámöI€ámöI®öIü®öIϲöIϲöI³×öIp³×öIQ÷IìQ÷I¡÷Iô¡÷Iá`÷I`á`÷IA…÷IÌA…÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIEöI�EöIãDöI ãDöI‘KöI‘KöInöI€nöI¯®öIü¯®öIcµöIcµöIã×öIpã×öI�÷Iì�÷I5÷Iô5÷I a÷I` a÷Ik…÷IÌk…÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI öI� öI§DöI §DöIýHöIýHöIÛmöI€ÛmöIy®öIüy®öIϲöIϲöI­×öIp­×öIK÷IìK÷I¡÷Iô¡÷Ií`÷I`í`÷IM…÷IÌM…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIEöI�EöIãDöI ãDöI‘KöI‘KöInöI€nöIµ®öIüµ®öIcµöIcµöIé×öIpé×öI‡÷Iì‡÷I5÷Iô5÷Ia÷I`a÷Iq…÷IÌq…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI öI� öI§DöI §DöIIöIIöIÕmöI€ÕmöIs®öIüs®öIÕ²öIÕ²öI§×öIp§×öIE÷IìE÷I§÷Iô§÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIKöI�KöIéDöI éDöI…KöI…KöInöI€nöIµ®öIüµ®öIWµöIWµöIé×öIpé×öI‡÷Iì‡÷I)÷Iô)÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¡DöI ¡DöIIöIIöIÕmöI€ÕmöIs®öIüs®öIÕ²öIÕ²öI§×öIp§×öIE÷IìE÷I§÷Iô§÷Iç`÷I`ç`÷IG…÷IÌG…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIKöI�KöIéDöI éDöI…KöI…KöInöI€nöI»®öIü»®öIWµöIWµöIï×öIpï×öI�÷Iì�÷I)÷Iô)÷Ia÷I`a÷Iq…÷IÌq…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¡DöI ¡DöIýHöIýHöIÏmöI€ÏmöIm®öIüm®öIϲöIϲöI¡×öIp¡×öI?÷Iì?÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIKöI�KöIéDöI éDöI‘KöI‘KöInöI€nöI»®öIü»®öIcµöIcµöIï×öIpï×öI�÷Iì�÷I5÷Iô5÷Ió`÷I`ó`÷IS…÷IÌS…÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúI—ûIT —ûIgÙûI˜gÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIýöI�ýöI›DöI ›DöIýHöIýHöIÏmöI€ÏmöIm®öIüm®öIϲöIϲöI¡×öIp¡×öI?÷Iì?÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Is¶÷Iìs¶÷I?øIL ?øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI}ûIT }ûIMØûI˜MØûIõêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIQöI�QöIïDöI ïDöI‘KöI‘KöInöI€nöI»®öIü»®öIcµöIcµöIï×öIpï×öI�÷Iì�÷I5÷Iô5÷Ia÷I`a÷Iq…÷IÌq…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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‡ïõI÷öI�÷öI•DöI •DöIIöIIöIÉmöI€ÉmöIg®öIüg®öIÕ²öIÕ²öI›×öIp›×öI9÷Iì9÷I§÷Iô§÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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‡ïõI÷öI�÷öI•DöI •DöIIöIIöIÃmöI€ÃmöIa®öIüa®öIÕ²öIÕ²öI•×öIp•×öI3÷Iì3÷I§÷Iô§÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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‡ïõIWöI�WöIõDöI õDöIyKöIyKöI#nöI€#nöIÁ®öIüÁ®öIKµöIKµöIõ×öIpõ×öI“÷Iì“÷I÷Iô÷I a÷I` a÷Ik…÷IÌk…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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�ïõIñöI�ñöI�DöI �DöIñHöIñHöIÃmöI€ÃmöIa®öIüa®öIòöIòöI•×öIp•×öI3÷Iì3÷I•÷Iô•÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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�ïõIWöI�WöIõDöI õDöI…KöI…KöI)nöI€)nöIÇ®öIüÇ®öIWµöIWµöIû×öIpû×öI™÷Iì™÷I)÷Iô)÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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�ïõIñöI�ñöI�DöI �DöIñHöIñHöI½möI€½möI[®öIü[®öIòöIòöI�×öIp�×öI-÷Iì-÷I•÷Iô•÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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�ïõIWöI�WöIõDöI õDöI…KöI…KöI)nöI€)nöIÇ®öIüÇ®öIWµöIWµöIû×öIpû×öI™÷Iì™÷I)÷Iô)÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúI³ûIT ³ûIƒØûI˜ƒØûI+ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIëöI�ëöI‰DöI ‰DöI÷HöI÷HöI½möI€½möI[®öIü[®öIɲöIɲöI�×öIp�×öI-÷Iì-÷I›÷Iô›÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI]öI�]öIûDöI ûDöIsKöIsKöI/nöI€/nöIÍ®öIüÍ®öIEµöIEµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIyûIT yûIIÙûI˜IÙûIñëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIåöI�åöIƒDöI ƒDöI÷HöI÷HöI·möI€·möIU®öIüU®öIɲöIɲöI‰×öIp‰×öI'÷Iì'÷I›÷Iô›÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõI]öI�]öIûDöI ûDöIsKöIsKöI/nöI€/nöIÍ®öIüÍ®öIEµöIEµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷I7_÷I`7_÷I—ƒ÷IÌ—ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIcöI�cöIEöI EöIyKöIyKöI/nöI€/nöIÍ®öIüÍ®öIKµöIKµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷Ii`÷I`i`÷IÉ„÷IÔÉ„÷Ia¶÷Iôa¶÷I-øIT -øI%£øI´ %£øIMwùIx MwùI+GúI| +GúI‘ûI\ ‘ûI‡×ûI ‡×ûI/êûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIßöI�ßöI}DöI }DöIåHöIåHöI±möI€±möIO®öIüO®öI·²öI·²öIƒ×öIpƒ×öI!÷Iì!÷I‰÷I: ‰÷I_÷I* _÷Isƒ÷I sƒ÷Ia¶÷Ir a¶÷I-øIJ -øI颸I" 颸IwùI wùIïFúIâ ïFúIqûIº qûIgÖûIZ gÖûIéûIz €} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIcöI�cöIEöI EöIyKöIyKöI5nöI€5nöIÓ®öIüÓ®öIKµöIKµöIØöIpØöI¥÷Iì¥÷I÷Iô÷I±`÷I`±`÷I…÷IÔ…÷Ia¶÷Iôa¶÷I-øIT -øI £øI´  £øI5wùIx 5wùIGúI| GúIûI\ ûI×ûI ×ûI·éûI¤€} 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„oïõI£öI£öIADöI|ADöI¥IöI„¥IöIÛLöIðÛLöIeVöI|eVöI›YöIH›YöIumöIÔumöI®öIP®öIw³öIXw³öI­¶öIÄ­¶öI7ÀöIP7ÀöImÃöImÃöIG×öI¨G×öIå÷I$ å÷II÷I, I÷I ÷I˜  ÷I *÷I$  *÷I?-÷Ið ?-÷I�Z÷I| �Z÷I«`÷IX«`÷I …÷IÔ …÷IM²÷IôM²÷IÛÀ÷IÐÛÀ÷I—õ÷I\—õ÷IwøI€wøIu øI¬u øI£øI8£øIëxùIüëxùI%eúI%eúIŸrúIàŸrúI+ûIl+ûIûØûI°ûØûI£ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI{öIè{öIEöIdEöIÇKöIlÇKöIQUöIèQUöI‡XöI´‡XöIGnöI@GnöIå®öI¼å®öI™µöIÄ™µöI#¿öI@#¿öIYÂöI YÂöIØöI˜ØöI·÷I·÷Ik÷I k÷Iõ(÷I˜ õ(÷I+,÷Id +,÷ImY÷Ið mY÷IÃ`÷IÌ Ã`÷I#…÷IH #…÷Ie²÷Ihe²÷IÁ÷IDÁ÷IÓõ÷IÐÓõ÷I£øIô£øI¡!øI ¡!øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIËûIàËûI›ØûI$›ØûICëûI(€} 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„oïõI£öI£öIADöI|ADöI¥IöI„¥IöIÛLöIðÛLöIeVöI|eVöI›YöIH›YöIomöIÔomöI ®öIP ®öIw³öIXw³öI­¶öIÄ­¶öI7ÀöIP7ÀöImÃöImÃöIA×öI¨A×öIß÷I$ ß÷II÷I, I÷I ÷I˜  ÷I *÷I$  *÷I?-÷Ið ?-÷I�Z÷I| �Z÷Iu`÷IXu`÷IÕ„÷IÔÕ„÷I²÷Iô²÷IwÁ÷IÐwÁ÷I3ö÷I\3ö÷IƒøI€ƒøI�#øI¬�#øIû¢øI8û¢øIÓxùIüÓxùI eúI eúI‡rúIà‡rúI³ûIl³ûIƒØûI°ƒØûI+ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxoïõI{öIè{öIEöIdEöIÇKöIlÇKöIQUöIèQUöI‡XöI´‡XöIMnöI@MnöIë®öI¼ë®öI™µöIÄ™µöI#¿öI@#¿öIYÂöI YÂöIØöI˜ØöI½÷I½÷Ik÷I k÷Iõ(÷I˜ õ(÷I+,÷Id +,÷ImY÷Ið mY÷I�`÷IÌ �`÷Ií„÷IH í„÷I/²÷Ih/²÷I}Á÷ID}Á÷I9ö÷IÐ9ö÷I•øIô•øI“#øI “#øI颸I¬颸IÁxùIpÁxùIûdúItûdúIurúITurúIkûIàkûI;ØûI$;ØûIãêûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIåöIèåöIƒDöIdƒDöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöI±möI@±möIO®öI¼O®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöIƒ×öI˜ƒ×öI!÷I!÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷IÉ`÷IÌ É`÷I)…÷IH )…÷Ik²÷Ihk²÷I½À÷ID½À÷Iyõ÷IÐyõ÷IÉøIôÉøIÇøI ÇøI£øI¬£øIñxùIpñxùI+eúIt+eúI¥rúIT¥rúIsûIàsûICÙûI$CÙûIëëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIoöIèoöI EöId EöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöIAnöI@AnöIß®öI¼ß®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöIØöI˜ØöI±÷I±÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷Iá`÷IÌ á`÷IA…÷IH A…÷Iƒ²÷Ihƒ²÷IùÀ÷IDùÀ÷Iµõ÷Iеõ÷IõøIôõøIó øI ó øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúI ûIà ûIÝØûI$ÝØûI…ëûI(€} 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$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`JbÀbÀd4e4g¨g¨ijl�l�nÐnÐpDqDs¸s¸u,v,x x zH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çxuïõIßöIèßöI}DöId}DöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöI±möI@±möIO®öI¼O®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöIƒ×öI˜ƒ×öI!÷I!÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷IŸ`÷IÌ Ÿ`÷Iÿ„÷IH ÿ„÷IA²÷IhA²÷I±À÷ID±À÷Imõ÷IÐmõ÷I�øIô�øI‹øI ‹øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIûIàûIÑØûI$ÑØûIyëûI(€} 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·l{ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷I¹¸÷I,¹¸÷ISøI˜SøI£øI$£øIñxùIèñxùI+eúIì+eúIûIÌûIOÙûIOÙûI÷ëûI€} 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·l{ïõIiöIÐiöIEöILEöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI;nöI(;nöIÙ®öI¤Ù®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI ØöI€ ØöI«÷Iü«÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Ia÷I´ a÷Ie…÷I0 e…÷I§²÷IP§²÷I§¸÷I,§¸÷I#øI˜#øI £øI$ £øIåxùIèåxùIeúIìeúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIXJXLÏLÏNFOFQ½Q½S4T4V«V«X"Y"[™[™]^`‡`‡bþbþdueugìgìicjclÚlÚnLoLq¾q¾s0t0v¢v¢xy{†{†}ø}øj€j‚܂܄N…N‡À‡À‰2Š2Œ¤Œ¤Ž�‘ˆ‘ˆ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIåöIÐåöIƒDöILƒDöIÍKöITÍKöIWUöIÐWUöI�XöIœ�XöI·möI(·möIU®öI¤U®öIŸµöI¬ŸµöI)¿öI()¿öI_ÂöIô_ÂöI‰×öI€‰×öI'÷Iü'÷Iq÷I q÷Iû(÷I€ û(÷I1,÷IL 1,÷IsY÷IØ sY÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IG»÷I,G»÷IãøI˜ãøI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIרûIרûIëûI€} 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$K%K'Á'Á)7*7,­,­.#/#1™1™346…6…8û8û:q;q=ç=ç?]@]BÓBÓDIEIG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‰‡‰‰ý‰ý‹qŒqŽåŽå�Y‘Y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIoöIÐoöI EöIL EöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöIAnöI(AnöIß®öI¤ß®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöIØöI€ØöI±÷Iü±÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I5»÷I,5»÷I³øI˜³øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIëöIÐëöI‰DöIL‰DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷I³¸÷I,³¸÷IqøI˜qøI £øI$ £øIåxùIèåxùIeúIìeúIIûIÌIûIÙûIÙûIÁëûI€} 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$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbåbådZeZgÏgÏiDjDl¹l¹n.o.q£q£stv�v�xy{w{w}ì}ìa€a‚Ö‚Ö„K…K‡À‡À‰5Š5ŒªŒªŽ�‘”‘”“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIcöIÐcöIEöILEöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI5nöI(5nöIÓ®öI¤Ó®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöI¥÷Iü¥÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Ia÷I´ a÷Ie…÷I0 e…÷I§²÷IP§²÷I­¸÷I,­¸÷IËøI˜ËøI£øI$£øIëxùIèëxùI%eúIì%eúIIûIÌIûIÙûIÙûIÁëûI€} 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·l{ïõIëöIÐëöI‰DöIL‰DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IA»÷I,A»÷IøI˜øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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$j%j'Þ'Þ)R*R,Æ,Æ.:/:1®1®3"4"6–6–8 9 ;~;~=ò=ò?f@fBÚBÚDNENGÂGÂI6J6L¨L¨NOQŒQŒSþSþUpVpXâXâZT[T]Æ]Æ_8`8bªbªdegŽgŽijlrlrnänäpVqVsÈsÈu:v:x¬x¬z{}�}�€‚t‚t„æ„æ†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIñöIÐñöI�DöIL�DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIÃmöI(ÃmöIa®öI¤a®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI•×öI€•×öI3÷Iü3÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷I³¸÷I,³¸÷I¿øI˜¿øI £øI$ £øIåxùIèåxùIeúIìeúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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$s%s'é'é)_*_,Õ,Õ.K/K1Á1Á37476­6­8#9#;™;™=>@ƒ@ƒB÷B÷DkEkGßGßISJSLÇLÇN;O;Q¯Q¯S#T#V—V—X Y [[]ó]ó_g`gbÛbÛdOeOgÃgÃi7j7l«l«noq“q“stv{v{xïxïzc{c}×}×K€K‚¿‚¿„3…3‡§‡§‰ŠŒ�Œ�Ž�‘s‘s“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI]öIÐ]öIûDöILûDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI/nöI(/nöIÍ®öI¤Í®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIØöI€ØöIŸ÷IüŸ÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷I³¸÷I,³¸÷IøI˜øI£øI$£øIÙxùIèÙxùIeúIìeúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$I%I'¾'¾)3*3,¨,¨./1’1’346|6|8ñ8ñ:f;f=Û=Û?N@NBÁBÁD4E4G§G§IJL�L�NOQsQsSæSæUYVYXÌXÌZ?[?]²]²_%`%b˜b˜d e g~g~iñiñkdldn×n×pJqJs½s½u0v0x£x£z{}‰}‰üü�o‚o„â„â†U‡U‰ljNj9Œ9ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIñöIÐñöI�DöIL�DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI½möI(½möI[®öI¤[®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI�×öI€�×öI-÷Iü-÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IG»÷I,G»÷IOøI˜OøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúI³ûI̳ûIƒØûIƒØûI+ëûI€} 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�ïõIcöIÐcöIEöILEöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI/nöI(/nöIÍ®öI¤Í®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIØöI€ØöIŸ÷IüŸ÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷IS»÷I,S»÷I!øI˜!øIï¢øI$ï¢øIÇxùIèÇxùIeúIìeúIqûIÌqûIAØûIAØûIéêûI€} 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‡ïõIWöIÐWöIõDöILõDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI#nöI(#nöIÁ®öI¤Á®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIõ×öI€õ×öI“÷Iü“÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I³¸÷I,³¸÷I3øI˜3øI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIïØûIïØûI—ëûI€} 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‡ïõI÷öIÐ÷öI•DöIL•DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöIÃmöI(ÃmöIa®öI¤a®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöI•×öI€•×öI3÷Iü3÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷IS»÷I,S»÷IøI˜øI £øI$ £øIåxùIèåxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIWöIÐWöIõDöILõDöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI)nöI()nöIÇ®öI¤Ç®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIû×öI€û×öI™÷Iü™÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I_»÷I,_»÷IÛøI˜ÛøI£øI$£øIßxùIèßxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIýöIÐýöI›DöIL›DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÉmöI(ÉmöIg®öI¤g®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI›×öI€›×öI9÷Iü9÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷I¹¸÷I,¹¸÷I'øI˜'øI+£øI$+£øIyùIèyùI=eúIì=eúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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$m%m'å'å)]*],Õ,Õ.M/M1Å1Å3=4=6µ6µ8-9-;¥;¥=>@•@•B C E…E…GýGýIuJuLëLëNaOaQ×Q×SMTMVÃVÃX9Y9[¯[¯]%^%`›`›bce‡e‡gýgýisjslélén_o_qÕqÕsKtKvÁvÁx7y7{­{­}#~#€™€™‚ƒ…………‡û‡û‰qŠqŒçŒçŽ]�]‘Ó‘Ó“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIQöIÐQöIïDöILïDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöInöI(nöI»®öI¤»®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIï×öI€ï×öI�÷Iü�÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷I¹¸÷I,¹¸÷IçøI˜çøI£øI$£øI÷xùIè÷xùI1eúIì1eúIIûIÌIûIÙûIÙûIÁëûI€} 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$a%a'Ø'Ø)O*O,Æ,Æ.=/=1´1´3+4+6¢6¢89;�;�=>@~@~BõBõDlElGãGãI\J\LÓLÓNJOJQÁQÁS8T8V¯V¯X&Y&[�[�]^`‹`‹bceyeygðgðigjglÞlÞnPoPqÂqÂs4t4v¦v¦xy{Š{Š}ü}ün€n‚à‚à„R…R‡ćĉ6Š6Œ¨Œ¨Ž�‘Œ‘Œ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI÷öIÐ÷öI•DöIL•DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÉmöI(ÉmöIg®öI¤g®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI›×öI€›×öI9÷Iü9÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷Ie»÷I,e»÷IÏøI˜ÏøI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIéØûIéØûI‘ëûI€} 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$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIQöIÐQöIïDöILïDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI#nöI(#nöIÁ®öI¤Á®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIõ×öI€õ×öI“÷Iü“÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷Iq»÷I,q»÷I›øI˜›øI £øI$ £øIåxùIèåxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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·l�ïõIöIÐöI¡DöIL¡DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÏmöI(ÏmöIm®öI¤m®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI¡×öI€¡×öI?÷Iü?÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷I¿¸÷I,¿¸÷IÛøI˜ÛøI%£øI$%£øIýxùIèýxùI7eúIì7eúI[ûIÌ[ûI+ÙûI+ÙûIÓëûI€} 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·l�ïõIEöIÐEöIãDöILãDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöInöI(nöIµ®öI¤µ®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIé×öI€é×öI‡÷Iü‡÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷I¿¸÷I,¿¸÷I›øI˜›øI%£øI$%£øIýxùIèýxùI7eúIì7eúIaûIÌaûI1ÙûI1ÙûIÙëûI€} 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·l�ïõIýöIÐýöI›DöIL›DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÏmöI(ÏmöIm®öI¤m®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI¡×öI€¡×öI?÷Iü?÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷Iq»÷I,q»÷I�øI˜�øI£øI$£øIëxùIèëxùI%eúIì%eúIõûIÌõûIÅØûIÅØûImëûI€} 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·l�ïõIKöIÐKöIéDöILéDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöInöI(nöI»®öI¤»®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIï×öI€ï×öI�÷Iü�÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷I}»÷I,}»÷I[øI˜[øI £øI$ £øIåxùIèåxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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$n%n'â'â)V*V,Ê,Ê.>/>1²1²3&4&6š6š89;‚;‚=ö=ö?j@jBÞBÞDRERGÆGÆI:J:L¬L¬NOQ�Q�STVtVtXæXæZX[X]Ê]Ê_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI¹DöI¹DöIIöI$IöIímöI�ímöI‹®öI‹®öIÕ²öI$Õ²öI¿×öI�¿×öI]÷I]÷I§÷I$§÷Ií`÷I�í`÷IM…÷IüM…÷I¶÷I ¶÷IKøI| KøI1£øIÜ 1£øI yùI   yùICeúI¤ CeúIOûI„ OûIÙûIÈÙûIÇëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIEöI�EöIãDöI ãDöI£KöI£KöInöI€nöI¯®öIü¯®öIuµöIuµöIã×öIpã×öI�÷Iì�÷IG÷IôG÷Ia÷I`a÷Iw…÷IÌw…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=æ=æ?Y@YBÌBÌD?E?G²G²I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†^‡^‰ЉЋBŒBŽ´Ž´�&‘&“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI³DöI³DöIIöI$IöIçmöI�çmöI…®öI…®öIÕ²öI$Õ²öI¹×öI�¹×öIW÷IW÷I§÷I$§÷Iù`÷I�ù`÷IY…÷IüY…÷I¶÷I ¶÷IKøI| KøI£øIÜ £øIñxùI  ñxùI+eúI¤ +eúI×ûI„ ×ûI§ØûIȧØûIOëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIEöI�EöIãDöI ãDöI�KöI�KöInöI€nöIµ®öIüµ®öIoµöIoµöIé×öIpé×öI‡÷Iì‡÷IA÷IôA÷Ia÷I`a÷Iw…÷IÌw…÷I¶÷Iì¶÷IKøIL KøI £øI¬  £øIåxùIp åxùIeúIt eúI�ûIT �ûI_ØûI˜_ØûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI³DöI ³DöIýHöIýHöIámöI€ámöI®öIü®öIϲöIϲöI³×öIp³×öIQ÷IìQ÷I¡÷Iô¡÷Iç`÷I`ç`÷IG…÷IÌG…÷Iy¶÷Iìy¶÷IEøIL EøI=£øI¬ =£øIyùIp yùIOeúIt OeúI—ûIT —ûIgÙûI˜gÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIKöI�KöIéDöI éDöI‘KöI‘KöInöI€nöIµ®öIüµ®öIcµöIcµöIé×öIpé×öI‡÷Iì‡÷I5÷Iô5÷I a÷I` a÷Ik…÷IÌk…÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI­DöI ­DöIýHöIýHöIámöI€ámöI®öIü®öIϲöIϲöI³×öIp³×öIQ÷IìQ÷I¡÷Iô¡÷Ió`÷I`ó`÷IS…÷IÌS…÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIKöI�KöIéDöI éDöI‘KöI‘KöInöI€nöI»®öIü»®öIcµöIcµöIï×öIpï×öI�÷Iì�÷I5÷Iô5÷Ia÷I`a÷Iw…÷IÌw…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI­DöI ­DöIIöIIöIÛmöI€ÛmöIy®öIüy®öIÕ²öIÕ²öI­×öIp­×öIK÷IìK÷I§÷Iô§÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Iy¶÷Iìy¶÷IEøIL EøI7£øI¬ 7£øIyùIp yùIIeúIt IeúI£ûIT £ûIsÙûI˜sÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIKöI�KöIéDöI éDöIKöIKöInöI€nöI»®öIü»®öIQµöIQµöIï×öIpï×öI�÷Iì�÷I#÷Iô#÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI öI� öI§DöI §DöI IöI IöIÛmöI€ÛmöIy®öIüy®öIÛ²öIÛ²öI­×öIp­×öIK÷IìK÷I­÷Iô­÷Ií`÷I`í`÷IM…÷IÌM…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIQöI�QöIïDöI ïDöIKöIKöInöI€nöI»®öIü»®öIQµöIQµöIï×öIpï×öI�÷Iì�÷I#÷Iô#÷Ia÷I`a÷Iw…÷IÌw…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¡DöI ¡DöI÷HöI÷HöIÕmöI€ÕmöIs®öIüs®öIɲöIɲöI§×öIp§×öIE÷IìE÷I›÷Iô›÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIQöI�QöIïDöI ïDöI‘KöI‘KöI#nöI€#nöIÁ®öIüÁ®öIcµöIcµöIõ×öIpõ×öI“÷Iì“÷I5÷Iô5÷Iù`÷I`ù`÷IY…÷IÌY…÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¡DöI ¡DöIýHöIýHöIÏmöI€ÏmöIm®öIüm®öIϲöIϲöI¡×öIp¡×öI?÷Iì?÷I¡÷Iô¡÷Ií`÷I`í`÷IM…÷IÌM…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIWöI�WöIõDöI õDöI‘KöI‘KöI#nöI€#nöIÁ®öIüÁ®öIcµöIcµöIõ×öIpõ×öI“÷Iì“÷I5÷Iô5÷Ia÷I`a÷Iq…÷IÌq…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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‡ïõIýöI�ýöI›DöI ›DöIIöIIöIÏmöI€ÏmöIm®öIüm®öIÕ²öIÕ²öI¡×öIp¡×öI?÷Iì?÷I§÷Iô§÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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‡ïõIWöI�WöIõDöI õDöIyKöIyKöI)nöI€)nöIÇ®öIüÇ®öIKµöIKµöIû×öIpû×öI™÷Iì™÷I÷Iô÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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‡ïõIýöI�ýöI›DöI ›DöIIöIIöIÉmöI€ÉmöIg®öIüg®öIÕ²öIÕ²öI›×öIp›×öI9÷Iì9÷I§÷Iô§÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI¹ûIT ¹ûI‰ØûI˜‰ØûI1ëûIœ€} 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�ïõI÷öI�÷öI•DöI •DöIñHöIñHöIÉmöI€ÉmöIg®öIüg®öIòöIòöI›×öIp›×öI9÷Iì9÷I•÷Iô•÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIyûIT yûIIÙûI˜IÙûIñëûIœ€} 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�ïõI]öI�]öIûDöI ûDöI…KöI…KöI/nöI€/nöIÍ®öIüÍ®öIWµöIWµöIØöIpØöIŸ÷IìŸ÷I)÷Iô)÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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�ïõIñöI�ñöI�DöI �DöIñHöIñHöIÃmöI€ÃmöIa®öIüa®öIòöIòöI•×öIp•×öI3÷Iì3÷I•÷Iô•÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIרûI˜רûIëûIœ€} 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�ïõI]öI�]öIûDöI ûDöI…KöI…KöI/nöI€/nöIÍ®öIüÍ®öIWµöIWµöIØöIpØöIŸ÷IìŸ÷I)÷Iô)÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Im¶÷Iìm¶÷I9øIL 9øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIñöI�ñöI�DöI �DöI÷HöI÷HöIÃmöI€ÃmöIa®öIüa®öIɲöIɲöI•×öIp•×öI3÷Iì3÷I›÷Iô›÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI…ûIT …ûIUÙûI˜UÙûIýëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIcöI�cöIEöI EöIsKöIsKöI/nöI€/nöIÍ®öIüÍ®öIEµöIEµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIëöI�ëöI‰DöI ‰DöI÷HöI÷HöI½möI€½möI[®öIü[®öIɲöIɲöI�×öIp�×öI-÷Iì-÷I›÷Iô›÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIcöI�cöIEöI EöIsKöIsKöI5nöI€5nöIÓ®öIüÓ®öIEµöIEµöIØöIpØöI¥÷Iì¥÷I÷Iô÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIëöI�ëöI‰DöI ‰DöIåHöIåHöI·möI€·möIU®öIüU®öI·²öI·²öI‰×öIp‰×öI'÷Iì'÷I‰÷Iô‰÷I{`÷I`{`÷IÛ„÷IÔÛ„÷Ia¶÷Iôa¶÷I-øIT -øI£øI´ £øIAwùIx AwùIGúI| GúIOûI\ OûIE×ûI E×ûIíéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIiöI�iöIEöI EöIyKöIyKöI5nöI€5nöIÓ®öIüÓ®öIKµöIKµöIØöIpØöI¥÷Iì¥÷I÷Iô÷I{`÷I`{`÷IÛ„÷IÔÛ„÷Ia¶÷Iôa¶÷I-øIT -øI£øI´ £øIGwùIx GwùI%GúI| %GúIUûI\ UûIK×ûI K×ûIóéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIåöI�åöIƒDöI ƒDöIåHöIåHöI·möI€·möIU®öIüU®öI·²öI·²öI‰×öIp‰×öI'÷Iì'÷I‰÷Iô‰÷I«`÷I`«`÷I …÷IÔ …÷Ia¶÷Iôa¶÷I-øIT -øI£øI´ £øI/wùIx /wùI GúI|  GúIéûI\ éûIßÖûI ßÖûI‡éûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIiöI�iöIEöI EöIKöIKöI;nöI€;nöIÙ®öIüÙ®öIQµöIQµöI ØöIp ØöI«÷Iì«÷I#÷Iô#÷IÃ`÷I`Ã`÷I#…÷IÔ#…÷Ia¶÷Iôa¶÷I-øIT -øI£øI´ £øI/wùIx /wùI GúI|  GúIÝûI\ ÝûIÓÖûI ÓÖûI{éûI¤€} 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„uïõI©öI©öIGDöI|GDöI«IöI„«IöIáLöIðáLöIkVöI|kVöI¡YöIH¡YöIumöIÔumöI®öIP®öI}³öIX}³öI³¶öIij¶öI=ÀöIP=ÀöIsÃöIsÃöIG×öI¨G×öIå÷I$ å÷IO÷I, O÷I… ÷I˜ … ÷I*÷I$ *÷IE-÷Ið E-÷I‡Z÷I| ‡Z÷I¥`÷IX¥`÷I…÷IÔ…÷IG²÷IôG²÷IÁ÷IÐÁ÷IÁõ÷I\Áõ÷IUøI€UøIS!øI¬S!øI£øI8£øIÙxùIüÙxùIeúIeúI�rúIà�rúIÑûIlÑûI¡ØûI°¡ØûIIëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI{öIè{öIEöIdEöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöIMnöI@MnöIë®öI¼ë®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöIØöI˜ØöI½÷I½÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷IÉ`÷IÌ É`÷I)…÷IH )…÷Ik²÷Ihk²÷IíÀ÷IDíÀ÷I©õ÷IЩõ÷IÅøIôÅøIà øI à øI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúI+ûIà+ûIûØûI$ûØûI£ëûI(€} 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„uïõI£öI£öIADöI|ADöI«IöI„«IöIáLöIðáLöIkVöI|kVöI¡YöIH¡YöIumöIÔumöI®öIP®öI}³öIX}³öI³¶öIij¶öI=ÀöIP=ÀöIsÃöIsÃöIG×öI¨G×öIå÷I$ å÷IO÷I, O÷I… ÷I˜ … ÷I*÷I$ *÷IE-÷Ið E-÷I‡Z÷I| ‡Z÷Io`÷IXo`÷IÏ„÷IÔÏ„÷I²÷Iô²÷I¡Á÷IСÁ÷I]ö÷I\]ö÷IaøI€aøI_$øI¬_$øI颸I8颸IÁxùIüÁxùIûdúIûdúIurúIàurúIqûIlqûIAØûI°AØûIéêûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI�öIè�öIEöIdEöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöISnöI@SnöIñ®öI¼ñ®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöI%ØöI˜%ØöIÃ÷IÃ÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷IŸ`÷IÌ Ÿ`÷Iÿ„÷IH ÿ„÷IA²÷IhA²÷IMÁ÷IDMÁ÷I ö÷IÐ ö÷I½øIô½øI»"øI »"øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúI³ûIà³ûIƒØûI$ƒØûI+ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIåöIèåöIƒDöIdƒDöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöI·möI@·möIU®öI¼U®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöI‰×öI˜‰×öI'÷I'÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷IÃ`÷IÌ Ã`÷I#…÷IH #…÷Ie²÷Ihe²÷IçÀ÷IDçÀ÷I£õ÷IУõ÷I¡øIô¡øIŸ øI Ÿ øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúIûIàûIãØûI$ãØûI‹ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIuöIèuöIEöIdEöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöIGnöI@GnöIå®öI¼å®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöIØöI˜ØöI·÷I·÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷Ií`÷IÌ í`÷IM…÷IH M…÷I�²÷Ih�²÷IÉÀ÷IDÉÀ÷I…õ÷IÐ…õ÷IøIôøI øI  øI£øI¬£øIñxùIpñxùI+eúIt+eúI¥rúIT¥rúIsûIàsûICÙûI$CÙûIëëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIåöIèåöIƒDöIdƒDöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöI·möI@·möIU®öI¼U®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöI‰×öI˜‰×öI'÷I'÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷IÃÀ÷IDÃÀ÷Iõ÷IÐõ÷IíøIôíøIëøI ëøIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIËûIàËûI›ØûI$›ØûICëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI{öIè{öIEöIdEöIÍKöIlÍKöIWUöIèWUöI�XöI´�XöIMnöI@MnöIë®öI¼ë®öIŸµöIÄŸµöI)¿öI@)¿öI_ÂöI _ÂöIØöI˜ØöI½÷I½÷Iq÷I q÷Iû(÷I˜ û(÷I1,÷Id 1,÷IsY÷Ið sY÷I«`÷IÌ «`÷I …÷IH  …÷IM²÷IhM²÷I·À÷ID·À÷Isõ÷IÐsõ÷I5øIô5øI3 øI 3 øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIûIàûIÑØûI$ÑØûIyëûI(€} 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·l{ïõIëöIÐëöI‰DöIL‰DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷IÕ½÷I,Õ½÷ImøI˜møI£øI$£øIßxùIèßxùIeúIìeúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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·l{ïõIoöIÐoöI EöIL EöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöIAnöI(AnöIß®öI¤ß®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIØöI€ØöI±÷Iü±÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I a÷I´  a÷Ik…÷I0 k…÷I­²÷IP­²÷Ií½÷I,í½÷IiøI˜iøI£øI$£øIñxùIèñxùI+eúIì+eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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·l{ïõIëöIÐëöI‰DöIL‰DöIÓKöITÓKöI]UöIÐ]UöI“XöIœ“XöI½möI(½möI[®öI¤[®öI¥µöI¬¥µöI/¿öI(/¿öIeÂöIôeÂöI�×öI€�×öI-÷Iü-÷Iw÷I w÷I)÷I€ )÷I7,÷IL 7,÷IyY÷IØ yY÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷I]À÷I,]À÷IýøI˜ýøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l{ïõIuöIÐuöIEöILEöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöIAnöI(AnöIß®öI¤ß®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöIØöI€ØöI±÷Iü±÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I‡À÷I,‡À÷IøI˜øI£øI$£øIÙxùIèÙxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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·l�ïõIñöIÐñöI�DöIL�DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöIÃmöI(ÃmöIa®öI¤a®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöI•×öI€•×öI3÷Iü3÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷IϽ÷I,Ͻ÷I‹øI˜‹øI£øI$£øIëxùIèëxùI%eúIì%eúIIûIÌIûIÙûIÙûIÁëûI€} 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·l�ïõIñöIÐñöI�DöIL�DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI½möI(½möI[®öI¤[®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöI�×öI€�×öI-÷Iü-÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷IcÀ÷I,cÀ÷IøI˜øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpnânâpTqTsÆsÆu8v8xªxªz{}Ž}Ž€‚r‚r„ä„ä†V‡V‰ȉȋ:Œ:ެެ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIoöIÐoöI EöIL EöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI;nöI(;nöIÙ®öI¤Ù®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI ØöI€ ØöI«÷Iü«÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷I{À÷I,{À÷IáøI˜áøI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Ö‰Ö‹IŒI޼޼�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI÷öIÐ÷öI•DöIL•DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIÉmöI(ÉmöIg®öI¤g®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI›×öI€›×öI9÷Iü9÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷IÛ½÷I,Û½÷IåøI˜åøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$E%E'º'º)/*/,¤,¤./1Ž1Ž346x6x8í8í:b;b=×=×?J@JB½B½D0E0G£G£IJL‰L‰NüNüPoQoSâSâUUVUXÈXÈZ;[;]®]®_!`!b”b”degzgziíiík`l`nÓnÓpFqFs¹s¹u,v,xŸxŸz{}…}…øø�k‚k„Þ„Þ†Q‡Q‰ÉË5Œ5ާާ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIcöIÐcöIEöILEöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI/nöI(/nöIÍ®öI¤Í®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöIŸ÷IüŸ÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷Iÿ½÷I,ÿ½÷I øI˜ øI£øI$£øIëxùIèëxùI%eúIì%eúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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�ïõI÷öIÐ÷öI•DöIL•DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÃmöI(ÃmöIa®öI¤a®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI•×öI€•×öI3÷Iü3÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷IoÀ÷I,oÀ÷I{øI˜{øI颸I$颸IÁxùIèÁxùIûdúIìûdúIqûIÌqûIAØûIAØûIéêûI€} 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�ïõIcöIÐcöIEöILEöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI5nöI(5nöIÓ®öI¤Ó®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöI¥÷Iü¥÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IŸÀ÷I,ŸÀ÷I­øI˜­øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúI¹ûI̹ûI‰ØûI‰ØûI1ëûI€} 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·l‡ïõIýöIÐýöI›DöIL›DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÉmöI(ÉmöIg®öI¤g®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI›×öI€›×öI9÷Iü9÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷Ió½÷I,ó½÷I±øI˜±øI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIéØûIéØûI‘ëûI€} 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$Z%Z'Ï'Ï)D*D,¹,¹../.1£1£346�6�89;w;w=ì=ì?a@aBÖBÖDKEKGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡Œ‡Œ‰ŠŒvŒvŽëŽë�`‘`“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI]öIÐ]öIûDöILûDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI)nöI()nöIÇ®öI¤Ç®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIû×öI€û×öI™÷Iü™÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷I¾÷I,¾÷IÑøI˜ÑøI+£øI$+£øIyùIèyùI=eúIì=eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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·l‡ïõI÷öIÐ÷öI•DöIL•DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÉmöI(ÉmöIg®öI¤g®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI›×öI€›×öI9÷Iü9÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷I“À÷I,“À÷IMøI˜MøI£øI$£øIßxùIèßxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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·l‡ïõI]öIÐ]öIûDöILûDöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöI/nöI(/nöIÍ®öI¤Í®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöIŸ÷IüŸ÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷IÃ`÷I´ Ã`÷I#…÷I0 #…÷Ie²÷IPe²÷IÃÀ÷I,ÃÀ÷IøI˜øI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIãØûIãØûI‹ëûI€} 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·l�ïõIöIÐöI¡DöIL¡DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÏmöI(ÏmöIm®öI¤m®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI¡×öI€¡×öI?÷Iü?÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I ¾÷I, ¾÷I}øI˜}øI£øI$£øI÷xùIè÷xùI1eúIì1eúIIûIÌIûIÙûIÙûIÁëûI€} 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·l�ïõIQöIÐQöIïDöILïDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI#nöI(#nöIÁ®öI¤Á®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIõ×öI€õ×öI“÷Iü“÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷Iu`÷I´ u`÷IÕ„÷I0 Õ„÷I²÷IP²÷I/¾÷I,/¾÷I�øI˜�øI1£øI$1£øI yùIè yùICeúIìCeúI—ûIÌ—ûIgÙûIgÙûIìûI€} 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·l�ïõIýöIÐýöI›DöIL›DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÏmöI(ÏmöIm®öI¤m®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI¡×öI€¡×öI?÷Iü?÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷I·À÷I,·À÷I%øI˜%øI £øI$ £øIåxùIèåxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIWöIÐWöIõDöILõDöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI)nöI()nöIÇ®öI¤Ç®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIû×öI€û×öI™÷Iü™÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷IçÀ÷I,çÀ÷IWøI˜WøI£øI$£øIñxùIèñxùI+eúIì+eúIûIÌûIïØûIïØûI—ëûI€} 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$Q%Q'Ç'Ç)=*=,³,³.)/)1Ÿ1Ÿ346‹6‹89;w;w=í=í?c@cBÙBÙDOEOGÅGÅI;J;L¯L¯N#O#Q—Q—S T VVXóXóZg[g]Û]Û_O`ObÃbÃd7e7g«g«ijl“l“noq{q{sïsïucvcx×x×zK{K}¿}¿3€3‚§‚§„…‡�‡�‰ŠŒwŒwŽëŽë�_‘_“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷I)¾÷I,)¾÷ICøI˜CøI%£øI$%£øIýxùIèýxùI7eúIì7eúI[ûIÌ[ûI+ÙûI+ÙûIÓëûI€} 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·l�ïõIKöIÐKöIéDöILéDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöInöI(nöI»®öI¤»®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIï×öI€ï×öI�÷Iü�÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Ii`÷I´ i`÷IÉ„÷I0 É„÷I ²÷IP ²÷IM¾÷I,M¾÷IiøI˜iøI+£øI$+£øIyùIèyùI=eúIì=eúIaûIÌaûI1ÙûI1ÙûIÙëûI€} 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·l�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷IÛÀ÷I,ÛÀ÷I÷øI˜÷øI £øI$ £øIåxùIèåxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$l%l'à'à)T*T,È,È.Œ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIQöIÐQöIïDöILïDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI#nöI(#nöIÁ®öI¤Á®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIõ×öI€õ×öI“÷Iü“÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IÁ÷I,Á÷I)øI˜)øI£øI$£øIëxùIèëxùI%eúIì%eúIõûIÌõûIÅØûIÅØûImëûI€} 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$Q%Q'Æ'Æ);*;,°,°.%/%1š1š346„6„8ù8ù:n;n=ã=ã?X@XBÍBÍDBEBG·G·I,J,LŸLŸNOQ…Q…SøSøUkVkXÞXÞZQ[Q]Ä]Ä_7`7bªbªdeg�g�ijlvlvnénép\q\sÏsÏuBvBxµxµz({(}›}›€‚�‚�„ô„ô†g‡g‰Ú‰Ú‹MŒMŽÀŽÀ�3‘3“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L™ïõI!öI�!öI¿DöI¿DöIIöI$IöIímöI�ímöI‹®öI‹®öIÕ²öI$Õ²öI¿×öI�¿×öI]÷I]÷I§÷I$§÷Iù`÷I�ù`÷IY…÷IüY…÷I¶÷I ¶÷IKøI| KøI%£øIÜ %£øIýxùI  ýxùI7eúI¤ 7eúIõûI„ õûIÅØûIÈÅØûImëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIKöI�KöIéDöI éDöI�KöI�KöInöI€nöIµ®öIüµ®öIoµöIoµöIé×öIpé×öI‡÷Iì‡÷IA÷IôA÷Ia÷I`a÷Iq…÷IÌq…÷I¶÷Iì¶÷IKøIL KøI1£øI¬ 1£øI yùIp  yùICeúIt CeúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIöI�öI¹DöI¹DöIIöI$IöIímöI�ímöI‹®öI‹®öIÕ²öI$Õ²öI¿×öI�¿×öI]÷I]÷I§÷I$§÷Iù`÷I�ù`÷IY…÷IüY…÷I¶÷I ¶÷IKøI| KøI £øIÜ  £øIåxùI  åxùIeúI¤ eúI�ûI„ �ûI_ØûIÈ_ØûIëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIKöI�KöIéDöI éDöI�KöI�KöInöI€nöI»®öIü»®öIoµöIoµöIï×öIpï×öI�÷Iì�÷IA÷IôA÷Ia÷I`a÷I}…÷IÌ}…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øIñxùIp ñxùI+eúIt +eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI¹DöI ¹DöIIöIIöIçmöI€çmöI…®öIü…®öIÕ²öIÕ²öI¹×öIp¹×öIW÷IìW÷I§÷Iô§÷Ií`÷I`í`÷IM…÷IÌM…÷I¶÷Iì¶÷IKøIL KøI+£øI¬ +£øIyùIp yùI=eúIt =eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIKöI�KöIéDöI éDöI‘KöI‘KöInöI€nöI»®öIü»®öIcµöIcµöIï×öIpï×öI�÷Iì�÷I5÷Iô5÷Ia÷I`a÷Ie…÷IÌe…÷I¶÷Iì¶÷IKøIL KøI7£øI¬ 7£øIyùIp yùIIeúIt IeúI—ûIT —ûIgÙûI˜gÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI³DöI ³DöIIöIIöIçmöI€çmöI…®öIü…®öIÕ²öIÕ²öI¹×öIp¹×öIW÷IìW÷I§÷Iô§÷Iù`÷I`ù`÷IY…÷IÌY…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIQöI�QöIïDöI ïDöI‘KöI‘KöInöI€nöI»®öIü»®öIcµöIcµöIï×öIpï×öI�÷Iì�÷I5÷Iô5÷Ia÷I`a÷Iw…÷IÌw…÷I¶÷Iì¶÷IKøIL KøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI­DöI ­DöI IöI IöIámöI€ámöI®öIü®öIÛ²öIÛ²öI³×öIp³×öIQ÷IìQ÷I­÷Iô­÷Iá`÷I`á`÷IA…÷IÌA…÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIQöI�QöIïDöI ïDöIKöIKöI#nöI€#nöIÁ®öIüÁ®öIQµöIQµöIõ×öIpõ×öI“÷Iì“÷I#÷Iô#÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI7£øI¬ 7£øIyùIp yùIIeúIt IeúI�ûIT �ûImÙûI˜mÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI­DöI ­DöI IöI IöIÛmöI€ÛmöIy®öIüy®öIÛ²öIÛ²öI­×öIp­×öIK÷IìK÷I­÷Iô­÷Iù`÷I`ù`÷IY…÷IÌY…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIWöI�WöIõDöI õDöIKöIKöI#nöI€#nöIÁ®öIüÁ®öIQµöIQµöIõ×öIpõ×öI“÷Iì“÷I#÷Iô#÷Ia÷I`a÷Iq…÷IÌq…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIWöI�WöIõDöI õDöI—KöI—KöI)nöI€)nöIÇ®öIüÇ®öIiµöIiµöIû×öIpû×öI™÷Iì™÷I;÷Iô;÷Ió`÷I`ó`÷IS…÷IÌS…÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI öI� öI§DöI §DöI÷HöI÷HöIÕmöI€ÕmöIs®öIüs®öIɲöIɲöI§×öIp§×öIE÷IìE÷I›÷Iô›÷Ió`÷I`ó`÷IS…÷IÌS…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIWöI�WöIõDöI õDöI—KöI—KöI)nöI€)nöIÇ®öIüÇ®öIiµöIiµöIû×öIpû×öI™÷Iì™÷I;÷Iô;÷Ia÷I`a÷Iq…÷IÌq…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¡DöI ¡DöIIöIIöIÕmöI€ÕmöIs®öIüs®öIÕ²öIÕ²öI§×öIp§×öIE÷IìE÷I§÷Iô§÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI]öI�]öIûDöI ûDöIyKöIyKöI/nöI€/nöIÍ®öIüÍ®öIKµöIKµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIýöI�ýöI›DöI ›DöIIöIIöIÏmöI€ÏmöIm®öIüm®öIÕ²öIÕ²öI¡×öIp¡×öI?÷Iì?÷I§÷Iô§÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Is¶÷Iìs¶÷I?øIL ?øIõ¢øI¬ õ¢øIÍxùIp ÍxùIeúIt eúIwûIT wûIGØûI˜GØûIïêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI]öI�]öIûDöI ûDöIyKöIyKöI/nöI€/nöIÍ®öIüÍ®öIKµöIKµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI¹ûIT ¹ûI‰ØûI˜‰ØûI1ëûIœ€} 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�ïõIýöI�ýöI›DöI ›DöIñHöIñHöIÏmöI€ÏmöIm®öIüm®öIòöIòöI¡×öIp¡×öI?÷Iì?÷I•÷Iô•÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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�ïõIcöI�cöIEöI EöI‹KöI‹KöI/nöI€/nöIÍ®öIüÍ®öI]µöI]µöIØöIpØöIŸ÷IìŸ÷I/÷Iô/÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIyûIT yûIIÙûI˜IÙûIñëûIœ€} 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�ïõI÷öI�÷öI•DöI •DöIñHöIñHöIÉmöI€ÉmöIg®öIüg®öIòöIòöI›×öIp›×öI9÷Iì9÷I•÷Iô•÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI÷öI�÷öI•DöI •DöIýHöIýHöIÃmöI€ÃmöIa®öIüa®öIϲöIϲöI•×öIp•×öI3÷Iì3÷I¡÷Iô¡÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIiöI�iöIEöI EöImKöImKöI5nöI€5nöIÓ®öIüÓ®öI?µöI?µöIØöIpØöI¥÷Iì¥÷I÷Iô÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIiöI�iöIEöI EöImKöImKöI;nöI€;nöIÙ®öIüÙ®öI?µöI?µöI ØöIp ØöI«÷Iì«÷I÷Iô÷I=_÷I`=_÷I�ƒ÷IÌ�ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIñöI�ñöI�DöI �DöIëHöIëHöI½möI€½möI[®öIü[®öI½²öI½²öI�×öIp�×öI-÷Iì-÷I�÷Iô�÷I{`÷I`{`÷IÛ„÷IÔÛ„÷Ig¶÷Iôg¶÷I3øIT 3øI£øI´ £øIAwùIx AwùIGúI| GúIOûI\ OûIE×ûI E×ûIíéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIiöI�iöIEöI EöIKöIKöI;nöI€;nöIÙ®öIüÙ®öIQµöIQµöI ØöIp ØöI«÷Iì«÷I#÷Iô#÷I{`÷I`{`÷IÛ„÷IÔÛ„÷Ig¶÷Iôg¶÷I3øIT 3øI£øI´ £øIGwùIx GwùI%GúI| %GúIUûI\ UûIK×ûI K×ûIóéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIëöI�ëöI‰DöI ‰DöIëHöIëHöI½möI€½möI[®öIü[®öI½²öI½²öI�×öIp�×öI-÷Iì-÷I�÷Iô�÷I¥`÷I`¥`÷I…÷IÔ…÷Ig¶÷Iôg¶÷I3øIT 3øI£øI´ £øI)wùIx )wùIGúI| GúIÝûI\ ÝûIÓÖûI ÓÖûI{éûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIoöI�oöI EöI  EöIKöIKöI;nöI€;nöIÙ®öIüÙ®öIQµöIQµöI ØöIp ØöI«÷Iì«÷I#÷Iô#÷IÛ`÷I`Û`÷I;…÷IÔ;…÷Ig¶÷Iôg¶÷I3øIT 3øI£øI´ £øI/wùIx /wùI GúI|  GúIéûI\ éûIßÖûI ßÖûI‡éûI¤€} 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$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I)J)LœLœNOQ‚Q‚SõSõUhVhXÛXÛZN[N]Á]Á_4`4b§b§deg�g�ijlslsnònòpeqesØsØuKvKx¾x¾z1{1}¤}¤€‚ЂЄý„ý†p‡p‰ã‰ã‹VŒVŽÉŽÉ�<‘<“ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„uïõI©öI©öIGDöI|GDöI«IöI„«IöIáLöIðáLöIkVöI|kVöI¡YöIH¡YöI{möIÔ{möI®öIP®öI}³öIX}³öI³¶öIij¶öI=ÀöIP=ÀöIsÃöIsÃöIM×öI¨M×öIë÷I$ ë÷IO÷I, O÷I… ÷I˜ … ÷I*÷I$ *÷IE-÷Ið E-÷I‡Z÷I| ‡Z÷Io`÷IXo`÷IÏ„÷IÔÏ„÷I²÷Iô²÷I½À÷IнÀ÷Iyõ÷I\yõ÷IÛøI€ÛøIÙøI¬ÙøI£øI8£øIñxùIüñxùI+eúI+eúI¥rúIà¥rúIOûIlOûIÙûI°ÙûIÇëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI�öIè�öIEöIdEöIÓKöIlÓKöI]UöIè]UöI“XöI´“XöISnöI@SnöIñ®öI¼ñ®öI¥µöIÄ¥µöI/¿öI@/¿öIeÂöI eÂöI%ØöI˜%ØöIÃ÷IÃ÷Iw÷I w÷I)÷I˜ )÷I7,÷Id 7,÷IyY÷Ið yY÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷IÃÀ÷IDÃÀ÷Iõ÷IÐõ÷IíøIôíøIëøI ëøI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúIûIàûIãØûI$ãØûI‹ëûI(€} 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„uïõI©öI©öIGDöI|GDöI«IöI„«IöIáLöIðáLöIkVöI|kVöI¡YöIH¡YöIumöIÔumöI®öIP®öI}³öIX}³öI³¶öIij¶öI=ÀöIP=ÀöIsÃöIsÃöIG×öI¨G×öIå÷I$ å÷IO÷I, O÷I… ÷I˜ … ÷I*÷I$ *÷IE-÷Ið E-÷I‡Z÷I| ‡Z÷I“`÷IX“`÷Ió„÷IÔó„÷I5²÷Iô5²÷I½À÷IнÀ÷Iyõ÷I\yõ÷IÛøI€ÛøIÙøI¬ÙøI£øI8£øIÙxùIüÙxùIeúIeúI�rúIà�rúIéûIléûI¹ØûI°¹ØûIaëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI‡öIè‡öI%EöId%EöIÓKöIlÓKöI]UöIè]UöI“XöI´“XöIYnöI@YnöI÷®öI¼÷®öI¥µöIÄ¥µöI/¿öI@/¿öIeÂöI eÂöI+ØöI˜+ØöIÉ÷IÉ÷Iw÷I w÷I)÷I˜ )÷I7,÷Id 7,÷IyY÷Ið yY÷I¥`÷IÌ ¥`÷I…÷IH …÷IG²÷IhG²÷I#Á÷ID#Á÷Ißõ÷IÐßõ÷IßøIôßøIÝ!øI Ý!øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIÑûIàÑûI¡ØûI$¡ØûIIëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIëöIèëöI‰DöId‰DöIÓKöIlÓKöI]UöIè]UöI“XöI´“XöI½möI@½möI[®öI¼[®öI¥µöIÄ¥µöI/¿öI@/¿öIeÂöI eÂöI�×öI˜�×öI-÷I-÷Iw÷I w÷I)÷I˜ )÷I7,÷Id 7,÷IyY÷Ið yY÷I½`÷IÌ ½`÷I…÷IH …÷I_²÷Ih_²÷IÁ÷IDÁ÷IÓõ÷IÐÓõ÷I…øIô…øIƒ!øI ƒ!øI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúI%ûIà%ûIõØûI$õØûI�ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI{öIè{öIEöIdEöIÓKöIlÓKöI]UöIè]UöI“XöI´“XöIMnöI@MnöIë®öI¼ë®öI¥µöIÄ¥µöI/¿öI@/¿öIeÂöI eÂöIØöI˜ØöI½÷I½÷Iw÷I w÷I)÷I˜ )÷I7,÷Id 7,÷IyY÷Ið yY÷Iù`÷IÌ ù`÷IY…÷IH Y…÷I›²÷Ih›²÷IÃÀ÷IDÃÀ÷Iõ÷IÐõ÷IÛøIôÛøIÙøI ÙøI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúI7ûIà7ûIÙûI$ÙûI¯ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõIëöIèëöI‰DöId‰DöIÓKöIlÓKöI]UöIè]UöI“XöI´“XöI½möI@½möI[®öI¼[®öI¥µöIÄ¥µöI/¿öI@/¿öIeÂöI eÂöI�×öI˜�×öI-÷I-÷Iw÷I w÷I)÷I˜ )÷I7,÷Id 7,÷IyY÷Ið yY÷IŸ`÷IÌ Ÿ`÷Iÿ„÷IH ÿ„÷IA²÷IhA²÷IóÀ÷IDóÀ÷I¯õ÷IЯõ÷IËøIôËøIÉ øI É øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúI­ûIà­ûI}ØûI$}ØûI%ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI�öIè�öIEöIdEöIÓKöIlÓKöI]UöIè]UöI“XöI´“XöIMnöI@MnöIë®öI¼ë®öI¥µöIÄ¥µöI/¿öI@/¿öIeÂöI eÂöIØöI˜ØöI½÷I½÷Iw÷I w÷I)÷I˜ )÷I7,÷Id 7,÷IyY÷Ið yY÷IÏ`÷IÌ Ï`÷I/…÷IH /…÷Iq²÷Ihq²÷IÃÀ÷IDÃÀ÷Iõ÷IÐõ÷IÛøIôÛøIÙøI ÙøIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIÝûIàÝûI­ØûI$­ØûIUëûI(€} 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$<%<'±'±)&*&,›,›./1…1…3ú3ú5o6o8ä8ä:Y;Y=Î=Î?C@CB¸B¸D-E-G¢G¢IJLŽLŽNOQxQxSíSíUbVbX×X×ZL[L]Á]Á_6`6b«b«d e g•g•i j lln¼n¼p.q.s s uvx„x„z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIñöIÐñöI�DöIL�DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöIÃmöI(ÃmöIa®öI¤a®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöI•×öI€•×öI3÷Iü3÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷IŸ÷I,Ÿ÷IaøI˜aøI£øI$£øIñxùIèñxùI+eúIì+eúIsûIÌsûICÙûICÙûIëëûI€} 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·l{ïõIuöIÐuöIEöILEöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIGnöI(GnöIå®öI¤å®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöI·÷Iü·÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I a÷I´  a÷Ik…÷I0 k…÷I­²÷IP­²÷I¹¸÷I,¹¸÷I1øI˜1øI£øI$£øIëxùIèëxùI%eúIì%eúIIûIÌIûIÙûIÙûIÁëûI€} 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·l{ïõIñöIÐñöI�DöIL�DöIÙKöITÙKöIcUöIÐcUöI™XöIœ™XöI½möI(½möI[®öI¤[®öI«µöI¬«µöI5¿öI(5¿öIkÂöIôkÂöI�×öI€�×öI-÷Iü-÷I}÷I }÷I)÷I€ )÷I=,÷IL =,÷IY÷IØ Y÷I½`÷I´ ½`÷I…÷I0 …÷I_²÷IP_²÷Ie»÷I,e»÷IøI˜øI£øI$£øIÙxùIèÙxùIeúIìeúIûIÌûIÑØûIÑØûIyëûI€} 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$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbäbädWeWgÊgÊi=j=l°l°n!o!q’q’stvtvtxåxåzV{V}Ç}Ç8€8‚©‚©„…‡Œ‡Œ‰ý‰ý‹nŒnŽߎß�P‘P“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI{öIÐ{öIEöILEöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIGnöI(GnöIå®öI¤å®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöI·÷Iü·÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷IS»÷I,S»÷IÍøI˜ÍøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$i%i'Ý'Ý)Q*Q,Å,Å.9/91­1­3!4!6•6•8 9 ;};}=ñ=ñ?e@eBÙBÙDMEMGÁGÁI5J5L§L§NOQ‹Q‹SýSýUoVoXáXáZS[S]Å]Å_7`7b©b©deg�g�iÿiÿkqlqnãnãpUqUsÇsÇu9v9x«x«z{}�}�€‚s‚s„å„å†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI÷öIÐ÷öI•DöIL•DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÉmöI(ÉmöIg®öI¤g®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI›×öI€›×öI9÷Iü9÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷IŸ÷I,Ÿ÷IøI˜øI£øI$£øI÷xùIè÷xùI1eúIì1eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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$j%j'â'â)Z*Z,Ò,Ò.J/J1Â1Â3:4:6²6²8*9*;¢;¢=>@’@’B C E‚E‚GúGúIrJrLèLèN^O^QÔQÔSJTJVÀVÀX6Y6[¬[¬]"^"`˜`˜bce„e„gúgúipjplælæn\o\qÒqÒsHtHv¾v¾x4y4{ª{ª} ~ €–€–‚ ƒ …‚…‚‡ø‡ø‰nŠnŒäŒäŽZ�Z‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIoöIÐoöI EöIL EöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI;nöI(;nöIÙ®öI¤Ù®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI ØöI€ ØöI«÷Iü«÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷I­¸÷I,­¸÷IøI˜øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?M@MBÀBÀD3E3G¦G¦IJLŒLŒNÿNÿPrQrSåSåUXVXXËXËZ>[>]±]±_$`$b—b—d e g}g}iðiðkclcnÖnÖpIqIs¼s¼u/v/x¢x¢z{}ˆ}ˆûû�n‚n„á„á†T‡T‰ƉƋ8Œ8ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI÷öIÐ÷öI•DöIL•DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÃmöI(ÃmöIa®öI¤a®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI•×öI€•×öI3÷Iü3÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I_»÷I,_»÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnoqvqvsêsêu^v^xÒxÒzF{F}º}º.€.‚¢‚¢„…‡ЇЉþ‰þ‹rŒr޿޿�Z‘Z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIoöIÐoöI EöIL EöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIAnöI(AnöIß®öI¤ß®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIØöI€ØöI±÷Iü±÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷IG»÷I,G»÷I¯øI˜¯øI颸I$颸IÁxùIèÁxùIûdúIìûdúIkûIÌkûI;ØûI;ØûIãêûI€} 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$X•X•Z[]w]w_è_èaYbYdÊdÊf;g;i¬i¬klnŽnŽpÿpÿrpspuáuáwRxRzÃzÃ|4}4¥¥�‚„‡„‡†ø†øˆh‰h‹Ø‹Ø�HŽH�¸�¸’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIýöIÐýöI›DöIL›DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIÉmöI(ÉmöIg®öI¤g®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI›×öI€›×öI9÷Iü9÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷I¿¸÷I,¿¸÷I�øI˜�øI£øI$£øIñxùIèñxùI+eúIì+eúIOûIÌOûIÙûIÙûIÇëûI€} 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$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝITJTLÉLÉN>O>Q³Q³S(T(V�V�XY[‡[‡]ü]ü_q`qbæbæd[e[gÐgÐiEjElºlºn/o/q¤q¤stvŽvŽxy{x{x}í}íb€b‚ׂׄL…L‡Á‡Á‰6Š6Œ«Œ«Ž � ‘•‘•“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIiöIÐiöIEöILEöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI5nöI(5nöIÓ®öI¤Ó®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIØöI€ØöI¥÷Iü¥÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I³¸÷I,³¸÷IÁøI˜ÁøI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIéØûIéØûI‘ëûI€} 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$]%]'Ò'Ò)G*G,¼,¼.1/11¦1¦346�6�89;z;z=ï=ï?d@dBÙBÙDNENGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒyŒyŽîŽî�c‘c“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI÷öIÐ÷öI•DöIL•DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIÉmöI(ÉmöIg®öI¤g®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI›×öI€›×öI9÷Iü9÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷I_»÷I,_»÷I9øI˜9øI£øI$£øIÙxùIèÙxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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"×"×$L%L'Á'Á)6*6,«,«. / 1•1•3 4 668ô8ô:i;i=Þ=Þ?S@SBÈBÈD=E=G²G²I'J'LšLšN O Q€Q€SóSóUfVfXÙXÙZL[L]¿]¿_2`2b¥b¥deg‹g‹iþiþkqlqnänäpWqWsÊsÊu=v=x°x°z#{#}–}– € ‚|‚|„ï„ï†b‡b‰Õ‰Õ‹HŒH޻޻�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIiöIÐiöIEöILEöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI;nöI(;nöIÙ®öI¤Ù®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI ØöI€ ØöI«÷Iü«÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷IY»÷I,Y»÷IcøI˜cøI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIöIÐöI¡DöIL¡DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÏmöI(ÏmöIm®öI¤m®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI¡×öI€¡×öI?÷Iü?÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷I¹¸÷I,¹¸÷I—øI˜—øI£øI$£øIëxùIèëxùI%eúIì%eúI1ûIÌ1ûIÙûIÙûI©ëûI€} 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$`%`'×'×)N*N,Å,Å.@}@}BôBôDkEkGâGâI[J[LÒLÒNIOIQÀQÀS7T7V®V®X%Y%[œ[œ]^`Š`ŠbcexexgïgïifjflÝlÝnOoOqÁqÁs3t3v¥v¥xy{‰{‰}û}ûm€m‚߂߄Q…Q‡ÇÉ5Š5Œ§Œ§Ž�‘‹‘‹“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIýöIÐýöI›DöIL›DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÏmöI(ÏmöIm®öI¤m®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI¡×öI€¡×öI?÷Iü?÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I_»÷I,_»÷I9øI˜9øI£øI$£øIÙxùIèÙxùIeúIìeúI¹ûI̹ûI‰ØûI‰ØûI1ëûI€} 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·l‡ïõIcöIÐcöIEöILEöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöI5nöI(5nöIÓ®öI¤Ó®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIØöI€ØöI¥÷Iü¥÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷Ie»÷I,e»÷I#øI˜#øI £øI$ £øIåxùIèåxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I¹¸÷I,¹¸÷I'øI˜'øI+£øI$+£øIyùIèyùI=eúIì=eúI…ûIÌ…ûIUÙûIUÙûIýëûI€} 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·l�ïõIWöIÐWöIõDöILõDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI)nöI()nöIÇ®öI¤Ç®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIû×öI€û×öI™÷Iü™÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷I¹¸÷I,¹¸÷I)øI˜)øI%£øI$%£øIýxùIèýxùI7eúIì7eúI[ûIÌ[ûI+ÙûI+ÙûIÓëûI€} 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·l�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I“`÷I´ “`÷Ió„÷I0 ó„÷I5²÷IP5²÷Ie»÷I,e»÷IÏøI˜ÏøI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIãØûIãØûI‹ëûI€} 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·l�ïõI]öIÐ]öIûDöILûDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI/nöI(/nöIÍ®öI¤Í®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIØöI€ØöIŸ÷IüŸ÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷Iw»÷I,w»÷IãøI˜ãøI £øI$ £øIåxùIèåxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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·l�ïõI öIÐ öI§DöIL§DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÛmöI(ÛmöIy®öI¤y®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI­×öI€­×öIK÷IüK÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I¿¸÷I,¿¸÷IÛøI˜ÛøI1£øI$1£øI yùIè yùICeúIìCeúI—ûIÌ—ûIgÙûIgÙûIìûI€} 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·l�ïõIQöIÐQöIïDöILïDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI#nöI(#nöIÁ®öI¤Á®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIõ×öI€õ×öI“÷Iü“÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I¿¸÷I,¿¸÷IÝøI˜ÝøI£øI$£øIëxùIèëxùI%eúIì%eúIãûIÌãûI³ØûI³ØûI[ëûI€} 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·l�ïõI öIÐ öI§DöIL§DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÛmöI(ÛmöIy®öI¤y®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI­×öI€­×öIK÷IüK÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷Iq»÷I,q»÷I�øI˜�øI£øI$£øIñxùIèñxùI+eúIì+eúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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·l�ïõIWöIÐWöIõDöILõDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI#nöI(#nöIÁ®öI¤Á®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIõ×öI€õ×öI“÷Iü“÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I a÷I´  a÷Ik…÷I0 k…÷I­²÷IP­²÷Iƒ»÷I,ƒ»÷I£øI˜£øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIƒûĨûISØûISØûIûêûI€} 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$@%@'³'³)&*&,™,™. / 113ò3ò5e6e8Ø8Ø:K;K=¾=¾?/@/B B DEG‚G‚IóIóKdLdNÕNÕPFQFS·S·U(V(X™X™Z [ ]{]{_ì_ìa]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L™ïõI'öI�'öIÅDöIÅDöIIöI$IöIómöI�ómöI‘®öI‘®öIÕ²öI$Õ²öIÅ×öI�Å×öIc÷Ic÷I§÷I$§÷I[_÷I�[_÷I»ƒ÷Iü»ƒ÷I¶÷I ¶÷IKøI| KøI=£øIÜ =£øIyùI  yùIOeúI¤ OeúIsûI„ sûICÙûIÈCÙûIëëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIKöI�KöIéDöI éDöI�KöI�KöInöI€nöI»®öIü»®öIoµöIoµöIï×öIpï×öI�÷Iì�÷IA÷IôA÷Ia÷I`a÷Iq…÷IÌq…÷I¶÷Iì¶÷IKøIL KøI+£øI¬ +£øIyùIp yùI=eúIt =eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI!öI�!öI¿DöI¿DöIIöI$IöIómöI�ómöI‘®öI‘®öIÕ²öI$Õ²öIÅ×öI�Å×öIc÷Ic÷I§÷I$§÷I[_÷I�[_÷I»ƒ÷Iü»ƒ÷I¶÷I ¶÷IKøI| KøI+£øIÜ +£øIyùI  yùI=eúI¤ =eúI ûI„  ûIÝØûIÈÝØûI…ëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIQöI�QöIïDöI ïDöI�KöI�KöInöI€nöI»®öIü»®öIoµöIoµöIï×öIpï×öI�÷Iì�÷IA÷IôA÷Ia÷I`a÷Iw…÷IÌw…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI¹DöI ¹DöIIöIIöIímöI€ímöI‹®öIü‹®öIÕ²öIÕ²öI¿×öIp¿×öI]÷Iì]÷I§÷Iô§÷Ió`÷I`ó`÷IS…÷IÌS…÷I¶÷Iì¶÷IKøIL KøI1£øI¬ 1£øI yùIp  yùICeúIt CeúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIQöI�QöIïDöI ïDöI—KöI—KöI#nöI€#nöIÁ®öIüÁ®öIiµöIiµöIõ×öIpõ×öI“÷Iì“÷I;÷Iô;÷Ia÷I`a÷Ie…÷IÌe…÷I¶÷Iì¶÷IKøIL KøI1£øI¬ 1£øI yùIp  yùICeúIt CeúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI¹DöI ¹DöIIöIIöIçmöI€çmöI…®öIü…®öIÕ²öIÕ²öI¹×öIp¹×öIW÷IìW÷I§÷Iô§÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIWöI�WöIõDöI õDöI—KöI—KöI#nöI€#nöIÁ®öIüÁ®öIiµöIiµöIõ×öIpõ×öI“÷Iì“÷I;÷Iô;÷Ia÷I`a÷Iw…÷IÌw…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIWöI�WöIõDöI õDöIKöIKöI)nöI€)nöIÇ®öIüÇ®öIQµöIQµöIû×öIpû×öI™÷Iì™÷I#÷Iô#÷Iù`÷I`ù`÷IY…÷IÌY…÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI ³DöI IöI IöIámöI€ámöI®öIü®öIÛ²öIÛ²öI³×öIp³×öIQ÷IìQ÷I­÷Iô­÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIWöI�WöIõDöI õDöIKöIKöI)nöI€)nöIÇ®öIüÇ®öIQµöIQµöIû×öIpû×öI™÷Iì™÷I#÷Iô#÷Ia÷I`a÷Iq…÷IÌq…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI­DöI ­DöI÷HöI÷HöIámöI€ámöI®öIü®öIɲöIɲöI³×öIp³×öIQ÷IìQ÷I›÷Iô›÷Iá`÷I`á`÷IA…÷IÌA…÷Is¶÷Iìs¶÷I?øIL ?øI7£øI¬ 7£øIyùIp yùIIeúIt IeúI�ûIT �ûImÙûI˜mÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI]öI�]öIûDöI ûDöI—KöI—KöI/nöI€/nöIÍ®öIüÍ®öIiµöIiµöIØöIpØöIŸ÷IìŸ÷I;÷Iô;÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI öI� öI§DöI §DöI÷HöI÷HöIÛmöI€ÛmöIy®öIüy®öIɲöIɲöI­×öIp­×öIK÷IìK÷I›÷Iô›÷Iù`÷I`ù`÷IY…÷IÌY…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI]öI�]öIûDöI ûDöI—KöI—KöI/nöI€/nöIÍ®öIüÍ®öIiµöIiµöIØöIpØöIŸ÷IìŸ÷I;÷Iô;÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷Is¶÷Iìs¶÷I?øIL ?øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI}ûIT }ûIMØûI˜MØûIõêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI öI� öI§DöI §DöI IöI IöIÛmöI€ÛmöIy®öIüy®öIÛ²öIÛ²öI­×öIp­×öIK÷IìK÷I­÷Iô­÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIcöI�cöIEöI EöIyKöIyKöI/nöI€/nöIÍ®öIüÍ®öIKµöIKµöIØöIpØöIŸ÷IìŸ÷I÷Iô÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¡DöI ¡DöIIöIIöIÕmöI€ÕmöIs®öIüs®öIÕ²öIÕ²öI§×öIp§×öIE÷IìE÷I§÷Iô§÷Iù`÷I`ù`÷IY…÷IÌY…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIcöI�cöIEöI EöIyKöIyKöI5nöI€5nöIÓ®öIüÓ®öIKµöIKµöIØöIpØöI¥÷Iì¥÷I÷Iô÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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�ïõIöI�öI¡DöI ¡DöIñHöIñHöIÏmöI€ÏmöIm®öIüm®öIòöIòöI¡×öIp¡×öI?÷Iì?÷I•÷Iô•÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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�ïõIiöI�iöIEöI EöI‹KöI‹KöI5nöI€5nöIÓ®öIüÓ®öI]µöI]µöIØöIpØöI¥÷Iì¥÷I/÷Iô/÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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�ïõIýöI�ýöI›DöI ›DöIñHöIñHöIÏmöI€ÏmöIm®öIüm®öIòöIòöI¡×öIp¡×öI?÷Iì?÷I•÷Iô•÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI³ûIT ³ûIƒØûI˜ƒØûI+ëûIœ€} 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�ïõIiöI�iöIEöI EöI‹KöI‹KöI;nöI€;nöIÙ®öIüÙ®öI]µöI]µöI ØöIp ØöI«÷Iì«÷I/÷Iô/÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIýöI�ýöI›DöI ›DöIýHöIýHöIÉmöI€ÉmöIg®öIüg®öIϲöIϲöI›×öIp›×öI9÷Iì9÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIyûIT yûIIÙûI˜IÙûIñëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIiöI�iöIEöI EöImKöImKöI;nöI€;nöIÙ®öIüÙ®öI?µöI?µöI ØöIp ØöI«÷Iì«÷I÷Iô÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI÷öI�÷öI•DöI •DöIýHöIýHöIÉmöI€ÉmöIg®öIüg®öIϲöIϲöI›×öIp›×öI9÷Iì9÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIßxùIp ßxùIeúIt eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIoöI�oöI EöI  EöImKöImKöI;nöI€;nöIÙ®öIüÙ®öI?µöI?µöI ØöIp ØöI«÷Iì«÷I÷Iô÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIñöI�ñöI�DöI �DöIëHöIëHöIÃmöI€ÃmöIa®öIüa®öI½²öI½²öI•×öIp•×öI3÷Iì3÷I�÷Iô�÷Iu`÷I`u`÷IÕ„÷IÔÕ„÷Ig¶÷Iôg¶÷I3øIT 3øI%£øI´ %£øIMwùIx MwùI+GúI| +GúI‹ûI\ ‹ûI�×ûI �×ûI)êûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIoöI�oöI EöI  EöIKöIKöIAnöI€AnöIß®öIüß®öIQµöIQµöIØöIpØöI±÷Iì±÷I#÷Iô#÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI/wùIp /wùI GúIt  GúI×ûIT ×ûIÍÖûI˜ÍÖûIuéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIuöI�uöIEöI EöI…KöI…KöIAnöI€AnöIß®öIüß®öIWµöIWµöIØöIpØöI±÷Iì±÷I)÷Iô)÷I_÷I`_÷Iyƒ÷IÌyƒ÷Ig¶÷Iìg¶÷I3øIL 3øIï¢øI¬ ï¢øIwùIp wùIõFúIt õFúIqûIT qûIgÖûI˜gÖûIéûIœ€} 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„uïõI¯öI¯öIMDöI|MDöI±IöI„±IöIçLöIðçLöIqVöI|qVöI§YöIH§YöI{möIÔ{möI®öIP®öIƒ³öIXƒ³öI¹¶öIŶöICÀöIPCÀöIyÃöIyÃöIM×öI¨M×öIë÷I$ ë÷IU÷I, U÷I‹ ÷I˜ ‹ ÷I*÷I$ *÷IK-÷Ið K-÷I�Z÷I| �Z÷IW`÷IXW`÷I·„÷IÔ·„÷Iù±÷Iôù±÷IÃÀ÷IÐÃÀ÷Iõ÷I\õ÷IÿøI€ÿøIýøI¬ýøI£øI8£øIñxùIüñxùI+eúI+eúI¥rúIà¥rúIOûIlOûIÙûI°ÙûIÇëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI‡öIè‡öI%EöId%EöIÓKöIlÓKöI]UöIè]UöI“XöI´“XöIYnöI@YnöI÷®öI¼÷®öI¥µöIÄ¥µöI/¿öI@/¿öIeÂöI eÂöI+ØöI˜+ØöIÉ÷IÉ÷Iw÷I w÷I)÷I˜ )÷I7,÷Id 7,÷IyY÷Ið yY÷Iá`÷IÌ á`÷IA…÷IH A…÷Iƒ²÷Ihƒ²÷IÃÀ÷IDÃÀ÷Iõ÷IÐõ÷IÿøIôÿøIýøI ýøI£øI¬£øI÷xùIp÷xùI1eúIt1eúI«rúIT«rúIyûIàyûIIÙûI$IÙûIñëûI(€} 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„uïõI©öI©öIGDöI|GDöI±IöI„±IöIçLöIðçLöIqVöI|qVöI§YöIH§YöI{möIÔ{möI®öIP®öIƒ³öIXƒ³öI¹¶öIŶöICÀöIPCÀöIyÃöIyÃöIM×öI¨M×öIë÷I$ ë÷IU÷I, U÷I‹ ÷I˜ ‹ ÷I*÷I$ *÷IK-÷Ið K-÷I�Z÷I| �Z÷I{`÷IX{`÷IÛ„÷IÔÛ„÷I²÷Iô²÷IÃÀ÷IÐÃÀ÷Iõ÷I\õ÷IÿøI€ÿøIýøI¬ýøI£øI8£øIÙxùIüÙxùIeúIeúI�rúIà�rúI×ûIl×ûI§ØûI°§ØûIOëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI�öIè�öI+EöId+EöIÓKöIlÓKöI]UöIè]UöI“XöI´“XöIYnöI@YnöI÷®öI¼÷®öI¥µöIÄ¥µöI/¿öI@/¿öIeÂöI eÂöI+ØöI˜+ØöIÉ÷IÉ÷Iw÷I w÷I)÷I˜ )÷I7,÷Id 7,÷IyY÷Ið yY÷I±`÷IÌ ±`÷I…÷IH …÷IS²÷IhS²÷IùÀ÷IDùÀ÷Iµõ÷Iеõ÷IøIôøI!øI !øI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúIûIàûIרûI$רûIëûI(€} 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$$%$'›'›)*,‰,‰./1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n1o1q¢q¢stv„v„xõxõzf{f}×}×H€H‚¹‚¹„*…*‡œ‡œ‰ Š Œ~Œ~ŽïŽï�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx{ïõIñöIèñöI�DöId�DöIÙKöIlÙKöIcUöIècUöI™XöI´™XöIÃmöI@ÃmöIa®öI¼a®öI«µöIÄ«µöI5¿öI@5¿öIkÂöI kÂöI•×öI˜•×öI3÷I3÷I}÷I }÷I)÷I˜ )÷I=,÷Id =,÷IY÷Ið Y÷I·`÷IÌ ·`÷I…÷IH …÷IY²÷IhY²÷IAÁ÷IDAÁ÷Iýõ÷IÐýõ÷I]øIô]øI["øI ["øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIËûIàËûI›ØûI$›ØûICëûI(€} 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$B%B'·'·),*,,¡,¡./1‹1‹346u6u8ê8ê:_;_=Ô=Ô?G@GBºBºD-E-G G IJL†L†NùNùPlQlSßSßUVX‹X‹ZþZþ\q]q_ä_äa›b›deg�g�iôiôkglgnÚnÚpMqMsÀsÀu3v3x¦x¦z{}Œ}Œÿÿ�r‚r„å„å†X‡X‰ʉʋ<Œ<ޮޮ� ‘ “  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI�öIè�öIEöIdEöIÙKöIlÙKöIcUöIècUöI™XöI´™XöISnöI@SnöIñ®öI¼ñ®öI«µöIÄ«µöI5¿öI@5¿öIkÂöI kÂöI%ØöI˜%ØöIÃ÷IÃ÷I}÷I }÷I)÷I˜ )÷I=,÷Id =,÷IY÷Ið Y÷Ia÷IÌ a÷Ie…÷IH e…÷I§²÷Ih§²÷IÉÀ÷IDÉÀ÷I…õ÷IÐ…õ÷IøIôøI øI  øI£øI¬£øIñxùIpñxùI+eúIt+eúI¥rúIT¥rúIûIàûIOÙûI$OÙûI÷ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõIñöIèñöI�DöId�DöIÙKöIlÙKöIcUöIècUöI™XöI´™XöI½möI@½möI[®öI¼[®öI«µöIÄ«µöI5¿öI@5¿öIkÂöI kÂöI�×öI˜�×öI-÷I-÷I}÷I }÷I)÷I˜ )÷I=,÷Id =,÷IY÷Ið Y÷I�`÷IÌ �`÷Ií„÷IH í„÷I/²÷Ih/²÷IÁ÷IDÁ÷IÙõ÷IÐÙõ÷I©øIô©øI§!øI §!øI颸I¬颸IÁxùIpÁxùIûdúItûdúIurúITurúIkûIàkûI;ØûI$;ØûIãêûI(€} 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$8%8'«'«)*,‘,‘./1w1w3ê3ê5]6]8Ð8Ð:C;C=¶=¶?'@'B˜B˜D E GzGzIëIëK\L\NÍNÍP>Q>S¯S¯UåUåWVXVZÇZÇ\8]8_©_©a]b]dÎdÎf?g?i°i°k!l!n’n’pqststuåuåwVxVzÇzÇ|8}8©©�‚„‹„‹†ü†üˆl‰l‹Ü‹Ü�LŽL�¼�¼’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI‡öIè‡öI%EöId%EöIÙKöIlÙKöIcUöIècUöI™XöI´™XöISnöI@SnöIñ®öI¼ñ®öI«µöIÄ«µöI5¿öI@5¿öIkÂöI kÂöI%ØöI˜%ØöIÃ÷IÃ÷I}÷I }÷I)÷I˜ )÷I=,÷Id =,÷IY÷Ið Y÷IÛ`÷IÌ Û`÷I;…÷IH ;…÷I}²÷Ih}²÷IÉÀ÷IDÉÀ÷I…õ÷IÐ…õ÷IøIôøI øI  øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúI ûIà ûIÝØûI$ÝØûI…ëûI(€} 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$L%L'Â'Â)8*8,®,®.$/$1š1š346†6†8ü8ü:r;r=è=è?^@^BÔBÔDJEJGÀGÀI6J6LªLªNOQ’Q’STVzVzXîXîZb[b]Ö]Ö_J`Jb¾b¾d2e2g¦g¦ijlŽlŽnËnËp?q?s³s³u'v'x›x›zJ{J}¾}¾2€2‚¦‚¦„…‡އމŠŒvŒvŽêŽê�^‘^“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI÷öIÐ÷öI•DöIL•DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÉmöI(ÉmöIg®öI¤g®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI›×öI€›×öI9÷Iü9÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷Iÿ½÷I,ÿ½÷IŸøI˜ŸøI£øI$£øIÙxùIèÙxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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·l{ïõI{öIÐ{öIEöILEöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIGnöI(GnöIå®öI¤å®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöIØöI€ØöI·÷Iü·÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷Ia÷I´ a÷Iq…÷I0 q…÷I³²÷IP³²÷Iç½÷I,ç½÷I?øI˜?øI£øI$£øIëxùIèëxùI%eúIì%eúIIûIÌIûIÙûIÙûIÁëûI€} 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·l{ïõI÷öIÐ÷öI•DöIL•DöIßKöITßKöIiUöIÐiUöIŸXöIœŸXöIÃmöI(ÃmöIa®öI¤a®öI±µöI¬±µöI;¿öI(;¿öIqÂöIôqÂöI•×öI€•×öI3÷Iü3÷Iƒ÷I ƒ÷I )÷I€  )÷IC,÷IL C,÷I…Y÷IØ …Y÷I·`÷I´ ·`÷I…÷I0 …÷IY²÷IPY²÷IŸÀ÷I,ŸÀ÷I;øI˜;øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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·l{ïõI{öIÐ{öIEöILEöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIMnöI(MnöIë®öI¤ë®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIØöI€ØöI½÷Iü½÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷I{À÷I,{À÷IÛøI˜ÛøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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�ïõIýöIÐýöI›DöIL›DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÉmöI(ÉmöIg®öI¤g®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI›×öI€›×öI9÷Iü9÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷Iÿ½÷I,ÿ½÷I½øI˜½øI £øI$ £øIåxùIèåxùIeúIìeúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜIUJULÌLÌNCOCQºQºS1T1V¨V¨XY[–[–] ^ `„`„bûbûdrergégéi`j`l×l×nIoIq»q»s-t-vŸvŸxy{ƒ{ƒ}õ}õg€g‚قلK…K‡½‡½‰/Š/Œ¡Œ¡Ž�‘…‘…“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIuöIÐuöIEöILEöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIAnöI(AnöIß®öI¤ß®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIØöI€ØöI±÷Iü±÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷Ió½÷I,ó½÷I«øI˜«øI£øI$£øIëxùIèëxùI%eúIì%eúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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�ïõI÷öIÐ÷öI•DöIL•DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÉmöI(ÉmöIg®öI¤g®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI›×öI€›×öI9÷Iü9÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷IÛ`÷I´ Û`÷I;…÷I0 ;…÷I}²÷IP}²÷IŸÀ÷I,ŸÀ÷IYøI˜YøI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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�ïõIuöIÐuöIEöILEöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIGnöI(GnöIå®öI¤å®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIØöI€ØöI·÷Iü·÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷I�À÷I,�À÷IMøI˜MøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúI³ûI̳ûIƒØûIƒØûI+ëûI€} 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‡ïõIöIÐöI¡DöIL¡DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIÏmöI(ÏmöIm®öI¤m®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI¡×öI€¡×öI?÷Iü?÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷Iÿ½÷I,ÿ½÷IÛøI˜ÛøI£øI$£øIñxùIèñxùI+eúIì+eúIOûIÌOûIÙûIÙûIÇëûI€} 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‡ïõIiöIÐiöIEöILEöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI;nöI(;nöIÙ®öI¤Ù®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI ØöI€ ØöI«÷Iü«÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I ¾÷I, ¾÷IwøI˜wøI%£øI$%£øIýxùIèýxùI7eúIì7eúIûIÌûIOÙûIOÙûI÷ëûI€} 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‡ïõIýöIÐýöI›DöIL›DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIÏmöI(ÏmöIm®öI¤m®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI¡×öI€¡×öI?÷Iü?÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Iç`÷I´ ç`÷IG…÷I0 G…÷I‰²÷IP‰²÷IŸÀ÷I,ŸÀ÷I}øI˜}øI£øI$£øIÙxùIèÙxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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‡ïõIoöIÐoöI EöIL EöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIAnöI(AnöIß®öI¤ß®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIØöI€ØöI±÷Iü±÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷I±À÷I,±À÷IøI˜øI £øI$ £øIåxùIèåxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$_%_'Ö'Ö)M*M,Ä,Ä.;/;1²1²3)4)6 6 89;Ž;Ž=>@|@|BóBóDjEjGáGáIXJXLÍLÍNBOBQ·Q·S,T,V¡V¡XY[‹[‹]^`u`ubèbèd[e[gÎgÎiAjAl´l´n%o%q–q–stvxvxxéxézZ{Z}Ë}Ë<€<‚­‚­„…‡�‡�‰ŠŒrŒrŽãŽã�T‘T“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷Iÿ½÷I,ÿ½÷IÛøI˜ÛøI£øI$£øIßxùIèßxùIeúIìeúI×ûIÌ×ûI§ØûI§ØûIOëûI€} 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$F%F'»'»)0*0,¥,¥./1�1�346y6y8î8î:c;c=Ø=Ø?K@KB¾B¾D1E1G¤G¤IJLŠLŠNýNýPpQpSãSãUVVVXÉXÉZ<[<]¯]¯_"`"b•b•deg{g{iîiîkalanÔnÔpGqGsºsºu-v-x x z{}†}†ùù�l‚l„߄߆R‡R‰ĉċ6Œ6ލލ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIcöIÐcöIEöILEöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI5nöI(5nöIÓ®öI¤Ó®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIØöI€ØöI¥÷Iü¥÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷I#¾÷I,#¾÷ICøI˜CøI+£øI$+£øIyùIèyùI=eúIì=eúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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·l‡ïõIiöIÐiöIEöILEöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI;nöI(;nöIÙ®öI¤Ù®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI ØöI€ ØöI«÷Iü«÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I]`÷I´ ]`÷I½„÷I0 ½„÷Iÿ±÷IPÿ±÷IÛÀ÷I,ÛÀ÷I÷øI˜÷øI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIéØûIéØûI‘ëûI€} 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·l�ïõI öIÐ öI§DöIL§DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Ic`÷I´ c`÷IÄ÷I0 Ä÷I²÷IP²÷I¾÷I,¾÷I}øI˜}øI£øI$£øIñxùIèñxùI+eúIì+eúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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·l�ïõI]öIÐ]öIûDöILûDöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI/nöI(/nöIÍ®öI¤Í®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIØöI€ØöIŸ÷IüŸ÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Ii`÷I´ i`÷IÉ„÷I0 É„÷I ²÷IP ²÷IA¾÷I,A¾÷I øI˜ øI%£øI$%£øIýxùIèýxùI7eúIì7eúI[ûIÌ[ûI+ÙûI+ÙûIÓëûI€} 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·l�ïõI öIÐ öI§DöIL§DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷I·À÷I,·À÷I%øI˜%øI £øI$ £øIåxùIèåxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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$L—L—N O Q}Q}SðSðUcVcXÖXÖZI[I]¼]¼_/`/b¡b¡deg…g…i÷i÷kilinÛnÛpMqMs¿s¿u1v1x£x£z{}‡}‡ùù�k‚k„݄݆Q‡Q‰ĉċ7Œ7ŽªŽª�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIcöIÐcöIEöILEöIëKöITëKöIuUöIÐuUöI«XöIœ«XöI/nöI(/nöIÍ®öI¤Í®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIØöI€ØöIŸ÷IüŸ÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IK`÷I´ K`÷I«„÷I0 «„÷Ií±÷IPí±÷IÿÀ÷I,ÿÀ÷IÉøI˜ÉøI£øI$£øIëxùIèëxùI%eúIì%eúIõûIÌõûIÅØûIÅØûImëûI€} 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$P%P'Å'Å):*:,¯,¯.$/$1™1™346ƒ6ƒ8ø8ø:m;m=â=â?W@WBÌBÌDAEAG¶G¶I+J+LžLžNOQ„Q„S÷S÷UjVjXÝXÝZP[P]Ã]Ã_6`6b©b©deg�g�ijlulunènèp[q[sÎsÎuAvAx´x´z'{'}š}š € ‚€‚€„ó„ó†f‡f‰ىًLŒLŽ¿Ž¿�2‘2“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l“ïõIöIÐöI­DöIL­DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIámöI(ámöI®öI¤®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI³×öI€³×öIQ÷IüQ÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷IW`÷I´ W`÷I·„÷I0 ·„÷Iù±÷IPù±÷I)¾÷I,)¾÷ICøI˜CøI%£øI$%£øIýxùIèýxùI7eúIì7eúIOûIÌOûIÙûIÙûIÇëûI€} 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$^%^'Õ'Õ)L*L,Ã,Ã.:/:1±1±3(4(6Ÿ6Ÿ89;�;�=>@{@{BòBòDiEiGàGàIYJYLÐLÐNGOGQ¾Q¾S5T5V¬V¬X#Y#[š[š]^`ˆ`ˆbÿbÿdvevgígíidjdlÛlÛnMoMq¿q¿s1t1v£v£xy{‡{‡}ù}ùk€k‚݂݄O…O‡Á‡Á‰3Š3Œ¥Œ¥Ž�‘‰‘‰“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l“ïõIWöIÐWöIõDöILõDöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI)nöI()nöIÇ®öI¤Ç®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIû×öI€û×öI™÷Iü™÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷IS¾÷I,S¾÷I±øI˜±øI%£øI$%£øIýxùIèýxùI7eúIì7eúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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 $d €™} bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations10ll æ æ\\ÒÒ)H*H4¾4¾>2?2I¦I¦ST^Ž^Žhisvsv} b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k80_c768_r7_r85:5z9299Z=âyz}*}z�"�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 80, output_columns: 768 })outputElements6144input0Elements576input1Elements52488invocations1 b²)gÉoperation.6ipu_stack_attention_softmax_large_query_small_key_f16-:-z1211Z5âaze*ezi*ibmreasonOperatorKernelvalue10kernelSpecPlanned(AttentionSoftmax { head_dimension: 72, key_columns: 729, padded_key_columns: 768 })outputElements6272input0Elements6144invocations1 €é“ bJ7MÇoperation.6exchange :z2breasonOperatorInputsvalue8invocations1ðllææ``ÚÚ T T Î ÎHHÂÂ<<¶¶,,¢¢ Ž Ž"#%z%z'ð'ð)f*f,Ü,Ü.R/R1È1È3>4>6´6´8*9*; ; =>@Š@ŠBþBþDrErGæGæIZJZLÎLÎNBOBQ¶Q¶S*T*VžVžXY[†[†]ú]ú_n`nbâbâdVeVgÊgÊi>j>l²l²n&o&qšqšstv‚v‚xöxözj{j}Þ}ÞR€R‚ƂƄ:…:‡®‡®‰!Š!Œ”Œ”Ž�‘z‘z“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l“ïõIöIÐöI­DöIL­DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷IáÀ÷I,áÀ÷IýøI˜ýøI£øI$£øIëxùIèëxùI%eúIì%eúIõûIÌõûIÅØûIÅØûImëûI€} 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$g™g™ijlƒlƒnõnõpgqgsÙsÙuKvKx½x½z/{/}¡}¡€‚…‚…„÷„÷†i‡i‰Û‰Û‹MŒMŽ¿Ž¿�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l“ïõI]öIÐ]öIûDöILûDöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI)nöI()nöIÇ®öI¤Ç®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIû×öI€û×öI™÷Iü™÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷IÁ÷I,Á÷IwøI˜wøI £øI$ £øIåxùIèåxùIeúIìeúIÅûIÌÅûI•ØûI•ØûI=ëûI€} 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$Z%Z'Î'Î)B*B,¶,¶.*/*1ž1ž346†6†8ú8ú:n;n=â=â?T@TBÆBÆD8E8GªGªIJLŽLŽNOQrQrSäSäUVVVXÈXÈZ:[:]¬]¬_`b�b�degtgtiæiækXlXnÊnÊp‡>‰¯‰¯‹ Œ ޑޑ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L™ïõI'öI�'öIÅDöIÅDöI IöI$ IöIùmöI�ùmöI—®öI—®öIÛ²öI$Û²öIË×öI�Ë×öIi÷Ii÷I­÷I$­÷I[_÷I�[_÷I»ƒ÷Iü»ƒ÷I…¶÷I …¶÷IQøI| QøI=£øIÜ =£øIyùI  yùIOeúI¤ OeúIsûI„ sûICÙûIÈCÙûIëëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI'öI�'öIÅDöIÅDöI IöI$ IöIómöI�ómöI‘®öI‘®öIÛ²öI$Û²öIÅ×öI�Å×öIc÷Ic÷I­÷I$­÷I[_÷I�[_÷I»ƒ÷Iü»ƒ÷I…¶÷I …¶÷IQøI| QøI%£øIÜ %£øIýxùI  ýxùI7eúI¤ 7eúIûûI„ ûûIËØûIÈËØûIsëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIWöI�WöIõDöI õDöI�KöI�KöI#nöI€#nöIÁ®öIüÁ®öIoµöIoµöIõ×öIpõ×öI“÷Iì“÷IA÷IôA÷Ia÷I`a÷Iw…÷IÌw…÷I…¶÷Iì…¶÷IQøIL QøI+£øI¬ +£øIyùIp yùI=eúIt =eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI!öI�!öI¿DöI ¿DöIIöIIöIómöI€ómöI‘®öIü‘®öIÕ²öIÕ²öIÅ×öIpÅ×öIc÷Iìc÷I§÷Iô§÷Iù`÷I`ù`÷IY…÷IÌY…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI!öI�!öI¿DöI ¿DöIIöIIöIímöI€ímöI‹®öIü‹®öIÕ²öIÕ²öI¿×öIp¿×öI]÷Iì]÷I§÷Iô§÷Iù`÷I`ù`÷IY…÷IÌY…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øIßxùIp ßxùIeúIt eúI�ûIT �ûI_ØûI˜_ØûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIWöI�WöIõDöI õDöI—KöI—KöI)nöI€)nöIÇ®öIüÇ®öIiµöIiµöIû×öIpû×öI™÷Iì™÷I;÷Iô;÷Ia÷I`a÷Iq…÷IÌq…÷I¶÷Iì¶÷IKøIL KøI+£øI¬ +£øIyùIp yùI=eúIt =eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI¹DöI ¹DöI IöI IöIímöI€ímöI‹®öIü‹®öIÛ²öIÛ²öI¿×öIp¿×öI]÷Iì]÷I­÷Iô­÷Ió`÷I`ó`÷IS…÷IÌS…÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI]öI�]öIûDöI ûDöIKöIKöI/nöI€/nöIÍ®öIüÍ®öIQµöIQµöIØöIpØöIŸ÷IìŸ÷I#÷Iô#÷Iù`÷I`ù`÷IY…÷IÌY…÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI³DöI ³DöI IöI IöIçmöI€çmöI…®öIü…®öIÛ²öIÛ²öI¹×öIp¹×öIW÷IìW÷I­÷Iô­÷Ia÷I`a÷Ie…÷IÌe…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI]öI�]öIûDöI ûDöIKöIKöI/nöI€/nöIÍ®öIüÍ®öIQµöIQµöIØöIpØöIŸ÷IìŸ÷I#÷Iô#÷Ia÷I`a÷Iq…÷IÌq…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI ³DöIýHöIýHöIçmöI€çmöI…®öIü…®öIϲöIϲöI¹×öIp¹×öIW÷IìW÷I¡÷Iô¡÷Iç`÷I`ç`÷IG…÷IÌG…÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI­DöI ­DöIýHöIýHöIámöI€ámöI®öIü®öIϲöIϲöI³×öIp³×öIQ÷IìQ÷I¡÷Iô¡÷Ia÷I`a÷Ie…÷IÌe…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIcöI�cöIEöI EöI�KöI�KöI5nöI€5nöIÓ®öIüÓ®öIoµöIoµöIØöIpØöI¥÷Iì¥÷IA÷IôA÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIßxùIp ßxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI­DöI ­DöI IöI IöIÛmöI€ÛmöIy®öIüy®öIÛ²öIÛ²öI­×öIp­×öIK÷IìK÷I­÷Iô­÷Iá`÷I`á`÷IA…÷IÌA…÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIiöI�iöIEöI EöIyKöIyKöI5nöI€5nöIÓ®öIüÓ®öIKµöIKµöIØöIpØöI¥÷Iì¥÷I÷Iô÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷Is¶÷Iìs¶÷I?øIL ?øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIOÙûI˜OÙûI÷ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI öI� öI§DöI §DöI IöI IöIÛmöI€ÛmöIy®öIüy®öIÛ²öIÛ²öI­×öIp­×öIK÷IìK÷I­÷Iô­÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIiöI�iöIEöI EöIyKöIyKöI;nöI€;nöIÙ®öIüÙ®öIKµöIKµöI ØöIp ØöI«÷Iì«÷I÷Iô÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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‡ïõI öI� öI§DöI §DöIñHöIñHöIÕmöI€ÕmöIs®öIüs®öIòöIòöI§×öIp§×öIE÷IìE÷I•÷Iô•÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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‡ïõIiöI�iöIEöI EöI‘KöI‘KöI;nöI€;nöIÙ®öIüÙ®öIcµöIcµöI ØöIp ØöI«÷Iì«÷I5÷Iô5÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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‡ïõIöI�öI¡DöI ¡DöIñHöIñHöIÕmöI€ÕmöIs®öIüs®öIòöIòöI§×öIp§×öIE÷IìE÷I•÷Iô•÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIqûIT qûIAØûI˜AØûIéêûIœ€} 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�ïõIýöI�ýöI›DöI ›DöIýHöIýHöIÏmöI€ÏmöIm®öIüm®öIϲöIϲöI¡×öIp¡×öI?÷Iì?÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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�ïõIoöI�oöI EöI  EöImKöImKöIAnöI€AnöIß®öIüß®öI?µöI?µöIØöIpØöI±÷Iì±÷I÷Iô÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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�ïõIýöI�ýöI›DöI ›DöIýHöIýHöIÉmöI€ÉmöIg®öIüg®öIϲöIϲöI›×öIp›×öI9÷Iì9÷I¡÷Iô¡÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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�ïõIuöI�uöIEöI EöImKöImKöIAnöI€AnöIß®öIüß®öI?µöI?µöIØöIpØöI±÷Iì±÷I÷Iô÷IC_÷I`C_÷I£ƒ÷IÌ£ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIuöI�uöIEöI EöI…KöI…KöIGnöI€GnöIå®öIüå®öIWµöIWµöIØöIpØöI·÷Iì·÷I)÷Iô)÷I_÷I`_÷Iƒ÷Ĩ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIAwùIp AwùIGúIt GúI1ûIT 1ûI'×ûI˜'×ûIÏéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõI÷öI�÷öI•DöI •DöIëHöIëHöIÃmöI€ÃmöIa®öIüa®öI½²öI½²öI•×öIp•×öI3÷Iì3÷I�÷Iô�÷I™`÷I`™`÷Iù„÷IÔù„÷Ig¶÷Iôg¶÷I3øIT 3øI£øI´ £øI/wùIx /wùI GúI|  GúIéûI\ éûIßÖûI ßÖûI‡éûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõIuöI�uöIEöI EöI…KöI…KöIGnöI€GnöIå®öIüå®öIWµöIWµöIØöIpØöI·÷Iì·÷I)÷Iô)÷I_÷I`_÷Iƒ÷Ĩ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI)wùIp )wùIGúIt GúI¹ûIT ¹ûI¯ÖûI˜¯ÖûIWéûIœ€} 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$Y%Y'Í'Í)A*A,µ,µ.)/)1�1�346…6…8ù8ù:m;m=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b™b™d e g}g}iïiïkalanÓnÓpEqEs·s·u)v)x›x›z { }}ññ�c‚c„Õ„Õ†E‡E‰¶‰¶‹'Œ'Ž˜Ž˜� ‘ “ b’)wíoperation.6ipu_stack_gemm_f16_init_small_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements560input0Elements5376input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_small_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements504input0Elements560input1Elements5488invocations1„uïõI¯öI¯öIMDöI|MDöI±IöI„±IöIçLöIðçLöIqVöI|qVöI§YöIH§YöI�möIÔ�möI®öIP®öIƒ³öIXƒ³öI¹¶öIŶöICÀöIPCÀöIyÃöIyÃöIS×öI¨S×öIñ÷I$ ñ÷IU÷I, U÷I‹ ÷I˜ ‹ ÷I*÷I$ *÷IK-÷Ið K-÷I�Z÷I| �Z÷I“`÷IX“`÷Ió„÷IÔó„÷I5²÷Iô5²÷IÏÀ÷IÐÏÀ÷I‹õ÷I\‹õ÷I)øI€)øI' øI¬' øI%£øI8%£øIýxùIüýxùI7eúI7eúI±rúIà±rúI‹ûIl‹ûI[ÙûI°[ÙûIìûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI�öIè�öI+EöId+EöIÙKöIlÙKöIcUöIècUöI™XöI´™XöI_nöI@_nöIý®öI¼ý®öI«µöIÄ«µöI5¿öI@5¿öIkÂöI kÂöI1ØöI˜1ØöIÏ÷IÏ÷I}÷I }÷I)÷I˜ )÷I=,÷Id =,÷IY÷Ið Y÷Ií`÷IÌ í`÷IM…÷IH M…÷I�²÷Ih�²÷IÏÀ÷IDÏÀ÷I‹õ÷IЋõ÷I)øIô)øI' øI ' øI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúI=ûIà=ûI ÙûI$ ÙûIµëûI(€} 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„uïõI¯öI¯öIMDöI|MDöI±IöI„±IöIçLöIðçLöIqVöI|qVöI§YöIH§YöI{möIÔ{möI®öIP®öIƒ³öIXƒ³öI¹¶öIŶöICÀöIPCÀöIyÃöIyÃöIM×öI¨M×öIë÷I$ ë÷IU÷I, U÷I‹ ÷I˜ ‹ ÷I*÷I$ *÷IK-÷Ið K-÷I�Z÷I| �Z÷I]`÷IX]`÷I½„÷IÔ½„÷Iÿ±÷Iôÿ±÷IÏÀ÷IÐÏÀ÷I‹õ÷I\‹õ÷I)øI€)øI' øI¬' øI £øI8 £øIåxùIüåxùIeúIeúI™rúIà™rúIûIlûIãØûI°ãØûI‹ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI“öIè“öI1EöId1EöIÙKöIlÙKöIcUöIècUöI™XöI´™XöI_nöI@_nöIý®öI¼ý®öI«µöIÄ«µöI5¿öI@5¿öIkÂöI kÂöI1ØöI˜1ØöIÏ÷IÏ÷I}÷I }÷I)÷I˜ )÷I=,÷Id =,÷IY÷Ið Y÷I½`÷IÌ ½`÷I…÷IH …÷I_²÷Ih_²÷IÏÀ÷IDÏÀ÷I‹õ÷IЋõ÷I5øIô5øI3 øI 3 øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIéûIàéûI¹ØûI$¹ØûIaëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI÷öIè÷öI•DöId•DöIßKöIlßKöIiUöIèiUöIŸXöI´ŸXöIÉmöI@ÉmöIg®öI¼g®öI±µöIıµöI;¿öI@;¿öIqÂöI qÂöI›×öI˜›×öI9÷I9÷Iƒ÷I ƒ÷I )÷I˜  )÷IC,÷Id C,÷I…Y÷Ið …Y÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷IÕÀ÷IDÕÀ÷I‘õ÷IБõ÷I/øIô/øI- øI - øI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúIIûIàIûIÙûI$ÙûIÁëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI‡öIè‡öI%EöId%EöIßKöIlßKöIiUöIèiUöIŸXöI´ŸXöISnöI@SnöIñ®öI¼ñ®öI±µöIıµöI;¿öI@;¿öIqÂöI qÂöI%ØöI˜%ØöIÃ÷IÃ÷Iƒ÷I ƒ÷I )÷I˜  )÷IC,÷Id C,÷I…Y÷Ið …Y÷Ia÷IÌ a÷Iq…÷IH q…÷I³²÷Ih³²÷IÕÀ÷IDÕÀ÷I‘õ÷IБõ÷I/øIô/øI- øI - øI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúIIûIàIûIÙûI$ÙûIÁëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI÷öIè÷öI•DöId•DöIßKöIlßKöIiUöIèiUöIŸXöI´ŸXöIÃmöI@ÃmöIa®öI¼a®öI±µöIıµöI;¿öI@;¿öIqÂöI qÂöI•×öI˜•×öI3÷I3÷Iƒ÷I ƒ÷I )÷I˜  )÷IC,÷Id C,÷I…Y÷Ið …Y÷Ic`÷IÌ c`÷IÄ÷IH Ä÷I²÷Ih²÷IÕÀ÷IDÕÀ÷I‘õ÷IБõ÷I/øIô/øI- øI - øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúIãûIàãûI³ØûI$³ØûI[ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI‡öIè‡öI%EöId%EöIßKöIlßKöIiUöIèiUöIŸXöI´ŸXöIYnöI@YnöI÷®öI¼÷®öI±µöIıµöI;¿öI@;¿öIqÂöI qÂöI+ØöI˜+ØöIÉ÷IÉ÷Iƒ÷I ƒ÷I )÷I˜  )÷IC,÷Id C,÷I…Y÷Ið …Y÷Iç`÷IÌ ç`÷IG…÷IH G…÷I‰²÷Ih‰²÷IÕÀ÷IDÕÀ÷I‘õ÷IБõ÷I/øIô/øI- øI - øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIÑûIàÑûI¡ØûI$¡ØûIIëûI(€} 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$h%h'Ü'Ü)P*P,Ä,Ä.8/81¬1¬3 4 6”6”89;|;|=ð=ð?d@dBØBØDLELGÀGÀI4J4L¦L¦NOQŠQŠSüSüUnVnXàXàZR[R]Ä]Ä_6`6b¨b¨degŒgŒiþiþkplpn©n©pqs�s�uÿuÿwqxqz { }’}’€‚v‚v„è„è†Z‡Z‰̉̋>Œ>ްް�"‘"“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõIýöIÐýöI›DöIL›DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÉmöI(ÉmöIg®öI¤g®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI›×öI€›×öI9÷Iü9÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IŸ÷I,Ÿ÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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$q%q'ç'ç)]*],Ó,Ó.I/I1¿1¿35456«6«8!9!;—;—= > @�@�BõBõDiEiGÝGÝIQJQLÅLÅN9O9Q­Q­S!T!V•V•X Y [}[}]ñ]ñ_e`ebÙbÙdMeMgÁgÁi5j5l©l©noq‘q‘stvyvyxíxíza{a}Õ}ÕI€I‚½‚½„1…1‡¥‡¥‰ŠŒ‹Œ‹ŽþŽþ�q‘q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI�öIÐ�öIEöILEöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIMnöI(MnöIë®öI¤ë®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöIØöI€ØöI½÷Iü½÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷I³¸÷I,³¸÷I øI˜ øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIËûIÌËûI›ØûI›ØûICëûI€} 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$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI÷öIÐ÷öI•DöIL•DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÉmöI(ÉmöIg®öI¤g®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI›×öI€›×öI9÷Iü9÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I±`÷I´ ±`÷I…÷I0 …÷IS²÷IPS²÷IY»÷I,Y»÷I—øI˜—øIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI­ûIÌ­ûI}ØûI}ØûI%ëûI€} 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$S%S'Ç'Ç);*;,¯,¯.#/#1—1—3 4 668ó8ó:g;g=Û=Û?M@MB¿B¿D1E1G£G£IJL‡L‡NùNùPkQkSÝSÝUOVOXÁXÁZ3[3]¥]¥_`b‰b‰dûdûfmgmißißkQlQnÃnÃp5q5s§s§uvx‹x‹zýzý|o}oáá�S‚S„ńņ7‡7‰¨‰¨‹ŒŽŠŽŠ�û�û’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l{ïõI�öIÐ�öIEöILEöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöISnöI(SnöIñ®öI¤ñ®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI%ØöI€%ØöIÃ÷IüÃ÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷IŸ`÷I´ Ÿ`÷Iÿ„÷I0 ÿ„÷IA²÷IPA²÷IM»÷I,M»÷I©øI˜©øI颸I$颸IÁxùIèÁxùIûdúIìûdúIkûIÌkûI;ØûI;ØûIãêûI€} 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$;%;'®'®)!*!,”,”./1z1z3í3í5`6`8Ó8Ó:F;F=¹=¹?*@*B›B›D E G}G}IîIîK_L_NÐNÐPAQAS²S²U#V#X”X”Z[]v]v_ç_çaXbXdÉdÉf:g:i«i«kln�n�pþpþrosouàuàwQxQzÂzÂ|3}3¤¤�‚„†„††÷†÷ˆg‰g‹׋×�GŽG�·�·’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÏmöI(ÏmöIm®öI¤m®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI¡×öI€¡×öI?÷Iü?÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷I¹¸÷I,¹¸÷I×øI˜×øI£øI$£øI÷xùIè÷xùI1eúIì1eúIyûIÌyûIIÙûIIÙûIñëûI€} 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�ïõIuöIÐuöIEöILEöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIGnöI(GnöIå®öI¤å®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI·÷Iü·÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ii`÷I´ i`÷IÉ„÷I0 É„÷I ²÷IP ²÷I³¸÷I,³¸÷ImøI˜møI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIéØûIéØûI‘ëûI€} 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$\%\'Ñ'Ñ)F*F,»,».0/01¥1¥346�6�89;y;y=î=î?c@cBØBØDMEMGÂGÂI8J8L¬L¬N O Q”Q”STV|V|XðXðZd[d]Ø]Ø_L`LbÀbÀd4e4g¨g¨ijl�l�noqxqxsìsìu`v`xÔxÔzH{H}¼}¼0€0‚¤‚¤„…‡އމŠŒxŒxŽíŽí�b‘b“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIýöIÐýöI›DöIL›DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÏmöI(ÏmöIm®öI¤m®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI¡×öI€¡×öI?÷Iü?÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷IÕ`÷I´ Õ`÷I5…÷I0 5…÷Iw²÷IPw²÷IM»÷I,M»÷ImøI˜møI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIרûIרûIëûI€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbåbådXeXgËgËi>j>l±l±n"o"q“q“stvuvuxæxæzW{W}È}È9€9‚ª‚ª„…‡�‡�‰þ‰þ‹oŒoŽàŽà�Q‘Q“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI{öIÐ{öIEöILEöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIMnöI(MnöIë®öI¤ë®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI½÷Iü½÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷IS»÷I,S»÷IøI˜øI£øI$£øIÙxùIèÙxùIeúIìeúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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$H%H'½'½)2*2,§,§./1‘1‘346{6{8ð8ð:e;e=Ú=Ú?O@OBÄBÄD9E9G®G®I#J#L–L–N O Q|Q|SïSïUbVbXÕXÕZH[H]»]»_.`.b b deg„g„iöiökhlhnÚnÚpLqLs¾s¾u0v0x¢x¢z{}†}†øø�j‚j„܄܆P‡P‰ÉË6Œ6ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIöIÐöI¡DöIL¡DöILöITLöI�UöIÐ�UöIÃXöIœÃXöIÕmöI(ÕmöIs®öI¤s®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI§×öI€§×öIE÷IüE÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Iÿ`÷I´ ÿ`÷I_…÷I0 _…÷I¡²÷IP¡²÷I³¸÷I,³¸÷IøI˜øI%£øI$%£øIýxùIèýxùI7eúIì7eúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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$k%k'ã'ã)[*[,Ó,Ó.K/K1Ã1Ã3;4;6³6³8+9+;£;£=>@“@“B C EƒEƒGûGûIsJsLéLéN_O_QÕQÕSKTKVÁVÁX7Y7[­[­]#^#`™`™bce…e…gûgûiqjqlçlçn]o]qÓqÓsItIv¿v¿x5y5{«{«}!~!€—€—‚ ƒ …ƒ…ƒ‡ù‡ù‰oŠoŒåŒåŽ[�[‘ёѓ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIoöIÐoöI EöIL EöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIAnöI(AnöIß®öI¤ß®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöIØöI€ØöI±÷Iü±÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IW`÷I´ W`÷I·„÷I0 ·„÷Iù±÷IPù±÷I³¸÷I,³¸÷I!øI˜!øI£øI$£øIñxùIèñxùI+eúIì+eúICûIÌCûIÙûIÙûI»ëûI€} 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$[›[›]^`‰`‰bcewewgîgîiejelÜlÜnNoNqÀqÀs2t2v¤v¤xy{ˆ{ˆ}ú}úl€l‚Þ‚Þ„P…P‡‡‰4Š4Œ¦Œ¦Ž�‘БГ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIöIÐöI¡DöIL¡DöILöITLöI�UöIÐ�UöIÃXöIœÃXöIÕmöI(ÕmöIs®öI¤s®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI§×öI€§×öIE÷IüE÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷IM»÷I,M»÷I—øI˜—øI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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"×"×$M%M'Ã'Ã)9*9,¯,¯.%/%1›1›346‡6‡8ý8ý:s;s=é=é?_@_BÕBÕDKEKGÁGÁI7J7L«L«NOQ“Q“STV{V{XïXïZc[c]×]×_K`Kb¿b¿d3e3g§g§ijl�l�noqwqwsësëu_v_xÓxÓzG{G}»}»/€/‚£‚£„…‡‹‡‹‰ÿ‰ÿ‹sŒsŽçŽç�[‘[“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIuöIÐuöIEöILEöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIGnöI(GnöIå®öI¤å®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIØöI€ØöI·÷Iü·÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I�`÷I´ �`÷Ií„÷I0 í„÷I/²÷IP/²÷I_»÷I,_»÷IÏøI˜ÏøI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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$?%?'´'´))*),ž,ž./1ˆ1ˆ3ý3ý5r6r8ç8ç:\;\=Ñ=Ñ?F@FB»B»D0E0G¥G¥IJL‘L‘NOQ{Q{SðSðUeVeXÚXÚZO[O]Ä]Ä_9`9b®b®d#e#g˜g˜i j l‚l‚nônôpfqfsØsØuJvJx¼x¼z.{.} } €‚„‚„„ö„ö†h‡h‰Ú‰Ú‹LŒL޾޾�0‘0“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI öIÐ öI§DöIL§DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Iÿ`÷I´ ÿ`÷I_…÷I0 _…÷I¡²÷IP¡²÷I¹¸÷I,¹¸÷IµøI˜µøI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»n0o0q¥q¥stv�v�xy{y{y}î}îc€c‚؂؄M…M‡‡‰7Š7Œ¬Œ¬Ž!�!‘–‘–“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIiöIÐiöIEöILEöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI;nöI(;nöIÙ®öI¤Ù®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI ØöI€ ØöI«÷Iü«÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷IE`÷I´ E`÷I¥„÷I0 ¥„÷Iç±÷IPç±÷I¹¸÷I,¹¸÷IÕøI˜ÕøI£øI$£øI÷xùIè÷xùI1eúIì1eúI[ûIÌ[ûI+ÙûI+ÙûIÓëûI€} 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·l‡ïõI öIÐ öI§DöIL§DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷IY»÷I,Y»÷IWøI˜WøI £øI$ £øIåxùIèåxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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$M%M'Â'Â)7*7,¬,¬.!/!1–1–3 4 6€6€8õ8õ:j;j=ß=ß?T@TBÉBÉD>E>G³G³I(J(L›L›NOQ�Q�SôSôUgVgXÚXÚZM[M]À]À_3`3b¦b¦degŒgŒiÿiÿkrlrnånåpXqXsËsËu>v>x±x±z${$}—}— € ‚}‚}„ð„ð†c‡c‰Ö‰Ö‹IŒI޼޼�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIoöIÐoöI EöIL EöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI;nöI(;nöIÙ®öI¤Ù®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI ØöI€ ØöI«÷Iü«÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷Iq»÷I,q»÷I‰øI˜‰øI£øI$£øIßxùIèßxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$l%l'à'à)T*T,È,È.Œ>ްް�"‘"“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI­DöIL­DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIámöI(ámöI®öI¤®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI³×öI€³×öIQ÷IüQ÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷I¿¸÷I,¿¸÷IiøI˜iøI£øI$£øI÷xùIè÷xùI1eúIì1eúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIcöIÐcöIEöILEöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI5nöI(5nöIÓ®öI¤Ó®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIØöI€ØöI¥÷Iü¥÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I a÷I´  a÷Ik…÷I0 k…÷I­²÷IP­²÷I¿¸÷I,¿¸÷I‰øI˜‰øI £øI$ £øIåxùIèåxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$K%K'À'À)5*5,ª,ª./1”1”3 4 6~6~8ó8ó:h;h=Ý=Ý?P@PBÃBÃD6E6G©G©IJL�L�NOQuQuSèSèU[V[XÎXÎZA[A]´]´_'`'bšbšd e g€g€ióiókflfnÙnÙpLqLs¿s¿u2v2x¥x¥z{}‹}‹þþ�q‚q„ä„ä†W‡W‰ɉɋ;Œ;Ž­Ž­�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI­DöIL­DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷IK`÷I´ K`÷I«„÷I0 «„÷Ií±÷IPí±÷Ik»÷I,k»÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúI¿ûIÌ¿ûI�ØûI�ØûI7ëûI€} 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$W%W'Ë'Ë)?*?,³,³.'/'1›1›346ƒ6ƒ8÷8÷:k;k=ß=ß?Q@QBÃBÃD5E5G§G§IJL‹L‹NýNýPoQoSáSáUSVSXÅXÅZ7[7]©]©_`b�b�dÿdÿfqgqiãiãkUlUnÇnÇp9q9s«s«uvx�x�z{}s}såå�W‚W„ɄɆ;‡;‰¬‰¬‹ŒŽŽŽŽ�ÿ�ÿ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIiöIÐiöIEöILEöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI5nöI(5nöIÓ®öI¤Ó®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIØöI€ØöI¥÷Iü¥÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Iÿ`÷I´ ÿ`÷I_…÷I0 _…÷I¡²÷IP¡²÷I}»÷I,}»÷IIøI˜IøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúI}ûIÌ}ûIMØûIMØûIõêûI€} 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·l“ïõIöIÐöI³DöIL³DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIçmöI(çmöI…®öI¤…®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI¹×öI€¹×öIW÷IüW÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I]`÷I´ ]`÷I½„÷I0 ½„÷Iÿ±÷IPÿ±÷I¿¸÷I,¿¸÷IøI˜øI1£øI$1£øI yùIè yùICeúIìCeúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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·l“ïõI]öIÐ]öIûDöILûDöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöI)nöI()nöIÇ®öI¤Ç®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIû×öI€û×öI™÷Iü™÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I¿¸÷I,¿¸÷IøI˜øI£øI$£øI÷xùIè÷xùI1eúIì1eúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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·l“ïõI]öIÐ]öIûDöILûDöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöI/nöI(/nöIÍ®öI¤Í®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöIŸ÷IüŸ÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I‰»÷I,‰»÷IåøI˜åøI£øI$£øIëxùIèëxùI%eúIì%eúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI-öI�-öIËDöIËDöI IöI$ IöIÿmöI�ÿmöI�®öI�®öIÛ²öI$Û²öIÑ×öI�Ñ×öIo÷Io÷I­÷I$­÷Ia_÷I�a_÷IÁƒ÷IüÁƒ÷I…¶÷I …¶÷IQøI| QøII£øIÜ I£øI!yùI  !yùI[eúI¤ [eúIµûI„ µûI…ÙûIÈ…ÙûI-ìûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIWöI�WöIõDöI õDöI�KöI�KöI)nöI€)nöIÇ®öIüÇ®öIoµöIoµöIû×öIpû×öI™÷Iì™÷IA÷IôA÷I a÷I` a÷Ik…÷IÌk…÷I…¶÷Iì…¶÷IQøIL QøI7£øI¬ 7£øIyùIp yùIIeúIt IeúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI-öI�-öIËDöIËDöI IöI$ IöIùmöI�ùmöI—®öI—®öIÛ²öI$Û²öIË×öI�Ë×öIi÷Ii÷I­÷I$­÷Ia_÷I�a_÷IÁƒ÷IüÁƒ÷I…¶÷I …¶÷IQøI| QøI1£øIÜ 1£øI yùI   yùICeúI¤ CeúI=ûI„ =ûI ÙûIÈ ÙûIµëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIWöI�WöIõDöI õDöI�KöI�KöI)nöI€)nöIÇ®öIüÇ®öIoµöIoµöIû×öIpû×öI™÷Iì™÷IA÷IôA÷Ia÷I`a÷Iw…÷IÌw…÷I…¶÷Iì…¶÷IQøIL QøI+£øI¬ +£øIyùIp yùI=eúIt =eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI'öI�'öIÅDöI ÅDöIIöIIöIùmöI€ùmöI—®öIü—®öIÕ²öIÕ²öIË×öIpË×öIi÷Iìi÷I§÷Iô§÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷I¶÷Iì¶÷IKøIL KøI7£øI¬ 7£øIyùIp yùIIeúIt IeúImûIT mûI=ÙûI˜=ÙûIåëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI]öI�]öIûDöI ûDöI�KöI�KöI/nöI€/nöIÍ®öIüÍ®öIoµöIoµöIØöIpØöIŸ÷IìŸ÷IA÷IôA÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷I¶÷Iì¶÷IKøIL KøI7£øI¬ 7£øIyùIp yùIIeúIt IeúImûIT mûI=ÙûI˜=ÙûIåëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI!öI�!öI¿DöI ¿DöIIöIIöIómöI€ómöI‘®öIü‘®öIÕ²öIÕ²öIÅ×öIpÅ×öIc÷Iìc÷I§÷Iô§÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷I¶÷Iì¶÷IKøIL KøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI]öI�]öIûDöI ûDöI�KöI�KöI/nöI€/nöIÍ®öIüÍ®öIoµöIoµöIØöIpØöIŸ÷IìŸ÷IA÷IôA÷Ia÷I`a÷Iq…÷IÌq…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI!öI�!öI¿DöI ¿DöI IöI IöIómöI€ómöI‘®öIü‘®öIÛ²öIÛ²öIÅ×öIpÅ×öIc÷Iìc÷I­÷Iô­÷Iù`÷I`ù`÷IY…÷IÌY…÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIcöI�cöIEöI EöIKöIKöI/nöI€/nöIÍ®öIüÍ®öIQµöIQµöIØöIpØöIŸ÷IìŸ÷I#÷Iô#÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIöI�öI¹DöI ¹DöI IöI IöIímöI€ímöI‹®öIü‹®öIÛ²öIÛ²öI¿×öIp¿×öI]÷Iì]÷I­÷Iô­÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIcöI�cöIEöI EöIKöIKöI5nöI€5nöIÓ®öIüÓ®öIQµöIQµöIØöIpØöI¥÷Iì¥÷I#÷Iô#÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIÙxùIp ÙxùIeúIt eúIƒûIT ƒûISØûI˜SØûIûêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöIýHöIýHöIçmöI€çmöI…®öIü…®öIϲöIϲöI¹×öIp¹×öIW÷IìW÷I¡÷Iô¡÷Ió`÷I`ó`÷IS…÷IÌS…÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIiöI�iöIEöI EöI�KöI�KöI5nöI€5nöIÓ®öIüÓ®öIoµöIoµöIØöIpØöI¥÷Iì¥÷IA÷IôA÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI ³DöIýHöIýHöIçmöI€çmöI…®öIü…®öIϲöIϲöI¹×öIp¹×öIW÷IìW÷I¡÷Iô¡÷Ia÷I`a÷Ie…÷IÌe…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIiöI�iöIEöI EöI�KöI�KöI;nöI€;nöIÙ®öIüÙ®öIoµöIoµöI ØöIp ØöI«÷Iì«÷IA÷IôA÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Iy¶÷Iìy¶÷IEøIL EøI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI³DöI ³DöI IöI IöIámöI€ámöI®öIü®öIÛ²öIÛ²öI³×öIp³×öIQ÷IìQ÷I­÷Iô­÷Iç`÷I`ç`÷IG…÷IÌG…÷Is¶÷Iìs¶÷I?øIL ?øI1£øI¬ 1£øI yùIp  yùICeúIt CeúI—ûIT —ûIgÙûI˜gÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI­DöI ­DöI IöI IöIámöI€ámöI®öIü®öIÛ²öIÛ²öI³×öIp³×öIQ÷IìQ÷I­÷Iô­÷Ia÷I`a÷Ie…÷IÌe…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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‡ïõI öI� öI§DöI §DöIñHöIñHöIÛmöI€ÛmöIy®öIüy®öIòöIòöI­×öIp­×öIK÷IìK÷I•÷Iô•÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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‡ïõIoöI�oöI EöI  EöI‘KöI‘KöIAnöI€AnöIß®öIüß®öIcµöIcµöIØöIpØöI±÷Iì±÷I5÷Iô5÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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‡ïõI öI� öI§DöI §DöIñHöIñHöIÕmöI€ÕmöIs®öIüs®öIòöIòöI§×öIp§×öIE÷IìE÷I•÷Iô•÷Ia÷I`a÷Ie…÷IÌe…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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‡ïõIuöI�uöIEöI EöI‘KöI‘KöIAnöI€AnöIß®öIüß®öIcµöIcµöIØöIpØöI±÷Iì±÷I5÷Iô5÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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�ïõIöI�öI¡DöI ¡DöIýHöIýHöIÕmöI€ÕmöIs®öIüs®öIϲöIϲöI§×öIp§×öIE÷IìE÷I¡÷Iô¡÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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�ïõIöI�öI¡DöI ¡DöIýHöIýHöIÏmöI€ÏmöIm®öIüm®öIϲöIϲöI¡×öIp¡×öI?÷Iì?÷I¡÷Iô¡÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI³ûIT ³ûIƒØûI˜ƒØûI+ëûIœ€} 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�ïõIuöI�uöIEöI EöImKöImKöIGnöI€GnöIå®öIüå®öI?µöI?µöIØöIpØöI·÷Iì·÷I÷Iô÷Iá`÷I`á`÷IA…÷IÌA…÷Im¶÷Iìm¶÷I9øIL 9øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIqûIT qûIAØûI˜AØûIéêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L{ïõIýöI�ýöI›DöI ›DöIëHöIëHöIÏmöI€ÏmöIm®öIüm®öI½²öI½²öI¡×öIp¡×öI?÷Iì?÷I�÷Iô�÷I{`÷I`{`÷IÛ„÷IÔÛ„÷Ig¶÷Iôg¶÷I3øIT 3øI%£øI´ %£øIMwùIx MwùI+GúI| +GúIûI\ ûIu×ûI u×ûIêûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõI{öI�{öIEöI EöI…KöI…KöIMnöI€MnöIë®öIüë®öIWµöIWµöIØöIpØöI½÷Iì½÷I)÷Iô)÷I_÷I`_÷Iƒ÷Ĩ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI;wùIp ;wùIGúIt GúIûIT ûI×ûI˜×ûI½éûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõI÷öI�÷öI•DöI •DöIëHöIëHöIÉmöI€ÉmöIg®öIüg®öI½²öI½²öI›×öIp›×öI9÷Iì9÷I�÷Iô�÷I“`÷I`“`÷Ió„÷IÔó„÷Ig¶÷Iôg¶÷I3øIT 3øI £øI´  £øI5wùIx 5wùIGúI| GúI ûI\  ûI×ûI ×ûI«éûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L{ïõI{öI�{öIEöI EöI‹KöI‹KöIMnöI€MnöIë®öIüë®öI]µöI]µöIØöIpØöI½÷Iì½÷I/÷Iô/÷I_÷I`_÷Iƒ÷Ĩ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI/wùIp /wùI GúIt  GúI×ûIT ×ûIÍÖûI˜ÍÖûIuéûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI“öIè“öI1EöId1EöIßKöIlßKöIiUöIèiUöIŸXöI´ŸXöI_nöI@_nöIý®öI¼ý®öI±µöIıµöI;¿öI@;¿öIqÂöI qÂöI1ØöI˜1ØöIÏ÷IÏ÷Iƒ÷I ƒ÷I )÷I˜  )÷IC,÷Id C,÷I…Y÷Ið …Y÷Iù`÷IÌ ù`÷IY…÷IH Y…÷I›²÷Ih›²÷IÕÀ÷IDÕÀ÷I‘õ÷IБõ÷ISøIôSøIQ øI Q øI%£øI¬%£øIýxùIpýxùI7eúIt7eúI±rúIT±rúI‹ûIà‹ûI[ÙûI$[ÙûIìûI(€} 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„uïõI¯öI¯öIMDöI|MDöI·IöI„·IöIíLöIðíLöIwVöI|wVöI­YöIH­YöI�möIÔ�möI®öIP®öI‰³öIX‰³öI¿¶öIÄ¿¶öIIÀöIPIÀöIÃöIÃöIS×öI¨S×öIñ÷I$ ñ÷I[÷I, [÷I‘ ÷I˜ ‘ ÷I*÷I$ *÷IQ-÷Ið Q-÷I“Z÷I| “Z÷I�`÷IX�`÷Ií„÷IÔí„÷I/²÷Iô/²÷IçÀ÷IÐçÀ÷I£õ÷I\£õ÷I¡øI€¡øIŸ øI¬Ÿ øI£øI8£øIßxùIüßxùIeúIeúI“rúIà“rúIéûIléûI¹ØûI°¹ØûIaëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI“öIè“öI1EöId1EöIßKöIlßKöIiUöIèiUöIŸXöI´ŸXöIenöI@enöI¯öI¼¯öI±µöIıµöI;¿öI@;¿öIqÂöI qÂöI7ØöI˜7ØöIÕ÷IÕ÷Iƒ÷I ƒ÷I )÷I˜  )÷IC,÷Id C,÷I…Y÷Ið …Y÷IÉ`÷IÌ É`÷I)…÷IH )…÷Ik²÷Ihk²÷IÕÀ÷IDÕÀ÷I‘õ÷IБõ÷ISøIôSøIQ øI Q øI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúIûIàûIãØûI$ãØûI‹ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõIýöIèýöI›DöId›DöIåKöIlåKöIoUöIèoUöI¥XöI´¥XöIÉmöI@ÉmöIg®öI¼g®öI·µöIÄ·µöIA¿öI@A¿öIwÂöI wÂöI›×öI˜›×öI9÷I9÷I‰÷I ‰÷I)÷I˜ )÷II,÷Id I,÷I‹Y÷Ið ‹Y÷I�`÷IÌ �`÷Ií„÷IH í„÷I/²÷Ih/²÷IÛÀ÷IDÛÀ÷I—õ÷IЗõ÷IYøIôYøIW øI W øI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúIIûIàIûIÙûI$ÙûIÁëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI�öIè�öI+EöId+EöIåKöIlåKöIoUöIèoUöI¥XöI´¥XöIYnöI@YnöI÷®öI¼÷®öI·µöIÄ·µöIA¿öI@A¿öIwÂöI wÂöI+ØöI˜+ØöIÉ÷IÉ÷I‰÷I ‰÷I)÷I˜ )÷II,÷Id I,÷I‹Y÷Ið ‹Y÷Ia÷IÌ a÷Iq…÷IH q…÷I³²÷Ih³²÷IÛÀ÷IDÛÀ÷I—õ÷IЗõ÷IYøIôYøIW øI W øI£øI¬£øIëxùIpëxùI%eúIt%eúIŸrúITŸrúIIûIàIûIÙûI$ÙûIÁëûI(€} 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$\%\'Ó'Ó)J*J,Á,Á.8/81¯1¯3&4&6�6�89;‹;‹=>@y@yBðBðDgEgGÞGÞIUJULÊLÊN?O?Q´Q´S)T)VžVžXY[ˆ[ˆ]ý]ý_r`rbçbçd\e\gÑgÑiFjFl»l»nônôpiqisÞsÞuSvSxÈxÈz{{{}ð}ðe€e‚Ú‚Ú„O…O‡ćĉ9Š9Œ®Œ®Ž#�#‘˜‘˜“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI÷öIè÷öI•DöId•DöIåKöIlåKöIoUöIèoUöI¥XöI´¥XöIÉmöI@ÉmöIg®öI¼g®öI·µöIÄ·µöIA¿öI@A¿öIwÂöI wÂöI›×öI˜›×öI9÷I9÷I‰÷I ‰÷I)÷I˜ )÷II,÷Id I,÷I‹Y÷Ið ‹Y÷I“`÷IÌ “`÷Ió„÷IH ó„÷I5²÷Ih5²÷IÛÀ÷IDÛÀ÷I—õ÷IЗõ÷IYøIôYøIW øI W øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIÑûIàÑûI¡ØûI$¡ØûIIëûI(€} 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$f%f'Ú'Ú)N*N,Â,Â.6/61ª1ª346’6’89;z;z=î=î?b@bBÖBÖDJEJG¾G¾I2J2L¤L¤NOQˆQˆSúSúUlVlXÞXÞZP[P]Â]Â_4`4bgbgdÙdÙfKgKi½i½k/l/nènèpZqZsÌsÌu>v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI�öIè�öI+EöId+EöIåKöIlåKöIoUöIèoUöI¥XöI´¥XöI_nöI@_nöIý®öI¼ý®öI·µöIÄ·µöIA¿öI@A¿öIwÂöI wÂöI1ØöI˜1ØöIÏ÷IÏ÷I‰÷I ‰÷I)÷I˜ )÷II,÷Id I,÷I‹Y÷Ið ‹Y÷Ií`÷IÌ í`÷IM…÷IH M…÷I�²÷Ih�²÷IÛÀ÷IDÛÀ÷I—õ÷IЗõ÷IYøIôYøIW øI W øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIãûIàãûI³ØûI$³ØûI[ëûI(€} 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·l�ïõI�öIÐ�öIEöILEöIëKöITëKöIuUöIÐuUöI«XöIœ«XöISnöI(SnöIñ®öI¤ñ®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI%ØöI€%ØöIÃ÷IüÃ÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I‡`÷I´ ‡`÷Iç„÷I0 ç„÷I)²÷IP)²÷Iá½÷I,á½÷I-øI˜-øI £øI$ £øIåxùIèåxùIeúIìeúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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$u%u'ë'ë)a*a,×,×.M/M1Ã1Ã39496¯6¯8%9%;›;›=>@…@…BùBùDmEmGáGáIUJULÉLÉN=O=Q±Q±S%T%V™V™X Y [�[�]õ]õ_i`ibÝbÝdQeQgÅgÅi9j9l­l­n!o!q•q•s t v}v}xñxñze{e}Ù}ÙM€M‚Á‚Á„5…5‡©‡©‰ŠŒ�Œ�Ž�‘u‘u“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIýöIÐýöI›DöIL›DöIåKöITåKöIoUöIÐoUöI¥XöIœ¥XöIÏmöI(ÏmöIm®öI¤m®öI·µöI¬·µöIA¿öI(A¿öIwÂöIôwÂöI¡×öI€¡×öI?÷Iü?÷I‰÷I ‰÷I)÷I€ )÷II,÷IL I,÷I‹Y÷IØ ‹Y÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷I‡À÷I,‡À÷IÃøI˜ÃøIã¢øI$ã¢øI»xùIè»xùIõdúIìõdúIkûIÌkûI;ØûI;ØûIãêûI€} 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$9%9'¬'¬)*,’,’./1x1x3ë3ë5^6^8Ñ8Ñ:D;D=·=·?(@(B™B™D E G{G{IìIìK]L]NÎNÎP?Q?S°S°U!V!X’X’Z[]t]t_å_åaVbVdÇdÇf8g8i©i©kln‹n‹püpürmsmuÞuÞwOxOzÀzÀ|1}1¢¢�‚„„„„†õ†õˆe‰e‹Õ‹Õ�EŽE�µ�µ’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI‡öIЇöI%EöIL%EöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIYnöI(YnöI÷®öI¤÷®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI+ØöI€+ØöIÉ÷IüÉ÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I«`÷I´ «`÷I …÷I0  …÷IM²÷IPM²÷IuÀ÷I,uÀ÷IÃøI˜ÃøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI³ûI̳ûIƒØûIƒØûI+ëûI€} 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$U%U'É'É)=*=,±,±.%/%1™1™3 4 6�6�8õ8õ:i;i=Ý=Ý?O@OBÁBÁD3E3G¥G¥IJL‰L‰NûNûPmQmSßSßUQVQXÃXÃZ5[5]§]§_`b‹b‹dýdýfogoiáiákSlSnÅnÅp7q7s©s©uvx�x�zÿzÿ|q}qãã�U‚U„DŽdž9‡9‰ª‰ª‹ŒŽŒŽŒ�ý�ý’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷Iá½÷I,á½÷IýøI˜ýøI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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$X%X'Í'Í)B*B,·,·.,/,1¡1¡346‹6‹89;u;u=ê=ê?_@_BÔBÔDIEIG¾G¾I4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb¼b¼d0e0g¤g¤ijlŒlŒnoqtqtsèsèu\v\xÐxÐzD{D}¸}¸,€,‚ ‚ „…‡ЇЉÿ‰ÿ‹tŒtŽéŽé�^‘^“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI{öIÐ{öIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIMnöI(MnöIë®öI¤ë®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöIØöI€ØöI½÷Iü½÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷I{`÷I´ {`÷IÛ„÷I0 Û„÷I²÷IP²÷Iù½÷I,ù½÷IùøI˜ùøI%£øI$%£øIýxùIèýxùI7eúIì7eúIûIÌûIOÙûIOÙûI÷ëûI€} 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�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷IÏ`÷I´ Ï`÷I/…÷I0 /…÷Iq²÷IPq²÷I{À÷I,{À÷I“øI˜“øI£øI$£øIÙxùIèÙxùIeúIìeúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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�ïõI�öIÐ�öIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöISnöI(SnöIñ®öI¤ñ®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI%ØöI€%ØöIÃ÷IüÃ÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷I™`÷I´ ™`÷Iù„÷I0 ù„÷I;²÷IP;²÷IŸÀ÷I,ŸÀ÷I›øI˜›øI £øI$ £øIåxùIèåxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$q•q•stvwvwxèxèzY{Y}Ê}Ê;€;‚¬‚¬„…‡�‡�‰ŠŒqŒqŽâŽâ�S‘S“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI öIÐ öI§DöIL§DöI LöIT LöI“UöIГUöIÉXöIœÉXöIÛmöI(ÛmöIy®öI¤y®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöI­×öI€­×öIK÷IüK÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Ie…÷I0 e…÷I§²÷IP§²÷Iç½÷I,ç½÷I3øI˜3øI£øI$£øIñxùIèñxùI+eúIì+eúICûIÌCûIÙûIÙûI»ëûI€} 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$[%['Ò'Ò)I*I,À,À.7/71®1®3%4%6œ6œ89;Š;Š=>@x@xBïBïDfEfGÝGÝIVJVLÍLÍNDODQ»Q»S2T2V©V©X Y [—[—]^`…`…bübüdsesgêgêiajalØlØnJoJq¼q¼s.t.v v xy{„{„}ö}öh€h‚Ú‚Ú„L…L‡¾‡¾‰0Š0Œ¢Œ¢Ž�‘†‘†“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIuöIÐuöIEöILEöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIGnöI(GnöIå®öI¤å®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIØöI€ØöI·÷Iü·÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Ii`÷I´ i`÷IÉ„÷I0 É„÷I ²÷IP ²÷I¾÷I,¾÷I¿øI˜¿øI+£øI$+£øIyùIèyùI=eúIì=eúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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��  "÷"÷$o%o'ç'ç)_*_,×,×.O/O1Ç1Ç3?4?6·6·8/9/;§;§=>@—@—BCE‡E‡GÿGÿIwJwLíLíNcOcQÙQÙSOTOVÅVÅX;Y;[±[±]'^'`�`�bce‰e‰gÿgÿiujulëlënaoaq×q×sMtMvÃvÃx9y9{¯{¯}%~%€›€›‚ƒ…‡…‡‡ý‡ý‰sŠsŒéŒéŽ_�_‘Õ‘Õ“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI öIÐ öI§DöIL§DöI LöIT LöI“UöIГUöIÉXöIœÉXöIÛmöI(ÛmöIy®öI¤y®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöI­×öI€­×öIK÷IüK÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷I{À÷I,{À÷IÉøI˜ÉøI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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$=%='²'²)'*',œ,œ./1†1†3û3û5p6p8å8å:Z;Z=Ï=Ï?D@DB¹B¹D.E.G£G£IJL�L�NOQyQySîSîUcVcXØXØZM[M]Â]Â_7`7b¬b¬d!e!g–g–i j l€l€nònòpdqdsÖsÖuHvHxºxºz,{,}ž}ž€‚‚‚‚„ô„ô†f‡f‰؉؋JŒJ޼޼�.‘.“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI{öIÐ{öIEöILEöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIGnöI(GnöIå®öI¤å®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöIØöI€ØöI·÷Iü·÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷IÃÀ÷I,ÃÀ÷ImøI˜møI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIéØûIéØûI‘ëûI€} 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$O%O'Å'Å);*;,±,±.'/'1�1�346‰6‰8ÿ8ÿ:u;u=ë=ë?a@aB×B×DMEMGÃGÃI9J9L­L­N!O!Q•Q•S T V}V}XñXñZe[e]Ù]Ù_M`MbÁbÁd5e5g©g©ijl‘l‘noqyqysísíuavaxÕxÕzI{I}½}½1€1‚¥‚¥„…‡�‡�‰ŠŒuŒuŽéŽé�]‘]“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI­DöIL­DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIámöI(ámöI®öI¤®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI³×öI€³×öIQ÷IüQ÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ia÷I´ a÷Ie…÷I0 e…÷I§²÷IP§²÷Iÿ½÷I,ÿ½÷IùøI˜ùøI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l�ïõIoöIÐoöI EöIL EöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIAnöI(AnöIß®öI¤ß®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI±÷Iü±÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I]`÷I´ ]`÷I½„÷I0 ½„÷Iÿ±÷IPÿ±÷I/¾÷I,/¾÷I‹øI˜‹øI%£øI$%£øIýxùIèýxùI7eúIì7eúI[ûIÌ[ûI+ÙûI+ÙûIÓëûI€} 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·l�ïõIöIÐöI­DöIL­DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIÛmöI(ÛmöIy®öI¤y®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI­×öI€­×öIK÷IüK÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷I¥À÷I,¥À÷I¡øI˜¡øI£øI$£øIßxùIèßxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l�ïõIuöIÐuöIEöILEöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIAnöI(AnöIß®öI¤ß®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI±÷Iü±÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷IçÀ÷I,çÀ÷IEøI˜EøI £øI$ £øIåxùIèåxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI³DöIL³DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIçmöI(çmöI…®öI¤…®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI¹×öI€¹×öIW÷IüW÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷I¾÷I,¾÷IÅøI˜ÅøI £øI$ £øIåxùIèåxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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$G%G'¼'¼)1*1,¦,¦./1�1�346z6z8ï8ï:d;d=Ù=Ù?L@LB¿B¿D2E2G¥G¥IJL‹L‹NþNþPqQqSäSäUWVWXÊXÊZ=[=]°]°_#`#b–b–d e g|g|iïiïkblbnÕnÕpHqHs»s»u.v.x¡x¡z{}‡}‡úú�m‚m„à„à†S‡S‰ʼnŋ7Œ7ީީ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIiöIÐiöIEöILEöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöI5nöI(5nöIÓ®öI¤Ó®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI¥÷Iü¥÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷IG¾÷I,G¾÷IWøI˜WøI£øI$£øI÷xùIè÷xùI1eúIì1eúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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·l�ïõIöIÐöI³DöIL³DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIámöI(ámöI®öI¤®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI³×öI€³×öIQ÷IüQ÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷IW`÷I´ W`÷I·„÷I0 ·„÷Iù±÷IPù±÷IÉÀ÷I,ÉÀ÷IsøI˜søIû¢øI$û¢øIÓxùIèÓxùI eúIì eúI}ûIÌ}ûIMØûIMØûIõêûI€} 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·l“ïõIöIÐöI¹DöIL¹DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIçmöI(çmöI…®öI¤…®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI¹×öI€¹×öIW÷IüW÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷I5¾÷I,5¾÷I‘øI˜‘øI£øI$£øI÷xùIè÷xùI1eúIì1eúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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·l“ïõIcöIÐcöIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöI/nöI(/nöIÍ®öI¤Í®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöIØöI€ØöIŸ÷IüŸ÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷Ie¾÷I,e¾÷IøI˜øI7£øI$7£øIyùIèyùIIeúIìIeúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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·l“ïõIöIÐöI³DöIL³DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIçmöI(çmöI…®öI¤…®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI¹×öI€¹×öIW÷IüW÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷IE`÷I´ E`÷I¥„÷I0 ¥„÷Iç±÷IPç±÷IíÀ÷I,íÀ÷IKøI˜KøI£øI$£øIëxùIèëxùI%eúIì%eúIãûIÌãûI³ØûI³ØûI[ëûI€} 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$J%J'¿'¿)4*4,©,©./1“1“346}6}8ò8ò:g;g=Ü=Ü?Q@QBÆBÆD;E;G°G°I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b¢b¢deg†g†iøiøkjljnÜnÜpNqNsÀsÀu2v2x¤x¤z{}ˆ}ˆúú�l‚l„Þ„Þ†R‡R‰ʼnŋ8Œ8ޫޫ�‘“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l“ïõIcöIÐcöIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöI5nöI(5nöIÓ®öI¤Ó®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöIØöI€ØöI¥÷Iü¥÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I/Á÷I,/Á÷IïøI˜ïøI£øI$£øI÷xùIè÷xùI1eúIì1eúIûIÌûIïØûIïØûI—ëûI€} 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$b%b'Ù'Ù)P*P,Ç,Ç.>/>1µ1µ3,4,6£6£89;‘;‘=>@@BöBöDmEmGäGäI[J[LÐLÐNEOEQºQºS/T/V¤V¤XY[Ž[Ž]^`x`xbëbëd^e^gÑgÑiDjDl·l·n(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡“‡“‰ŠŒuŒu޿޿�W‘W“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17L™ïõI3öI�3öIÑDöIÑDöI IöI$ IöInöI�nöI£®öI£®öIÛ²öI$Û²öI××öI�××öIu÷Iu÷I­÷I$­÷Ia_÷I�a_÷IÁƒ÷IüÁƒ÷I…¶÷I …¶÷IQøI| QøI=£øIÜ =£øIyùI  yùIOeúI¤ OeúIgûI„ gûI7ÙûIÈ7ÙûIßëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI]öI�]öIûDöI ûDöI�KöI�KöI/nöI€/nöIÍ®öIüÍ®öIoµöIoµöIØöIpØöIŸ÷IìŸ÷IA÷IôA÷I a÷I` a÷Ik…÷IÌk…÷I…¶÷Iì…¶÷IQøIL QøII£øI¬ I£øI!yùIp !yùI[eúIt [eúI¯ûIT ¯ûIÙûI˜ÙûI'ìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI-öI�-öIËDöIËDöI IöI$ IöIÿmöI�ÿmöI�®öI�®öIÛ²öI$Û²öIÑ×öI�Ñ×öIo÷Io÷I­÷I$­÷Ia_÷I�a_÷IÁƒ÷IüÁƒ÷I…¶÷I …¶÷IQøI| QøI+£øIÜ +£øIyùI  yùI=eúI¤ =eúI ûI„  ûIÝØûIÈÝØûI…ëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI]öI�]öIûDöI ûDöI�KöI�KöI/nöI€/nöIÍ®öIüÍ®öIoµöIoµöIØöIpØöIŸ÷IìŸ÷IA÷IôA÷Ia÷I`a÷Iq…÷IÌq…÷I…¶÷Iì…¶÷IQøIL QøI1£øI¬ 1£øI yùIp  yùICeúIt CeúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI-öI�-öIËDöI ËDöIIöIIöIÿmöI€ÿmöI�®öIü�®öIÕ²öIÕ²öIÑ×öIpÑ×öIo÷Iìo÷I§÷Iô§÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI7£øI¬ 7£øIyùIp yùIIeúIt IeúIsûIT sûICÙûI˜CÙûIëëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIcöI�cöIEöI EöI�KöI�KöI/nöI€/nöIÍ®öIüÍ®öIoµöIoµöIØöIpØöIŸ÷IìŸ÷IA÷IôA÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷I¶÷Iì¶÷IKøIL KøI7£øI¬ 7£øIyùIp yùIIeúIt IeúImûIT mûI=ÙûI˜=ÙûIåëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI'öI�'öIÅDöI ÅDöIIöIIöIùmöI€ùmöI—®öIü—®öIÕ²öIÕ²öIË×öIpË×öIi÷Iìi÷I§÷Iô§÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIcöI�cöIEöI EöI�KöI�KöI5nöI€5nöIÓ®öIüÓ®öIoµöIoµöIØöIpØöI¥÷Iì¥÷IA÷IôA÷Ia÷I`a÷Iq…÷IÌq…÷I¶÷Iì¶÷IKøIL KøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI'öI�'öIÅDöI ÅDöI IöI IöIómöI€ómöI‘®öIü‘®öIÛ²öIÛ²öIÅ×öIpÅ×öIc÷Iìc÷I­÷Iô­÷Ió`÷I`ó`÷IS…÷IÌS…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIiöI�iöIEöI EöIyKöIyKöI5nöI€5nöIÓ®öIüÓ®öIKµöIKµöIØöIpØöI¥÷Iì¥÷I÷Iô÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷I¶÷Iì¶÷IKøIL KøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúICûIT CûIÙûI˜ÙûI»ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI!öI�!öI¿DöI ¿DöI IöI IöIómöI€ómöI‘®öIü‘®öIÛ²öIÛ²öIÅ×öIpÅ×öIc÷Iìc÷I­÷Iô­÷Iù`÷I`ù`÷IY…÷IÌY…÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øIßxùIp ßxùIeúIt eúI‰ûIT ‰ûIYØûI˜YØûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIiöI�iöIEöI EöIyKöIyKöI;nöI€;nöIÙ®öIüÙ®öIKµöIKµöI ØöIp ØöI«÷Iì«÷I÷Iô÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIËûIT ËûI›ØûI˜›ØûICëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI ¿DöIýHöIýHöIímöI€ímöI‹®öIü‹®öIϲöIϲöI¿×öIp¿×öI]÷Iì]÷I¡÷Iô¡÷Iù`÷I`ù`÷IY…÷IÌY…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIiöI�iöIEöI EöI£KöI£KöI;nöI€;nöIÙ®öIüÙ®öIuµöIuµöI ØöIp ØöI«÷Iì«÷IG÷IôG÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIoöI�oöI EöI  EöI£KöI£KöI;nöI€;nöIÙ®öIüÙ®öIuµöIuµöI ØöIp ØöI«÷Iì«÷IG÷IôG÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI ³DöI IöI IöIçmöI€çmöI…®öIü…®öIÛ²öIÛ²öI¹×öIp¹×öIW÷IìW÷I­÷Iô­÷Ií`÷I`í`÷IM…÷IÌM…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIoöI�oöI EöI  EöIsKöIsKöIAnöI€AnöIß®öIüß®öIEµöIEµöIØöIpØöI±÷Iì±÷I÷Iô÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúI‘ûIT ‘ûIaÙûI˜aÙûI ìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI³DöI ³DöI IöI IöIámöI€ámöI®öIü®öIÛ²öIÛ²öI³×öIp³×öIQ÷IìQ÷I­÷Iô­÷I a÷I` a÷Ik…÷IÌk…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI­DöI ­DöI÷HöI÷HöIámöI€ámöI®öIü®öIɲöIɲöI³×öIp³×öIQ÷IìQ÷I›÷Iô›÷Iç`÷I`ç`÷IG…÷IÌG…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIUûIT UûI%ÙûI˜%ÙûIÍëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIuöI�uöIEöI EöI—KöI—KöIGnöI€GnöIå®öIüå®öIiµöIiµöIØöIpØöI·÷Iì·÷I;÷Iô;÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI­DöI ­DöI÷HöI÷HöIÛmöI€ÛmöIy®öIüy®öIɲöIɲöI­×öIp­×öIK÷IìK÷I›÷Iô›÷I a÷I` a÷Ik…÷IÌk…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIuöI�uöIEöI EöI—KöI—KöIGnöI€GnöIå®öIüå®öIiµöIiµöIØöIpØöI·÷Iì·÷I;÷Iô;÷IO_÷I`O_÷I¯ƒ÷I̯ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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�ïõI öI� öI§DöI §DöIýHöIýHöIÛmöI€ÛmöIy®öIüy®öIϲöIϲöI­×öIp­×öIK÷IìK÷I¡÷Iô¡÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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�ïõI{öI�{öIEöI EöImKöImKöIMnöI€MnöIë®öIüë®öI?µöI?µöIØöIpØöI½÷Iì½÷I÷Iô÷IÏ`÷I`Ï`÷I/…÷IÌ/…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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�ïõIöI�öI¡DöI ¡DöIýHöIýHöIÕmöI€ÕmöIs®öIüs®öIϲöIϲöI§×öIp§×öIE÷IìE÷I¡÷Iô¡÷II_÷I`I_÷I©ƒ÷IÌ©ƒ÷Im¶÷Iìm¶÷I9øIL 9øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIqûIT qûIAØûI˜AØûIéêûIœ€} 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�ïõI{öI�{öIEöI EöImKöImKöIMnöI€MnöIë®öIüë®öI?µöI?µöIØöIpØöI½÷Iì½÷I÷Iô÷Iç`÷I`ç`÷IG…÷IÌG…÷Im¶÷Iìm¶÷I9øIL 9øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI¹ûIT ¹ûI‰ØûI˜‰ØûI1ëûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L�ïõI�öI��öIEöI EöI‹KöI‹KöIMnöI€MnöIë®öIüë®öI]µöI]µöIØöIpØöI½÷Iì½÷I/÷Iô/÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI+£øI¬ +£øISwùIp SwùI1GúIt 1GúI…ûIT …ûI{×ûI˜{×ûI#êûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L�ïõIýöI�ýöI›DöI ›DöIëHöIëHöIÏmöI€ÏmöIm®öIüm®öI½²öI½²öI¡×öIp¡×öI?÷Iì?÷I�÷Iô�÷I�`÷I`�`÷Ií„÷IÔí„÷Ig¶÷Iôg¶÷I3øIT 3øI£øI´ £øI/wùIx /wùI GúI|  GúI×ûI\ ×ûIÍÖûI ÍÖûIuéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L�ïõI�öI��öIEöI EöI‹KöI‹KöISnöI€SnöIñ®öIüñ®öI]µöI]µöI%ØöIp%ØöIÃ÷IìÃ÷I/÷Iô/÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øI;wùIp ;wùIGúIt GúIûIT ûI ×ûI˜ ×ûI±éûIœ€} 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„uïõIµöIµöISDöI|SDöI·IöI„·IöIíLöIðíLöIwVöI|wVöI­YöIH­YöI‡möIÔ‡möI%®öIP%®öI‰³öIX‰³öI¿¶öIÄ¿¶öIIÀöIPIÀöIÃöIÃöIY×öI¨Y×öI÷÷I$ ÷÷I[÷I, [÷I‘ ÷I˜ ‘ ÷I*÷I$ *÷IQ-÷Ið Q-÷I“Z÷I| “Z÷I]`÷IX]`÷I½„÷IÔ½„÷Iÿ±÷Iôÿ±÷IáÀ÷IÐáÀ÷I�õ÷I\�õ÷I}øI€}øI{ øI¬{ øI%£øI8%£øIýxùIüýxùI7eúI7eúI±rúIà±rúIûIlûIOÙûI°OÙûI÷ëûI´€} 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„uïõIµöIµöISDöI|SDöI·IöI„·IöIíLöIðíLöIwVöI|wVöI­YöIH­YöI�möIÔ�möI®öIP®öI‰³öIX‰³öI¿¶öIÄ¿¶öIIÀöIPIÀöIÃöIÃöIS×öI¨S×öIñ÷I$ ñ÷I[÷I, [÷I‘ ÷I˜ ‘ ÷I*÷I$ *÷IQ-÷Ið Q-÷I“Z÷I| “Z÷I�`÷IX�`÷Ií„÷IÔí„÷I/²÷Iô/²÷IÁ÷IÐÁ÷IÓõ÷I\Óõ÷IøI€øI}!øI¬}!øI £øI8 £øIåxùIüåxùIeúIeúI™rúIà™rúI ûIl ûIÝØûI°ÝØûI…ëûI´€} 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 })outputElements576input0Elements640input1Elements6272invocations1çxuïõI™öIè™öI7EöId7EöIåKöIlåKöIoUöIèoUöI¥XöI´¥XöIknöI@knöI ¯öI¼ ¯öI·µöIÄ·µöIA¿öI@A¿öIwÂöI wÂöI=ØöI˜=ØöIÛ÷IÛ÷I‰÷I ‰÷I)÷I˜ )÷II,÷Id I,÷I‹Y÷Ið ‹Y÷IÕ`÷IÌ Õ`÷I5…÷IH 5…÷Iw²÷Ihw²÷IáÀ÷IDáÀ÷I�õ÷IÐ�õ÷I}øIô}øI{ øI { øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúI×ûIà×ûI§ØûI$§ØûIOëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõIöIèöI¡DöId¡DöIëKöIlëKöIuUöIèuUöI«XöI´«XöIÏmöI@ÏmöIm®öI¼m®öI½µöIĽµöIG¿öI@G¿öI}ÂöI }ÂöI¡×öI˜¡×öI?÷I?÷I�÷I �÷I)÷I˜ )÷IO,÷Id O,÷I‘Y÷Ið ‘Y÷Ic`÷IÌ c`÷IÄ÷IH Ä÷I²÷Ih²÷IçÀ÷IDçÀ÷I£õ÷IУõ÷IƒøIôƒøI� øI � øI£øI¬£øI÷xùIp÷xùI1eúIt1eúI«rúIT«rúI‹ûIà‹ûI[ÙûI$[ÙûIìûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõIýöIèýöI›DöId›DöIëKöIlëKöIuUöIèuUöI«XöI´«XöIÏmöI@ÏmöIm®öI¼m®öI½µöIĽµöIG¿öI@G¿öI}ÂöI }ÂöI¡×öI˜¡×öI?÷I?÷I�÷I �÷I)÷I˜ )÷IO,÷Id O,÷I‘Y÷Ið ‘Y÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷IçÀ÷IDçÀ÷I£õ÷IУõ÷IƒøIôƒøI� øI � øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúIûIàûIãØûI$ãØûI‹ëûI(€} 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$J%J'À'À)6*6,¬,¬."/"1˜1˜346„6„8ú8ú:p;p=æ=æ?\@\BÒBÒDHEHG¾G¾I4J4L¨L¨NOQ�Q�STVxVxXìXìZ`[`]Ô]Ô_H`Hb{b{dïdïfcgci×i×kKlKnoq|q|sðsðudvdxØxØzL{L}À}À4€4‚¨‚¨„…‡�‡�‰ŠŒxŒxŽìŽì�`‘`“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI“öIè“öI1EöId1EöIëKöIlëKöIuUöIèuUöI«XöI´«XöIenöI@enöI¯öI¼¯öI½µöIĽµöIG¿öI@G¿öI}ÂöI }ÂöI7ØöI˜7ØöIÕ÷IÕ÷I�÷I �÷I)÷I˜ )÷IO,÷Id O,÷I‘Y÷Ið ‘Y÷Iu`÷IÌ u`÷IÕ„÷IH Õ„÷I²÷Ih²÷I Á÷ID Á÷IÇõ÷IÐÇõ÷I7øIô7øI5!øI 5!øI颸I¬颸IÁxùIpÁxùIûdúItûdúIurúITurúIkûIàkûI;ØûI$;ØûIãêûI(€} 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$:%:'­'­) * ,“,“./1y1y3ì3ì5_6_8Ò8Ò:E;E=¸=¸?)@)BšBšD E G|G|IíIíK^L^NÏNÏP@Q@S±S±U"V"X“X“Z[]u]u_æ_æabd‹d‹füfühmimkÞkÞm�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷I³¸÷I,³¸÷IíøI˜íøI£øI$£øIëxùIèëxùI%eúIì%eúIIûIÌIûIÙûIÙûIÁëûI€} 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·l�ïõIöIÐöI¡DöIL¡DöIëKöITëKöIuUöIÐuUöI«XöIœ«XöIÕmöI(ÕmöIs®öI¤s®öI½µöI¬½µöIG¿öI(G¿öI}ÂöIô}ÂöI§×öI€§×öIE÷IüE÷I�÷I �÷I)÷I€ )÷IO,÷IL O,÷I‘Y÷IØ ‘Y÷I¥`÷I´ ¥`÷I…÷I0 …÷IG²÷IPG²÷I;»÷I,;»÷IwøI˜wøIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIãûIÌãûI³ØûI³ØûI[ëûI€} 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·l�ïõI�öIÐ�öI+EöIL+EöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöI_nöI(_nöIý®öI¤ý®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI1ØöI€1ØöIÏ÷IüÏ÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I]`÷I´ ]`÷I½„÷I0 ½„÷Iÿ±÷IPÿ±÷IG»÷I,G»÷I—øI˜—øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúIÑûIÌÑûI¡ØûI¡ØûIIëûI€} 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�ïõI öIÐ öI§DöIL§DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷I³¸÷I,³¸÷I�øI˜�øI £øI$ £øIåxùIèåxùIeúIìeúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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�ïõI�öIÐ�öIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöISnöI(SnöIñ®öI¤ñ®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI%ØöI€%ØöIÃ÷IüÃ÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷I�`÷I´ �`÷Iá„÷I0 á„÷I#²÷IP#²÷I³¸÷I,³¸÷I¯øI˜¯øI£øI$£øIñxùIèñxùI+eúIì+eúICûIÌCûIÙûIÙûI»ëûI€} 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�ïõI öIÐ öI§DöIL§DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷IÉ`÷I´ É`÷I)…÷I0 )…÷Ik²÷IPk²÷IA»÷I,A»÷IøI˜øIû¢øI$û¢øIÓxùIèÓxùI eúIì eúI³ûI̳ûIƒØûIƒØûI+ëûI€} 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$T%T'È'È)<*<,°,°.$/$1˜1˜3 4 6€6€8ô8ô:h;h=Ü=Ü?N@NBÀBÀD2E2G¤G¤IJLˆLˆNúNúPlQlSÞSÞUPVPXÂXÂZ4[4]¦]¦_`bŠbŠdüdüfngniàiàkRlRnÄnÄp6q6s¨s¨uvxŒxŒzþzþ|p}pââ�T‚T„ƄƆ8‡8‰©‰©‹ŒŽ‹Ž‹�ü�ü’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI‡öIЇöI%EöIL%EöILöITLöI�UöIÐ�UöIÃXöIœÃXöISnöI(SnöIñ®öI¤ñ®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI%ØöI€%ØöIÃ÷IüÃ÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷IE`÷I´ E`÷I¥„÷I0 ¥„÷Iç±÷IPç±÷IY»÷I,Y»÷IQøI˜QøI£øI$£øIßxùIèßxùIeúIìeúIéûIÌéûI¹ØûI¹ØûIaëûI€} 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$>%>'³'³)(*(,�,�./1‡1‡3ü3ü5q6q8æ8æ:[;[=Ð=Ð?E@EBºBºD/E/G¤G¤IJL�L�NOQzQzSïSïUdVdXÙXÙZN[N]Ã]Ã_8`8b­b­d"e"g—g—i j l�l�nónópeqes×s×uIvIx»x»z-{-}Ÿ}Ÿ€‚ƒ‚ƒ„õ„õ†g‡g‰ىًKŒK޽޽�/‘/“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõIöIÐöI­DöIL­DöILöITLöI™UöIЙUöIÏXöIœÏXöIámöI(ámöI®öI¤®öIáµöI¬áµöIk¿öI(k¿öI¡ÂöIô¡ÂöI³×öI€³×öIQ÷IüQ÷I³÷I ³÷I=)÷I€ =)÷Is,÷IL s,÷IµY÷IØ µY÷I a÷I´  a÷Ik…÷I0 k…÷I­²÷IP­²÷I¹¸÷I,¹¸÷ICøI˜CøI%£øI$%£øIýxùIèýxùI7eúIì7eúIûIÌûIOÙûIOÙûI÷ëûI€} 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·l‡ïõI{öIÐ{öIEöILEöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIMnöI(MnöIë®öI¤ë®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI½÷Iü½÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Io`÷I´ o`÷IÏ„÷I0 Ï„÷I²÷IP²÷I¹¸÷I,¹¸÷IcøI˜cøI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l‡ïõIöIÐöI­DöIL­DöI LöIT LöI“UöIГUöIÉXöIœÉXöIÛmöI(ÛmöIy®öI¤y®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöI­×öI€­×öIK÷IüK÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ió`÷I´ ó`÷IS…÷I0 S…÷I•²÷IP•²÷IS»÷I,S»÷IßøI˜ßøI £øI$ £øIåxùIèåxùIeúIìeúI ûIÌ ûIÝØûIÝØûI…ëûI€} 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$]%]'Ô'Ô)K*K,Â,Â.9/91°1°3'4'6ž6ž89;Œ;Œ=>@z@zBñBñDhEhGßGßIVJVLËLËN@O@QµQµS*T*VŸVŸXY[‰[‰]þ]þ_s`sbæbædYeYgÌgÌi?j?l²l²n#o#q”q”stvvvvxçxçzX{X}É}É:€:‚«‚«„…‡އމÿ‰ÿ‹pŒpŽáŽá�R‘R“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI�öIÐ�öIEöILEöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIMnöI(MnöIë®öI¤ë®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI½÷Iü½÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I?`÷I´ ?`÷IŸ„÷I0 Ÿ„÷Iá±÷IPá±÷Ie»÷I,e»÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõIöIÐöI³DöIL³DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIçmöI(çmöI…®öI¤…®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI¹×öI€¹×öIW÷IüW÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I a÷I´  a÷Ik…÷I0 k…÷I­²÷IP­²÷I¹¸÷I,¹¸÷I÷øI˜÷øI+£øI$+£øIyùIèyùI=eúIì=eúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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·l�ïõIuöIÐuöIEöILEöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIAnöI(AnöIß®öI¤ß®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI±÷Iü±÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ia÷I´ a÷Ie…÷I0 e…÷I§²÷IP§²÷I¹¸÷I,¹¸÷IøI˜øI £øI$ £øIåxùIèåxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l�ïõIöIÐöI³DöIL³DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIámöI(ámöI®öI¤®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI³×öI€³×öIQ÷IüQ÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Iÿ`÷I´ ÿ`÷I_…÷I0 _…÷I¡²÷IP¡²÷I_»÷I,_»÷IŸøI˜ŸøI£øI$£øIëxùIèëxùI%eúIì%eúIûIÌûIïØûIïØûI—ëûI€} 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·l�ïõIuöIÐuöIEöILEöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIGnöI(GnöIå®öI¤å®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöIØöI€ØöI·÷Iü·÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ií`÷I´ í`÷IM…÷I0 M…÷I�²÷IP�²÷Iw»÷I,w»÷IÑøI˜ÑøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúI}ûIÌ}ûIMØûIMØûIõêûI€} 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·l�ïõIöIÐöI¹DöIL¹DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIçmöI(çmöI…®öI¤…®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI¹×öI€¹×öIW÷IüW÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Ia÷I´ a÷Ie…÷I0 e…÷I§²÷IP§²÷I¿¸÷I,¿¸÷I«øI˜«øI%£øI$%£øIýxùIèýxùI7eúIì7eúI[ûIÌ[ûI+ÙûI+ÙûIÓëûI€} 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·l�ïõIoöIÐoöI EöIL EöIýKöITýKöI‡UöIЇUöI½XöIœ½XöI;nöI(;nöIÙ®öI¤Ù®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI ØöI€ ØöI«÷Iü«÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I¿¸÷I,¿¸÷IËøI˜ËøI£øI$£øI÷xùIè÷xùI1eúIì1eúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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·l�ïõIöIÐöI³DöIL³DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIçmöI(çmöI…®öI¤…®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI¹×öI€¹×öIW÷IüW÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Ia÷I´ a÷Ie…÷I0 e…÷I§²÷IP§²÷Iq»÷I,q»÷I_øI˜_øI£øI$£øIëxùIèëxùI%eúIì%eúIõûIÌõûIÅØûIÅØûImëûI€} 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·l�ïõIoöIÐoöI EöIL EöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIAnöI(AnöIß®öI¤ß®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöIØöI€ØöI±÷Iü±÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Ia÷I´ a÷Iq…÷I0 q…÷I³²÷IP³²÷Iƒ»÷I,ƒ»÷I‘øI˜‘øI£øI$£øIëxùIèëxùI%eúIì%eúIãûIÌãûI³ØûI³ØûI[ëûI€} 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·l“ïõIcöIÐcöIEöILEöILöITLöI�UöIÐ�UöIÃXöIœÃXöI5nöI(5nöIÓ®öI¤Ó®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöIØöI€ØöI¥÷Iü¥÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷IŸ÷I,Ÿ÷IøI˜øI+£øI$+£øIyùIèyùI=eúIì=eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l“ïõIöIÐöI¹DöIL¹DöILöITLöI�UöIÐ�UöIÃXöIœÃXöIímöI(ímöI‹®öI¤‹®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI¿×öI€¿×öI]÷Iü]÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷I]`÷I´ ]`÷I½„÷I0 ½„÷Iÿ±÷IPÿ±÷I}»÷I,}»÷IøI˜øI £øI$ £øIåxùIèåxùIeúIìeúIËûIÌËûI›ØûI›ØûICëûI€} 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·l“ïõIiöIÐiöIEöILEöILöITLöI�UöIÐ�UöIÃXöIœÃXöI;nöI(;nöIÙ®öI¤Ù®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI ØöI€ ØöI«÷Iü«÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I•»÷I,•»÷IQøI˜QøI£øI$£øIñxùIèñxùI+eúIì+eúIûûIÌûûIËØûIËØûIsëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI3öI�3öIÑDöIÑDöI IöI$ IöInöI�nöI£®öI£®öIÛ²öI$Û²öI××öI�××öIu÷Iu÷I­÷I$­÷Ig_÷I�g_÷Iǃ÷Iüǃ÷I…¶÷I …¶÷IQøI| QøI1£øIÜ 1£øI yùI   yùICeúI¤ CeúI1ûI„ 1ûIÙûIÈÙûI©ëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIcöI�cöIEöI EöI—KöI—KöI5nöI€5nöIÓ®öIüÓ®öIiµöIiµöIØöIpØöI¥÷Iì¥÷I;÷Iô;÷Ia÷I`a÷Iq…÷IÌq…÷I…¶÷Iì…¶÷IQøIL QøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI3öI�3öIÑDöI ÑDöIIöIIöIÿmöI€ÿmöI�®öIü�®öIÕ²öIÕ²öIÑ×öIpÑ×öIo÷Iìo÷I§÷Iô§÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøII£øI¬ I£øI!yùIp !yùI[eúIt [eúI¯ûIT ¯ûIÙûI˜ÙûI'ìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIiöI�iöIEöI EöI£KöI£KöI5nöI€5nöIÓ®öIüÓ®öIuµöIuµöIØöIpØöI¥÷Iì¥÷IG÷IôG÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIiöI�iöIEöI EöI£KöI£KöI;nöI€;nöIÙ®öIüÙ®öIuµöIuµöI ØöIp ØöI«÷Iì«÷IG÷IôG÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI £øI¬  £øIåxùIp åxùIeúIt eúI�ûIT �ûI_ØûI˜_ØûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI-öI�-öIËDöI ËDöI IöI IöIùmöI€ùmöI—®öIü—®öIÛ²öIÛ²öIË×öIpË×öIi÷Iìi÷I­÷Iô­÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷I¶÷Iì¶÷IKøIL KøI1£øI¬ 1£øI yùIp  yùICeúIt CeúImûIT mûI=ÙûI˜=ÙûIåëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI'öI�'öIÅDöI ÅDöIIöIIöIùmöI€ùmöI—®öIü—®öIá²öIá²öIË×öIpË×öIi÷Iìi÷I³÷Iô³÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIoöI�oöI EöI  EöIyKöIyKöI;nöI€;nöIÙ®öIüÙ®öIKµöIKµöI ØöIp ØöI«÷Iì«÷I÷Iô÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI ¿DöIýHöIýHöIómöI€ómöI‘®öIü‘®öIϲöIϲöIÅ×öIpÅ×öIc÷Iìc÷I¡÷Iô¡÷Ió`÷I`ó`÷IS…÷IÌS…÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIoöI�oöI EöI  EöI£KöI£KöIAnöI€AnöIß®öIüß®öIuµöIuµöIØöIpØöI±÷Iì±÷IG÷IôG÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷Iy¶÷Iìy¶÷IEøIL EøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI ¿DöIýHöIýHöIímöI€ímöI‹®öIü‹®öIϲöIϲöI¿×öIp¿×öI]÷Iì]÷I¡÷Iô¡÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI £øI¬  £øIåxùIp åxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIuöI�uöIEöI EöI£KöI£KöIAnöI€AnöIß®öIüß®öIuµöIuµöIØöIpØöI±÷Iì±÷IG÷IôG÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöI IöI IöIímöI€ímöI‹®öIü‹®öIÛ²öIÛ²öI¿×öIp¿×öI]÷Iì]÷I­÷Iô­÷Ió`÷I`ó`÷IS…÷IÌS…÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúI‹ûIT ‹ûI[ÙûI˜[ÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIuöI�uöIEöI EöIsKöIsKöIGnöI€GnöIå®öIüå®öIEµöIEµöIØöIpØöI·÷Iì·÷I÷Iô÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöI IöI IöIçmöI€çmöI…®öIü…®öIÛ²öIÛ²öI¹×öIp¹×öIW÷IìW÷I­÷Iô­÷Ia÷I`a÷Ie…÷IÌe…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIuöI�uöIEöI EöIsKöIsKöIGnöI€GnöIå®öIüå®öIEµöIEµöIØöIpØöI·÷Iì·÷I÷Iô÷IU_÷I`U_÷Iµƒ÷I̵ƒ÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIãûIT ãûI³ØûI˜³ØûI[ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI³DöI ³DöI÷HöI÷HöIçmöI€çmöI…®öIü…®öIɲöIɲöI¹×öIp¹×öIW÷IìW÷I›÷Iô›÷Ií`÷I`í`÷IM…÷IÌM…÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúI‘ûIT ‘ûIaÙûI˜aÙûI ìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI{öI�{öIEöI EöI—KöI—KöIMnöI€MnöIë®öIüë®öIiµöIiµöIØöIpØöI½÷Iì½÷I;÷Iô;÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIÙxùIp ÙxùIeúIt eúIÑûIT ÑûI¡ØûI˜¡ØûIIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI­DöI ­DöI÷HöI÷HöIámöI€ámöI®öIü®öIɲöIɲöI³×öIp³×öIQ÷IìQ÷I›÷Iô›÷Ia÷I`a÷Iq…÷IÌq…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI{öI�{öIEöI EöI—KöI—KöIMnöI€MnöIë®öIüë®öIiµöIiµöIØöIpØöI½÷Iì½÷I;÷Iô;÷Ií`÷I`í`÷IM…÷IÌM…÷Is¶÷Iìs¶÷I?øIL ?øIï¢øI¬ ï¢øIÇxùIp ÇxùIeúIt eúIqûIT qûIAØûI˜AØûIéêûIœ€} 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�ïõIöI�öI­DöI ­DöIýHöIýHöIámöI€ámöI®öIü®öIϲöIϲöI³×öIp³×öIQ÷IìQ÷I¡÷Iô¡÷Iá`÷I`á`÷IA…÷IÌA…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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�ïõI�öI��öIEöI EöImKöImKöIMnöI€MnöIë®öIüë®öI?µöI?µöIØöIpØöI½÷Iì½÷I÷Iô÷IÕ`÷I`Õ`÷I5…÷IÌ5…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIïØûI˜ïØûI—ëûIœ€} 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�ïõI öI� öI§DöI §DöIýHöIýHöIÛmöI€ÛmöIy®öIüy®öIϲöIϲöI­×öIp­×öIK÷IìK÷I¡÷Iô¡÷Ia÷I`a÷Iq…÷IÌq…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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�ïõI�öI��öIEöI EöImKöImKöISnöI€SnöIñ®öIüñ®öI?µöI?µöI%ØöIp%ØöIÃ÷IìÃ÷I÷Iô÷Ió`÷I`ó`÷IS…÷IÌS…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIÙxùIp ÙxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI öI� öI§DöI §DöIëHöIëHöIÕmöI€ÕmöIs®öIüs®öI½²öI½²öI§×öIp§×öIE÷IìE÷I�÷Iô�÷I‡`÷I`‡`÷Iç„÷IÔç„÷Ig¶÷Iôg¶÷I3øIT 3øI£øI´ £øIAwùIx AwùIGúI| GúI1ûI\ 1ûI'×ûI '×ûIÏéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L�ïõI‡öI�‡öI%EöI %EöI‹KöI‹KöISnöI€SnöIñ®öIüñ®öI]µöI]µöI%ØöIp%ØöIÃ÷IìÃ÷I/÷Iô/÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI£øI¬ £øIGwùIp GwùI%GúIt %GúIIûIT IûI?×ûI˜?×ûIçéûIœ€} 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 })outputElements504input0Elements560input1Elements5488invocations17L�ïõIöI�öI¡DöI ¡DöIëHöIëHöIÕmöI€ÕmöIs®öIüs®öI½²öI½²öI§×öIp§×öIE÷IìE÷I�÷Iô�÷I‡`÷I`‡`÷Iç„÷IÔç„÷Ig¶÷Iôg¶÷I3øIT 3øI£øI´ £øI)wùIx )wùIGúI| GúI¹ûI\ ¹ûI¯ÖûI ¯ÖûIWéûI¤€} 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 })outputElements504input0Elements560input1Elements5488invocations17L�ïõI‡öI�‡öI%EöI %EöI‘KöI‘KöIYnöI€YnöI÷®öIü÷®öIcµöIcµöI+ØöIp+ØöIÉ÷IìÉ÷I5÷Iô5÷I%_÷I`%_÷I…ƒ÷IÌ…ƒ÷Ig¶÷Iìg¶÷I3øIL 3øI £øI¬  £øI5wùIp 5wùIGúIt GúIïûIT ïûIåÖûI˜åÖûI�éûIœ€} 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„{ïõI»öI»öIYDöI|YDöI½IöI„½IöIóLöIðóLöI}VöI|}VöI³YöIH³YöI‡möIÔ‡möI%®öIP%®öI�³öIX�³öIŶöIÄŶöIOÀöIPOÀöI…ÃöI…ÃöIY×öI¨Y×öI÷÷I$ ÷÷Ia÷I, a÷I— ÷I˜ — ÷I!*÷I$ !*÷IW-÷Ið W-÷I™Z÷I| ™Z÷I‡`÷IX‡`÷Iç„÷IÜç„÷I)²÷Iü)²÷IçÀ÷IØçÀ÷I£õ÷Id£õ÷I§øIˆ§øI¥ øI´¥ øI£øI@£øIëxùIëxùI%eúI%eúIŸrúIèŸrúIûItûIïØûI¸ïØûI—ëûI¼€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI™öIè™öI7EöId7EöIëKöIlëKöIuUöIèuUöI«XöI´«XöIknöI@knöI ¯öI¼ ¯öI½µöIĽµöIG¿öI@G¿öI}ÂöI }ÂöI=ØöI˜=ØöIÛ÷IÛ÷I�÷I �÷I)÷I˜ )÷IO,÷Id O,÷I‘Y÷Ið ‘Y÷Ia÷IÌ a÷Iq…÷IH q…÷I³²÷Ih³²÷IçÀ÷IDçÀ÷I£õ÷IУõ÷I§øIô§øI¥ øI ¥ øI£øI¬£øIñxùIpñxùI+eúIt+eúI¥rúIT¥rúIOûIàOûIÙûI$ÙûIÇëûI(€} 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„{ïõIµöIµöISDöI|SDöI½IöI„½IöIóLöIðóLöI}VöI|}VöI³YöIH³YöI‡möIÔ‡möI%®öIP%®öI�³öIX�³öIŶöIÄŶöIOÀöIPOÀöI…ÃöI…ÃöIY×öI¨Y×öI÷÷I$ ÷÷Ia÷I, a÷I— ÷I˜ — ÷I!*÷I$ !*÷IW-÷Ið W-÷I™Z÷I| ™Z÷I-`÷IX-`÷I�„÷IÜ�„÷Iϱ÷Iüϱ÷IAÁ÷IØAÁ÷Iýõ÷Idýõ÷I]øIˆ]øI["øI´["øI£øI@£øIßxùIßxùIeúIeúI“rúIè“rúI×ûIt×ûI§ØûI¸§ØûIOëûI¼€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõIöIèöI¡DöId¡DöIñKöIlñKöI{UöIè{UöI±XöI´±XöIÕmöI@ÕmöIs®öI¼s®öIõöIÄõöIM¿öI@M¿öIƒÂöI ƒÂöI§×öI˜§×öIE÷IE÷I•÷I •÷I)÷I˜ )÷IU,÷Id U,÷I—Y÷Ið —Y÷I�`÷IÌ �`÷Ií„÷IP í„÷I/²÷Ip/²÷IíÀ÷ILíÀ÷I©õ÷IØ©õ÷I§øIü§øI¥ øI(¥ øI£øI´£øIëxùIxëxùI%eúI|%eúIŸrúI\ŸrúI=ûIè=ûI ÙûI, ÙûIµëûI0€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI“öIè“öI1EöId1EöIëKöIlëKöIuUöIèuUöI«XöI´«XöIenöI@enöI¯öI¼¯öI½µöIĽµöIG¿öI@G¿öI}ÂöI }ÂöI7ØöI˜7ØöIÕ÷IÕ÷I�÷I �÷I)÷I˜ )÷IO,÷Id O,÷I‘Y÷Ið ‘Y÷Ic`÷IÌ c`÷IÄ÷IH Ä÷I²÷Ih²÷IÁ÷IDÁ÷IÓõ÷IÐÓõ÷I…øIô…øIƒ!øI ƒ!øI £øI¬ £øIåxùIpåxùIeúIteúI™rúIT™rúI%ûIà%ûIõØûI$õØûI�ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI™öIè™öI7EöId7EöIëKöIlëKöIuUöIèuUöI«XöI´«XöIknöI@knöI ¯öI¼ ¯öI½µöIĽµöIG¿öI@G¿öI}ÂöI }ÂöI=ØöI˜=ØöIÛ÷IÛ÷I�÷I �÷I)÷I˜ )÷IO,÷Id O,÷I‘Y÷Ið ‘Y÷Io`÷IÌ o`÷IÏ„÷IH Ï„÷I²÷Ih²÷IíÀ÷IDíÀ÷I©õ÷IЩõ÷I§øIô§øI¥ øI ¥ øIõ¢øI¬õ¢øIÍxùIpÍxùIeúIteúI�rúIT�rúI³ûIà³ûIƒØûI$ƒØûI+ëûI(€} 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�ïõI öIÐ öI§DöIL§DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I{`÷I´ {`÷IÛ„÷I8 Û„÷I²÷IX²÷I½½÷I4½½÷IûøI ûøI£øI,£øIëxùIðëxùI%eúIô%eúIIûIÔIûIÙûIÙûIÁëûI€} 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�ïõI�öIÐ�öI+EöIL+EöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöI_nöI(_nöIý®öI¤ý®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI1ØöI€1ØöIÏ÷IüÏ÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷IK`÷I´ K`÷I«„÷I0 «„÷Ií±÷IPí±÷Ií½÷I,í½÷IuøI˜uøI£øI$£øI÷xùIè÷xùI1eúIì1eúIyûIÌyûIIÙûIIÙûIñëûI€} 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�ïõI öIÐ öI§DöIL§DöIñKöITñKöI{UöIÐ{UöI±XöIœ±XöIÛmöI(ÛmöIy®öI¤y®öIõöI¬õöIM¿öI(M¿öIƒÂöIôƒÂöI­×öI€­×öIK÷IüK÷I•÷I •÷I)÷I€ )÷IU,÷IL U,÷I—Y÷IØ —Y÷I'`÷I´ '`÷I‡„÷I8 ‡„÷Iɱ÷IXɱ÷IEÀ÷I4EÀ÷IøI øIû¢øI,û¢øIÓxùIðÓxùI eúIô eúIÑûIÔÑûI¡ØûI¡ØûIIëûI€} 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�ïõI“öIГöI1EöIL1EöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöI_nöI(_nöIý®öI¤ý®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI1ØöI€1ØöIÏ÷IüÏ÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Ic`÷I´ c`÷IÄ÷I0 Ä÷I²÷IP²÷I‡À÷I,‡À÷IøI˜øI£øI$£øIßxùIèßxùIeúIìeúIûIÌûIרûIרûIëûI€} 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‡ïõIöIÐöI­DöIL­DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIámöI(ámöI®öI¤®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI³×öI€³×öIQ÷IüQ÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷Iç`÷I´ ç`÷IG…÷I8 G…÷I‰²÷IX‰²÷IÕ½÷I4Õ½÷I¯øI ¯øI£øI,£øIÙxùIðÙxùIeúIôeúIÑûIÔÑûI¡ØûI¡ØûIIëûI€} 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‡ïõI‡öIЇöI%EöIL%EöILöITLöI�UöIÐ�UöIÃXöIœÃXöIYnöI(YnöI÷®öI¤÷®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI+ØöI€+ØöIÉ÷IüÉ÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷I9`÷I´ 9`÷I™„÷I0 ™„÷IÛ±÷IPÛ±÷I¾÷I,¾÷IAøI˜AøI%£øI$%£øIýxùIèýxùI7eúIì7eúI‹ûIÌ‹ûI[ÙûI[ÙûIìûI€} 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‡ïõIöIÐöI­DöIL­DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIÛmöI(ÛmöIy®öI¤y®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI­×öI€­×öIK÷IüK÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷IÃ`÷I´ Ã`÷I#…÷I8 #…÷Ie²÷IXe²÷IiÀ÷I4iÀ÷IEøI EøIï¢øI,ï¢øIÇxùIðÇxùIeúIôeúIqûIÔqûIAØûIAØûIéêûI€} 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‡ïõI�öIÐ�öI+EöIL+EöILöITLöI�UöIÐ�UöIÃXöIœÃXöIYnöI(YnöI÷®öI¤÷®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI+ØöI€+ØöIÉ÷IüÉ÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷IW`÷I´ W`÷I·„÷I0 ·„÷Iù±÷IPù±÷I«À÷I,«À÷IéøI˜éøI £øI$ £øIåxùIèåxùIeúIìeúIûIÌûIãØûIãØûI‹ëûI€} 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·l‡ïõIöIÐöI³DöIL³DöILöITLöI™UöIЙUöIÏXöIœÏXöIçmöI(çmöI…®öI¤…®öIáµöI¬áµöIk¿öI(k¿öI¡ÂöIô¡ÂöI¹×öI€¹×öIW÷IüW÷I³÷I ³÷I=)÷I€ =)÷Is,÷IL s,÷IµY÷IØ µY÷Ia÷I´ a÷Ie…÷I8 e…÷I§²÷IX§²÷Ií½÷I4í½÷I{øI {øI£øI,£øIëxùIðëxùI%eúIô%eúIûIÔûIïØûIïØûI—ëûI€} 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·l‡ïõI�öIÐ�öIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIMnöI(MnöIë®öI¤ë®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöIØöI€ØöI½÷Iü½÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷I!`÷I´ !`÷I�„÷I0 �„÷Iñ÷IPñ÷I¾÷I,¾÷I øI˜ øI£øI$£øI÷xùIè÷xùI1eúIì1eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l‡ïõIöIÐöI³DöIL³DöILöITLöI™UöIЙUöIÏXöIœÏXöIámöI(ámöI®öI¤®öIáµöI¬áµöIk¿öI(k¿öI¡ÂöIô¡ÂöI³×öI€³×öIQ÷IüQ÷I³÷I ³÷I=)÷I€ =)÷Is,÷IL s,÷IµY÷IØ µY÷Ií`÷I´ í`÷IM…÷I8 M…÷I�²÷IX�²÷I�À÷I4�À÷IøI øI£øI,£øIßxùIðßxùIeúIôeúI×ûIÔ×ûI§ØûI§ØûIOëûI€} 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·l‡ïõI�öIÐ�öIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöISnöI(SnöIñ®öI¤ñ®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI%ØöI€%ØöIÃ÷IüÃ÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷IE`÷I´ E`÷I¥„÷I0 ¥„÷Iç±÷IPç±÷IÏÀ÷I,ÏÀ÷I»øI˜»øI £øI$ £øIåxùIèåxùIeúIìeúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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·l�ïõIöIÐöI¹DöIL¹DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIçmöI(çmöI…®öI¤…®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI¹×öI€¹×öIW÷IüW÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Ia÷I´ a÷Iq…÷I8 q…÷I³²÷IX³²÷I ¾÷I4 ¾÷IGøI GøI£øI,£øI÷xùIð÷xùI1eúIô1eúIIûIÔIûIÙûIÙûIÁëûI€} 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·l�ïõI{öIÐ{öIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIGnöI(GnöIå®öI¤å®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöIØöI€ØöI·÷Iü·÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Ia÷I´ a÷Iq…÷I0 q…÷I³²÷IP³²÷I;¾÷I,;¾÷IÓøI˜ÓøI£øI$£øI÷xùIè÷xùI1eúIì1eúI7ûIÌ7ûIÙûIÙûI¯ëûI€} 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·l�ïõI{öIÐ{öIEöILEöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIMnöI(MnöIë®öI¤ë®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöIØöI€ØöI½÷Iü½÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷IóÀ÷I,óÀ÷I“øI˜“øI£øI$£øIßxùIèßxùIeúIìeúI¿ûIÌ¿ûI�ØûI�ØûI7ëûI€} 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·l“ïõI!öIÐ!öI¿DöIL¿DöILöITLöI�UöIÐ�UöIÃXöIœÃXöIímöI(ímöI‹®öI¤‹®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI¿×öI€¿×öI]÷Iü]÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷I a÷I´  a÷Ik…÷I8 k…÷I­²÷IX­²÷I#¾÷I4#¾÷I øI  øI%£øI,%£øIýxùIðýxùI7eúIô7eúI[ûIÔ[ûI+ÙûI+ÙûIÓëûI€} 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·l“ïõIoöIÐoöI EöIL EöILöITLöI�UöIÐ�UöIÃXöIœÃXöIAnöI(AnöIß®öI¤ß®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöIØöI€ØöI±÷Iü±÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷IS¾÷I,S¾÷IŸøI˜ŸøI1£øI$1£øI yùIè yùICeúIìCeúI‘ûIÌ‘ûIaÙûIaÙûI ìûI€} 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��  "÷"÷$k%k'ß'ß)S*S,Ç,Ç.;/;1¯1¯3#4#6—6—8 9 ;;=ó=ó?g@gBÛBÛDOEOGÃGÃI7J7L©L©NOQ�Q�SÿSÿUqVqXãXãZU[U]Ç]Ç_9`9b«b«deg�g�ijlslsnånåpWqWsÉsÉu;v;x­x­z{}‘}‘€‚u‚u„ç„ç†Y‡Y‰ˉˋ=Œ=ޝޝ�!‘!“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l“ïõIuöIÐuöIEöILEöILöITLöI�UöIÐ�UöIÃXöIœÃXöIGnöI(GnöIå®öI¤å®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöIØöI€ØöI·÷Iü·÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Ia÷I´ a÷Iq…÷I0 q…÷I³²÷IP³²÷IÁ÷I,Á÷IeøI˜eøI£øI$£øI÷xùIè÷xùI1eúIì1eúIûIÌûIéØûIéØûI‘ëûI€} 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·l“ïõI!öIÐ!öI¿DöIL¿DöI LöIT LöI“UöIГUöIÉXöIœÉXöIómöI(ómöI‘®öI¤‘®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIÅ×öI€Å×öIc÷Iüc÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷I!`÷I´ !`÷I�„÷I8 �„÷Iñ÷IXñ÷I;¾÷I4;¾÷IÙøI ÙøI£øI,£øIëxùIðëxùI%eúIô%eúIãûIÔãûI³ØûI³ØûI[ëûI€} 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·l“ïõIiöIÐiöIEöILEöI LöIT LöI“UöIГUöIÉXöIœÉXöI;nöI(;nöIÙ®öI¤Ù®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöI ØöI€ ØöI«÷Iü«÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷Ik¾÷I,k¾÷IkøI˜køI=£øI$=£øIyùIèyùIOeúIìOeúI£ûIÌ£ûIsÙûIsÙûIìûI€} 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$>%>'±'±)$*$,—,—. / 1}1}3ð3ð5c6c8Ö8Ö:I;I=¼=¼?-@-BžBžDEG€G€IñIñKbLbNÓNÓPDQDSµSµU&V&X—X—Z[]y]y_ê_êa[b[dÌdÌf=g=i®i®kln�n�pqsrsruãuãwTxTzÅzÅ|6}6§§�‚„‰„‰†ú†úˆj‰j‹Ú‹Ú�JŽJ�º�º’ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l“ïõIoöIÐoöI EöIL EöI LöIT LöI“UöIГUöIÉXöIœÉXöIAnöI(AnöIß®öI¤ß®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIØöI€ØöI±÷Iü±÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷IAÁ÷I,AÁ÷I=øI˜=øI%£øI$%£øIýxùIèýxùI7eúIì7eúI+ûIÌ+ûIûØûIûØûI£ëûI€} 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 })outputElements576input0Elements640input1Elements6272invocations17LŸïõIiöI�iöIEöI EöI—KöI—KöI5nöI€5nöIÓ®öIüÓ®öIiµöIiµöIØöIpØöI¥÷Iì¥÷I;÷Iô;÷Ia÷I`a÷Ie…÷IÌe…÷I…¶÷Iì…¶÷IQøIL QøI=£øI¬ =£øIyùIp yùIOeúIt OeúIsûIT sûICÙûI˜CÙûIëëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LŸïõI9öI�9öI×DöI×DöI IöI$ IöI nöI� nöI©®öI©®öIÛ²öI$Û²öIÝ×öI�Ý×öI{÷I{÷I­÷I$­÷Ig_÷I�g_÷Iǃ÷Iüǃ÷I…¶÷I …¶÷IQøI| QøI+£øIÜ +£øIyùI  yùI=eúI¤ =eúIûûI„ ûûIËØûIÈËØûIsëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LŸïõIiöI�iöIEöI EöI—KöI—KöI;nöI€;nöIÙ®öIüÙ®öIiµöIiµöI ØöIp ØöI«÷Iì«÷I;÷Iô;÷Ia÷I`a÷Iq…÷IÌq…÷I…¶÷Iì…¶÷IQøIL QøI+£øI¬ +£øIyùIp yùI=eúIt =eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI9öI�9öI×DöI ×DöI IöI IöInöI€nöI£®öIü£®öIÛ²öIÛ²öI××öIp××öIu÷Iìu÷I­÷Iô­÷Ig_÷I`g_÷Iǃ÷IÌǃ÷I…¶÷Iì…¶÷IQøIL QøI7£øI¬ 7£øIyùIp yùIIeúIt IeúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIiöI�iöIEöI EöI£KöI£KöI;nöI€;nöIÙ®öIüÙ®öIuµöIuµöI ØöIp ØöI«÷Iì«÷IG÷IôG÷Ig_÷I`g_÷Iǃ÷IÌǃ÷I…¶÷Iì…¶÷IQøIL QøI1£øI¬ 1£øI yùIp  yùICeúIt CeúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI3öI�3öIÑDöI ÑDöI IöI IöInöI€nöI£®öIü£®öIÛ²öIÛ²öI××öIp××öIu÷Iìu÷I­÷Iô­÷Ig_÷I`g_÷Iǃ÷IÌǃ÷I…¶÷Iì…¶÷IQøIL QøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIרûI˜רûIëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIoöI�oöI EöI  EöI£KöI£KöI;nöI€;nöIÙ®öIüÙ®öIuµöIuµöI ØöIp ØöI«÷Iì«÷IG÷IôG÷Ig_÷I`g_÷Iǃ÷IÌǃ÷I…¶÷Iì…¶÷IQøIL QøI£øI¬ £øIñxùIp ñxùI+eúIt +eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI-öI�-öIËDöI ËDöIIöIIöIÿmöI€ÿmöI�®öIü�®öIá²öIá²öIÑ×öIpÑ×öIo÷Iìo÷I³÷Iô³÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI7£øI¬ 7£øIyùIp yùIIeúIt IeúImûIT mûI=ÙûI˜=ÙûIåëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIoöI�oöI EöI  EöIyKöIyKöIAnöI€AnöIß®öIüß®öIKµöIKµöIØöIpØöI±÷Iì±÷I÷Iô÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI7£øI¬ 7£øIyùIp yùIIeúIt IeúI‘ûIT ‘ûIaÙûI˜aÙûI ìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI-öI�-öIËDöI ËDöIIöIIöIùmöI€ùmöI—®öIü—®öIá²öIá²öIË×öIpË×öIi÷Iìi÷I³÷Iô³÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIuöI�uöIEöI EöIyKöIyKöIAnöI€AnöIß®öIüß®öIKµöIKµöIØöIpØöI±÷Iì±÷I÷Iô÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI'öI�'öIÅDöI ÅDöIýHöIýHöIùmöI€ùmöI—®öIü—®öIϲöIϲöIË×öIpË×öIi÷Iìi÷I¡÷Iô¡÷Ií`÷I`í`÷IM…÷IÌM…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIuöI�uöIEöI EöI©KöI©KöIGnöI€GnöIå®öIüå®öI{µöI{µöIØöIpØöI·÷Iì·÷IM÷IôM÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷Iy¶÷Iìy¶÷IEøIL EøI7£øI¬ 7£øIyùIp yùIIeúIt IeúI�ûIT �ûImÙûI˜mÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI'öI�'öIÅDöI ÅDöIýHöIýHöIómöI€ómöI‘®öIü‘®öIϲöIϲöIÅ×öIpÅ×öIc÷Iìc÷I¡÷Iô¡÷Iù`÷I`ù`÷IY…÷IÌY…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIÙxùIp ÙxùIeúIt eúIƒûIT ƒûISØûI˜SØûIûêûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIuöI�uöIEöI EöI©KöI©KöIGnöI€GnöIå®öIüå®öI{µöI{µöIØöIpØöI·÷Iì·÷IM÷IôM÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI ¿DöI IöI IöIómöI€ómöI‘®öIü‘®öIÛ²öIÛ²öIÅ×öIpÅ×öIc÷Iìc÷I­÷Iô­÷Ió`÷I`ó`÷IS…÷IÌS…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúI%ûIT %ûIõØûI˜õØûI�ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõIöI�öI¹DöI ¹DöI IöI IöIímöI€ímöI‹®öIü‹®öIÛ²öIÛ²öI¿×öIp¿×öI]÷Iì]÷I­÷Iô­÷Ia÷I`a÷Ie…÷IÌe…÷Iy¶÷Iìy¶÷IEøIL EøI £øI¬  £øIåxùIp åxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI{öI�{öIEöI EöIsKöIsKöIMnöI€MnöIë®öIüë®öIEµöIEµöIØöIpØöI½÷Iì½÷I÷Iô÷I[_÷I`[_÷I»ƒ÷IÌ»ƒ÷Iy¶÷Iìy¶÷IEøIL EøI £øI¬  £øIåxùIp åxùIeúIt eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¹DöI ¹DöI÷HöI÷HöIímöI€ímöI‹®öIü‹®öIɲöIɲöI¿×öIp¿×öI]÷Iì]÷I›÷Iô›÷Ió`÷I`ó`÷IS…÷IÌS…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI�öI��öIEöI EöI�KöI�KöIMnöI€MnöIë®öIüë®öIoµöIoµöIØöIpØöI½÷Iì½÷IA÷IôA÷IÛ`÷I`Û`÷I;…÷IÌ;…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI³DöI ³DöI÷HöI÷HöIçmöI€çmöI…®öIü…®öIɲöIɲöI¹×öIp¹×öIW÷IìW÷I›÷Iô›÷I a÷I` a÷Ik…÷IÌk…÷Is¶÷Iìs¶÷I?øIL ?øI £øI¬  £øIåxùIp åxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI�öI��öIEöI EöI�KöI�KöISnöI€SnöIñ®öIüñ®öIoµöIoµöI%ØöIp%ØöIÃ÷IìÃ÷IA÷IôA÷Iù`÷I`ù`÷IY…÷IÌY…÷Is¶÷Iìs¶÷I?øIL ?øIû¢øI¬ û¢øIÓxùIp ÓxùI eúIt  eúI¹ûIT ¹ûI‰ØûI˜‰ØûI1ëûIœ€} 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‡ïõIöI�öI³DöI ³DöIIöIIöIámöI€ámöI®öIü®öIÕ²öIÕ²öI³×öIp³×öIQ÷IìQ÷I§÷Iô§÷Ií`÷I`í`÷IM…÷IÌM…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIñxùIp ñxùI+eúIt +eúIOûIT OûIÙûI˜ÙûIÇëûIœ€} 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‡ïõI‡öI�‡öI%EöI %EöIgKöIgKöISnöI€SnöIñ®öIüñ®öI9µöI9µöI%ØöIp%ØöIÃ÷IìÃ÷I ÷Iô ÷Iá`÷I`á`÷IA…÷IÌA…÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûIT ûIOÙûI˜OÙûI÷ëûIœ€} 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‡ïõIöI�öI­DöI ­DöIýHöIýHöIámöI€ámöI®öIü®öIϲöIϲöI³×öIp³×öIQ÷IìQ÷I¡÷Iô¡÷Ia÷I`a÷Iq…÷IÌq…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIßxùIp ßxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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‡ïõI‡öI�‡öI%EöI %EöIgKöIgKöIYnöI€YnöI÷®öIü÷®öI9µöI9µöI+ØöIp+ØöIÉ÷IìÉ÷I ÷Iô ÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúI ûIT  ûIÝØûI˜ÝØûI…ëûIœ€} 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�ïõI‡öI�‡öI%EöI %EöI‘KöI‘KöIYnöI€YnöI÷®öIü÷®öIcµöIcµöI+ØöIp+ØöIÉ÷IìÉ÷I5÷Iô5÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI1£øI¬ 1£øIYwùIp YwùI7GúIt 7GúI—ûIT —ûI�×ûI˜�×ûI5êûIœ€} 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�ïõI öI� öI§DöI §DöIñHöIñHöIÛmöI€ÛmöIy®öIüy®öIòöIòöI­×öIp­×öIK÷IìK÷I•÷Iô•÷I�`÷I`�`÷Iá„÷IÔá„÷Im¶÷Iôm¶÷I9øIT 9øIõ¢øI´ õ¢øIwùIx wùIûFúI| ûFúIwûI\ wûImÖûI mÖûIéûI¤€} 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�ïõI�öI��öI+EöI +EöI‘KöI‘KöIYnöI€YnöI÷®öIü÷®öIcµöIcµöI+ØöIp+ØöIÉ÷IìÉ÷I5÷Iô5÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIAwùIp AwùIGúIt GúIûIT ûI×ûI˜×ûI½éûIœ€} 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„{ïõI»öI»öIYDöI|YDöI½IöI„½IöIóLöIðóLöI}VöI|}VöI³YöIH³YöI�möIÔ�möI+®öIP+®öI�³öIX�³öIŶöIÄŶöIOÀöIPOÀöI…ÃöI…ÃöI_×öI¨_×öIý÷I$ ý÷Ia÷I, a÷I— ÷I˜ — ÷I!*÷I$ !*÷IW-÷Ið W-÷I™Z÷I| ™Z÷I3`÷IX3`÷I“„÷IÜ“„÷IÕ±÷IüÕ±÷IóÀ÷IØóÀ÷I¯õ÷Id¯õ÷I×øIˆ×øIÕ øI´Õ øI£øI@£øIñxùIñxùI+eúI+eúI¥rúIè¥rúI1ûIt1ûIÙûI¸ÙûI©ëûI¼€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõIŸöIèŸöI=EöId=EöIñKöIlñKöI{UöIè{UöI±XöI´±XöIqnöI@qnöI¯öI¼¯öIõöIÄõöIM¿öI@M¿öIƒÂöI ƒÂöICØöI˜CØöIá÷Iá÷I•÷I •÷I)÷I˜ )÷IU,÷Id U,÷I—Y÷Ið —Y÷I]`÷IÌ ]`÷I½„÷IH ½„÷Iÿ±÷Ihÿ±÷IÁ÷IDÁ÷IÍõ÷IÐÍõ÷I[øIô[øIY!øI Y!øI£øI¬£øIÙxùIpÙxùIeúIteúI�rúIT�rúI×ûIà×ûI§ØûI$§ØûIOëûI(€} 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„{ïõI»öI»öIYDöI|YDöI½IöI„½IöIóLöIðóLöI}VöI|}VöI³YöIH³YöI‡möIÔ‡möI%®öIP%®öI�³öIX�³öIŶöIÄŶöIOÀöIPOÀöI…ÃöI…ÃöIY×öI¨Y×öI÷÷I$ ÷÷Ia÷I, a÷I— ÷I˜ — ÷I!*÷I$ !*÷IW-÷Ið W-÷I•[÷I| •[÷I¥`÷IT¥`÷I…÷IÐ…÷IC³÷IðC³÷IqÁ÷IÈqÁ÷I-ö÷IT-ö÷I;øIx;øI9#øI¤9#øI£øI0£øIÙxùIôÙxùIeúIøeúI�rúIØ�rúI¹ûId¹ûI‰ØûI¨‰ØûI1ëûI¬€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI¥öIè¥öICEöIdCEöIñKöIlñKöI{UöIè{UöI±XöI´±XöIwnöI@wnöI¯öI¼¯öIõöIÄõöIM¿öI@M¿öIƒÂöI ƒÂöIIØöI˜IØöIç÷Iç÷I•÷I •÷I)÷I˜ )÷IU,÷Id U,÷I—Y÷Ið —Y÷I™`÷IÌ ™`÷Iù„÷IH ù„÷I;²÷Ih;²÷I•Á÷ID•Á÷IQö÷IÐQö÷IïøIôïøIí#øI í#øIï¢øI¬ï¢øIÇxùIpÇxùIeúIteúI{rúIT{rúIqûIàqûIAØûI$AØûIéêûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI öIè öI§DöId§DöIñKöIlñKöI{UöIè{UöI±XöI´±XöIÛmöI@ÛmöIy®öI¼y®öIõöIÄõöIM¿öI@M¿öIƒÂöI ƒÂöI­×öI˜­×öIK÷IK÷I•÷I •÷I)÷I˜ )÷IU,÷Id U,÷I—Y÷Ið —Y÷I-`÷IÌ -`÷I�„÷IP �„÷Iϱ÷Ipϱ÷IùÀ÷ILùÀ÷Iµõ÷Iصõ÷IÑøIüÑøIÏ øI(Ï øI£øI´£øI÷xùIx÷xùI1eúI|1eúI«rúI\«rúIûIèûIOÙûI,OÙûI÷ëûI0€} 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$E�E�GHJ�J�LMO}O}QôQôSkTkVâVâXYYY[Ð[Ð]G^G`¾`¾bïbïddedgÙgÙiNjNlÃlÃn„o„q÷q÷sjtjvÝvÝxPyP{Ã{Ã}6~6€©€©‚ƒ…�…�‡ˆŠtŠtŒæŒæŽX�X‘ʑʓ  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI™öIè™öI7EöId7EöIñKöIlñKöI{UöIè{UöI±XöI´±XöIknöI@knöI ¯öI¼ ¯öIõöIÄõöIM¿öI@M¿öIƒÂöI ƒÂöI=ØöI˜=ØöIÛ÷IÛ÷I•÷I •÷I)÷I˜ )÷IU,÷Id U,÷I—Y÷Ið —Y÷IQ`÷IÌ Q`÷I±„÷IH ±„÷Ió±÷Ihó±÷IùÀ÷IDùÀ÷Iµõ÷Iеõ÷IÑøIôÑøIÏ øI Ï øI£øI¬£øIßxùIpßxùIeúIteúI“rúIT“rúIûIàûIãØûI$ãØûI‹ëûI(€} 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 })outputElements576input0Elements640input1Elements6272invocations1çx{ïõI öIè öI§DöId§DöIñKöIlñKöI{UöIè{UöI±XöI´±XöIÛmöI@ÛmöIy®öI¼y®öIõöIÄõöIM¿öI@M¿öIƒÂöI ƒÂöI­×öI˜­×öIK÷IK÷I•÷I •÷I)÷I˜ )÷IU,÷Id U,÷I“Z÷Ið “Z÷IÏ`÷IÈ Ï`÷I/…÷ID /…÷Im³÷Idm³÷IùÀ÷I<ùÀ÷Iµõ÷Iȵõ÷IÑøIìÑøIÏ øIÏ øI £øI¤ £øIåxùIhåxùIeúIleúI™rúIL™rúIûIØûIרûIרûIëûI €} 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$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbžbždeg„g„i÷i÷kjljn(o(q™q™s t v{v{xìxìz]{]}Î}Î?€?‚°‚°„!…!‡“‡“‰ŠŒuŒu޿޿�W‘W“  b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1! bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1çx{ïõIŸöIèŸöI=EöId=EöIñKöIlñKöI{UöIè{UöI±XöI´±XöIknöI@knöI ¯öI¼ ¯öIõöIÄõöIM¿öI@M¿öIƒÂöI ƒÂöI=ØöI˜=ØöIÛ÷IÛ÷I•÷I •÷I)÷I˜ )÷IU,÷Id U,÷I—Y÷Ið —Y÷I‡`÷IÌ ‡`÷Iç„÷IH ç„÷I)²÷Ih)²÷IùÀ÷IDùÀ÷Iµõ÷Iеõ÷IÑøIôÑøIÏ øI Ï øIû¢øI¬û¢øIÓxùIpÓxùI eúIt eúI‡rúIT‡rúIÑûIàÑûI¡ØûI$¡ØûIIëûI(€} 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�ïõIöIÐöI­DöIL­DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIámöI(ámöI®öI¤®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI³×öI€³×öIQ÷IüQ÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I�Y÷IØ �Y÷I`÷I´ `÷I{„÷I8 {„÷I½±÷IX½±÷I³¸÷I4³¸÷Ià øI à øI£øI,£øI÷xùIð÷xùI1eúIô1eúI…ûIÔ…ûIUÙûIUÙûIýëûI€} 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�ïõI“öIГöI1EöIL1EöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIenöI(enöI¯öI¤¯öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI7ØöI€7ØöIÕ÷IüÕ÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷I?`÷I´ ?`÷IŸ„÷I0 Ÿ„÷Iá±÷IPá±÷I³¸÷I,³¸÷I=øI˜=øI£øI$£øIëxùIèëxùI%eúIì%eúI=ûIÌ=ûI ÙûI ÙûIµëûI€} 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�ïõIöIÐöI­DöIL­DöI÷KöIT÷KöI�UöIÐ�UöI·XöIœ·XöIÛmöI(ÛmöIy®öI¤y®öIɵöI¬ɵöIS¿öI(S¿öI‰ÂöIô‰ÂöI­×öI€­×öIK÷IüK÷I›÷I ›÷I%)÷I€ %)÷I[,÷IL [,÷I™Z÷IØ ™Z÷Ió`÷I° ó`÷IS…÷I, S…÷I‘³÷IL‘³÷I?º÷I$?º÷IM øI�M øI£øI£øIßxùIàßxùIeúIäeúI ûIÄ ûIÝØûIÝØûI…ëûI €} 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$K%K'Á'Á)7*7,­,­.#/#1™1™346…6…8û8û:q;q=ç=ç?]@]BÓBÓDIEIG¿G¿I5J5L©L©NOQ‘Q‘STVyVyXíXíZa[a]Õ]Õ_I`Ib½b½d1e1g¥g¥ijl�l�noququséséu]v]xÑxÑzE{E}¹}¹-€-‚¡‚¡„…‡‰‡‰‰ý‰ý‹qŒqŽåŽå�Y‘Y“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l�ïõI™öIЙöI7EöIL7EöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIenöI(enöI¯öI¤¯öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI7ØöI€7ØöIÕ÷IüÕ÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷Iu`÷I´ u`÷IÕ„÷I0 Õ„÷I²÷IP²÷IM»÷I,M»÷IÙøI˜ÙøI£øI$£øIÙxùIèÙxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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‡ïõIöIÐöI³DöIL³DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIçmöI(çmöI…®öI¤…®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI¹×öI€¹×öIW÷IüW÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷I£Y÷IØ £Y÷I`÷I´ `÷Io„÷I8 o„÷I±±÷IX±±÷I¹¸÷I4¹¸÷I# øI # øI£øI,£øIñxùIðñxùI+eúIô+eúIOûIÔOûIÙûIÙûIÇëûI€} 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$6›6›89;‰;‰=>@w@wBîBîDeEeGÜGÜISJSLÈLÈN=O=Q²Q²S'T'VœVœXY[†[†]û]û_p`pbåbådZeZgÏgÏiDjDl¹l¹n.o.q£q£stv�v�xy{w{w}ì}ìa€a‚Ö‚Ö„K…K‡À‡À‰5Š5ŒªŒªŽ�‘”‘”“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l‡ïõI�öIÐ�öI+EöIL+EöI LöIT LöI“UöIГUöIÉXöIœÉXöIYnöI(YnöI÷®öI¤÷®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöI+ØöI€+ØöIÉ÷IüÉ÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷I3`÷I´ 3`÷I“„÷I0 “„÷IÕ±÷IPÕ±÷I³¸÷I,³¸÷IñøI˜ñøI£øI$£øIñxùIèñxùI+eúIì+eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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‡ïõIöIÐöI³DöIL³DöIýKöITýKöI‡UöIЇUöI½XöIœ½XöIámöI(ámöI®öI¤®öIϵöI¬ϵöIY¿öI(Y¿öI�ÂöIô�ÂöI³×öI€³×öIQ÷IüQ÷I¡÷I ¡÷I+)÷I€ +)÷Ia,÷IL a,÷IŸZ÷IØ ŸZ÷Iÿ`÷I° ÿ`÷I_…÷I, _…÷I�³÷IL�³÷IQº÷I$Qº÷I¹ øI�¹ øI£øI£øIßxùIàßxùIeúIäeúIéûIÄéûI¹ØûI¹ØûIaëûI €} 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·l‡ïõIöIÐöI¹DöIL¹DöILöITLöIŸUöIПUöIÕXöIœÕXöIçmöI(çmöI…®öI¤…®öIçµöI¬çµöIq¿öI(q¿öI§ÂöIô§ÂöI¹×öI€¹×öIW÷IüW÷I¹÷I ¹÷IC)÷I€ C)÷Iy,÷IL y,÷I»Y÷IØ »Y÷Iÿ`÷I´ ÿ`÷I_…÷I8 _…÷I¡²÷IX¡²÷I¹¸÷I4¹¸÷IG øI G øI£øI,£øIñxùIðñxùI+eúIô+eúI1ûIÔ1ûIÙûIÙûI©ëûI€} 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·l‡ïõI‡öIЇöI%EöIL%EöILöITLöI�UöIÐ�UöIÃXöIœÃXöISnöI(SnöIñ®öI¤ñ®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI%ØöI€%ØöIÃ÷IüÃ÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Iù`÷I´ ù`÷IY…÷I0 Y…÷I›²÷IP›²÷I¹¸÷I,¹¸÷I¥øI˜¥øI£øI$£øIßxùIèßxùIeúIìeúIÝûIÌÝûI­ØûI­ØûIUëûI€} 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·l‡ïõIöIÐöI³DöIL³DöILöITLöIŸUöIПUöIÕXöIœÕXöIçmöI(çmöI…®öI¤…®öIçµöI¬çµöIq¿öI(q¿öI§ÂöIô§ÂöI¹×öI€¹×öIW÷IüW÷I¹÷I ¹÷IC)÷I€ C)÷Iy,÷IL y,÷I·Z÷IØ ·Z÷Iç`÷I° ç`÷IG…÷I, G…÷I…³÷IL…³÷IWº÷I$Wº÷IÝ øI�Ý øI£øI£øIÙxùIàÙxùIeúIäeúI¹ûIĹûI‰ØûI‰ØûI1ëûI €} 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·l‡ïõI‡öIЇöI%EöIL%EöILöITLöI�UöIÐ�UöIÃXöIœÃXöIYnöI(YnöI÷®öI¤÷®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI+ØöI€+ØöIÉ÷IüÉ÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Iá`÷I´ á`÷IA…÷I0 A…÷Iƒ²÷IPƒ²÷Ik»÷I,k»÷IYøI˜YøIõ¢øI$õ¢øIÍxùIèÍxùIeúIìeúIwûIÌwûIGØûIGØûIïêûI€} 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·l�ïõI{öIÐ{öIEöILEöILöITLöI�UöIÐ�UöIÃXöIœÃXöIMnöI(MnöIë®öI¤ë®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöIØöI€ØöI½÷Iü½÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I¿¸÷I,¿¸÷IYøI˜YøI£øI$£øIñxùIèñxùI+eúIì+eúI%ûIÌ%ûIõØûIõØûI�ëûI€} 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·l�ïõIöIÐöI¹DöIL¹DöILöITLöI�UöIÐ�UöIÃXöIœÃXöIímöI(ímöI‹®öI¤‹®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI¿×öI€¿×öI]÷Iü]÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I¥Z÷IØ ¥Z÷I a÷I°  a÷Ik…÷I, k…÷I©³÷IL©³÷Icº÷I$cº÷IC øI�C øI£øI£øIñxùIàñxùI+eúIä+eúIûIÄûIãØûIãØûI‹ëûI €} 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·l�ïõI�öIÐ�öIEöILEöILöITLöI�UöIÐ�UöIÃXöIœÃXöISnöI(SnöIñ®öI¤ñ®öIÕµöI¬ÕµöI_¿öI(_¿öI•ÂöIô•ÂöI%ØöI€%ØöIÃ÷IüÃ÷I§÷I §÷I1)÷I€ 1)÷Ig,÷IL g,÷I©Y÷IØ ©Y÷Ia÷I´ a÷Ie…÷I0 e…÷I§²÷IP§²÷I}»÷I,}»÷IøI˜øI £øI$ £øIåxùIèåxùIeúIìeúIãûIÌãûI³ØûI³ØûI[ëûI€} 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·l“ïõI!öIÐ!öI¿DöIL¿DöI LöIT LöI“UöIГUöIÉXöIœÉXöIómöI(ómöI‘®öI¤‘®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIÅ×öI€Å×öIc÷Iüc÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Iq…÷I8 q…÷I³²÷IX³²÷I¿¸÷I4¿¸÷Iû øI û øI1£øI,1£øI yùIð yùICeúIôCeúI—ûIÔ—ûIgÙûIgÙûIìûI€} 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·l“ïõIuöIÐuöIEöILEöI LöIT LöI“UöIГUöIÉXöIœÉXöIGnöI(GnöIå®öI¤å®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIØöI€ØöI·÷Iü·÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I¿¸÷I,¿¸÷I øI˜ øI%£øI$%£øIýxùIèýxùI7eúIì7eúIUûIÌUûI%ÙûI%ÙûIÍëûI€} 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·l“ïõI!öIÐ!öI¿DöIL¿DöI LöIT LöI“UöIГUöIÉXöIœÉXöIómöI(ómöI‘®öI¤‘®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIÅ×öI€Å×öIc÷Iüc÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I«Z÷IØ «Z÷Ia÷I° a÷Iq…÷I, q…÷I¯³÷IL¯³÷Iuº÷I$uº÷I© øI�© øI£øI£øI÷xùIà÷xùI1eúIä1eúI%ûIÄ%ûIõØûIõØûI�ëûI €} 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·l“ïõI{öIÐ{öIEöILEöI LöIT LöI“UöIГUöIÉXöIœÉXöIMnöI(MnöIë®öI¤ë®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIØöI€ØöI½÷Iü½÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Iq…÷I0 q…÷I³²÷IP³²÷I‰»÷I,‰»÷IÙøI˜ÙøI£øI$£øIëxùIèëxùI%eúIì%eúIûûIÌûûIËØûIËØûIsëûI€} 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·l“ïõI'öIÐ'öIÅDöILÅDöI LöIT LöI“UöIГUöIÉXöIœÉXöIùmöI(ùmöI—®öI¤—®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIË×öI€Ë×öIi÷Iüi÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Iq…÷I8 q…÷I³²÷IX³²÷IŸ÷I4Ÿ÷I[ øI [ øI+£øI,+£øIyùIðyùI=eúIô=eúIaûIÔaûI1ÙûI1ÙûIÙëûI€} 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·l“ïõIoöIÐoöI EöIL EöI LöIT LöI“UöIГUöIÉXöIœÉXöIAnöI(AnöIß®öI¤ß®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIØöI€ØöI±÷Iü±÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷IŸ÷I,Ÿ÷IÁøI˜ÁøI1£øI$1£øI yùIè yùICeúIìCeúIgûIÌgûI7ÙûI7ÙûIßëûI€} 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·l“ïõI'öIÐ'öIÅDöILÅDöI LöIT LöI“UöIГUöIÉXöIœÉXöIùmöI(ùmöI—®öI¤—®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIË×öI€Ë×öIi÷Iüi÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I«Z÷IØ «Z÷Ia÷I° a÷Iq…÷I, q…÷I¯³÷IL¯³÷I�º÷I$�º÷IøI�øI£øI£øIñxùIàñxùI+eúIä+eúIûûIÄûûIËØûIËØûIsëûI €} 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$O%O'Ä'Ä)9*9,®,®.#/#1˜1˜3 4 6‚6‚8÷8÷:l;l=á=á?V@VBËBËD@E@GµGµI*J*L�L�NOQƒQƒSöSöUiViXÜXÜZO[O]Â]Â_5`5b¨b¨degŽgŽijltltnçnçpZqZsÍsÍu@v@x³x³z&{&}™}™ € ‚‚„ò„ò†e‡e‰؉؋KŒK޾޾�1‘1“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations1·l“ïõIuöIÐuöIEöILEöI LöIT LöI“UöIГUöIÉXöIœÉXöIAnöI(AnöIß®öI¤ß®öIÛµöI¬ÛµöIe¿öI(e¿öI›ÂöIô›ÂöIØöI€ØöI±÷Iü±÷I­÷I ­÷I7)÷I€ 7)÷Im,÷IL m,÷I¯Y÷IØ ¯Y÷Ia÷I´ a÷Iw…÷I0 w…÷I¹²÷IP¹²÷I›»÷I,›»÷I“øI˜“øI£øI$£øIñxùIèñxùI+eúIì+eúIïûIÌïûI¿ØûI¿ØûIgëûI€} 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$n%n'â'â)V*V,Ê,Ê.>/>1²1²3&4&6š6š89;‚;‚=ö=ö?j@jBÞBÞDRERGÆGÆI:J:L¬L¬NOQ�Q�STVtVtXæXæZX[X]Ê]Ê_<`v>x°x°z"{"}”}”€‚x‚x„ê„ê†\‡\‰Ή΋@Œ@޲޲�$‘$“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LŸïõIEöI�EöIãDöIãDöI IöI$ IöInöI�nöI¯®öI¯®öIÛ²öI$Û²öIã×öI�ã×öI�÷I�÷I­÷I$­÷Im_÷I�m_÷I̓÷Iü̓÷I…¶÷I …¶÷IQøI| QøI=£øIÜ =£øIyùI  yùIOeúI¤ OeúIUûI„ UûI%ÙûIÈ%ÙûIÍëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LŸïõIiöI�iöIEöI EöI—KöI—KöI;nöI€;nöIÙ®öIüÙ®öIiµöIiµöI ØöIp ØöI«÷Iì«÷I;÷Iô;÷Im_÷I`m_÷I̓÷IÌ̓÷I…¶÷Iì…¶÷IQøIL QøI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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$V%V'Ë'Ë)@*@,µ,µ.*/*1Ÿ1Ÿ346‰6‰8þ8þ:s;s=æ=æ?Y@YBÌBÌD?E?G²G²I%J%L˜L˜N O Q~Q~SñSñUdVdX×X×ZJ[J]½]½_0`0b£b£deg‰g‰iüiükolonânâpUqUsÈsÈu;v;x®x®z!{!}”}”€‚z‚z„í„í†^‡^‰ЉЋBŒBŽ´Ž´�&‘&“ b’)wíoperation.6ipu_stack_gemm_f16_init_large_rows_k768_c80_r7_r85:5z9299Z=âyz}"}z�*�z…2…b‰reasonOperatorKernelvalue10kernelSpecPlanned(Gemm { multiply: F16, accumulate: F32, mode: Initialize, weights: Standard, inner_block: 768, output_columns: 80 })outputElements640input0Elements6144input1Elements52488invocations1 bR%wíoperation.6ipu_stack_attention_merge_large_query_f165:5z9299Z=2}z�"�z…"…z‰*‰b�reasonOperatorKernelvalue10kernelSpecPlanned(AttentionMerge { value_dimension: 72, padded_value_dimension: 80, key_block_columns: 768, initial: true, final_block: true })outputElements576input0Elements640input1Elements6272invocations17LŸïõI?öI�?öIÝDöIÝDöI IöI$ IöInöI�nöI¯®öI¯®öIÛ²öI$Û²öIã×öI�ã×öI�÷I�÷I­÷I$­÷Im_÷I�m_÷I̓÷Iü̓÷I…¶÷I …¶÷IQøI| QøI%£øIÜ %£øIýxùI  ýxùI7eúI¤ 7eúIÝûI„ ÝûI­ØûIÈ­ØûIUëûIÌ€} 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 })outputElements576input0Elements640input1Elements6272invocations17LŸïõIoöI�oöI EöI  EöI—KöI—KöI;nöI€;nöIÙ®öIüÙ®öIiµöIiµöI ØöIp ØöI«÷Iì«÷I;÷Iô;÷Im_÷I`m_÷I̓÷IÌ̓÷I…¶÷Iì…¶÷IQøIL QøI£øI¬ £øIëxùIp ëxùI%eúIt %eúI•ûIT •ûIeØûI˜eØûI ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI9öI�9öI×DöI ×DöI IöI IöI nöI€ nöI©®öIü©®öIÛ²öIÛ²öIÝ×öIpÝ×öI{÷Iì{÷I­÷Iô­÷Ig_÷I`g_÷Iǃ÷IÌǃ÷I…¶÷Iì…¶÷IQøIL QøIC£øI¬ C£øIyùIp yùIUeúIt UeúI�ûIT �ûImÙûI˜mÙûIìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIoöI�oöI EöI  EöI©KöI©KöIAnöI€AnöIß®öIüß®öI{µöI{µöIØöIpØöI±÷Iì±÷IM÷IôM÷Ig_÷I`g_÷Iǃ÷IÌǃ÷I…¶÷Iì…¶÷IQøIL QøI+£øI¬ +£øIyùIp yùI=eúIt =eúI7ûIT 7ûIÙûI˜ÙûI¯ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõI9öI�9öI×DöI ×DöI IöI IöInöI€nöI£®öIü£®öIÛ²öIÛ²öI××öIp××öIu÷Iìu÷I­÷Iô­÷Ig_÷I`g_÷Iǃ÷IÌǃ÷I…¶÷Iì…¶÷IQøIL QøI+£øI¬ +£øIyùIp yùI=eúIt =eúI+ûIT +ûIûØûI˜ûØûI£ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L™ïõIuöI�uöIEöI EöI©KöI©KöIAnöI€AnöIß®öIüß®öI{µöI{µöIØöIpØöI±÷Iì±÷IM÷IôM÷Ig_÷I`g_÷Iǃ÷IÌǃ÷I…¶÷Iì…¶÷IQøIL QøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIõûIT õûIÅØûI˜ÅØûImëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI3öI�3öIÑDöI ÑDöIIöIIöInöI€nöI£®öIü£®öIá²öIá²öI××öIp××öIu÷Iìu÷I³÷Iô³÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøIC£øI¬ C£øIyùIp yùIUeúIt UeúI©ûIT ©ûIyÙûI˜yÙûI!ìûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIuöI�uöIEöI EöIyKöIyKöIGnöI€GnöIå®öIüå®öIKµöIKµöIØöIpØöI·÷Iì·÷I÷Iô÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI1£øI¬ 1£øI yùIp  yùICeúIt CeúI[ûIT [ûI+ÙûI˜+ÙûIÓëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI3öI�3öIÑDöI ÑDöIIöIIöIÿmöI€ÿmöI�®öIü�®öIá²öIá²öIÑ×öIpÑ×öIo÷Iìo÷I³÷Iô³÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI+£øI¬ +£øIyùIp yùI=eúIt =eúI1ûIT 1ûIÙûI˜ÙûI©ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõIuöI�uöIEöI EöIyKöIyKöIGnöI€GnöIå®öIüå®öIKµöIKµöIØöIpØöI·÷Iì·÷I÷Iô÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷I¶÷Iì¶÷IKøIL KøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIûIT ûIÑØûI˜ÑØûIyëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI-öI�-öIËDöI ËDöIýHöIýHöIÿmöI€ÿmöI�®öIü�®öIϲöIϲöIÑ×öIpÑ×öIo÷Iìo÷I¡÷Iô¡÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷Iy¶÷Iìy¶÷IEøIL EøI1£øI¬ 1£øI yùIp  yùICeúIt CeúIgûIT gûI7ÙûI˜7ÙûIßëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI{öI�{öIEöI EöI©KöI©KöIMnöI€MnöIë®öIüë®öI{µöI{µöIØöIpØöI½÷Iì½÷IM÷IôM÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷Iy¶÷Iìy¶÷IEøIL EøI+£øI¬ +£øIyùIp yùI=eúIt =eúIaûIT aûI1ÙûI˜1ÙûIÙëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI'öI�'öIÅDöI ÅDöIýHöIýHöIùmöI€ùmöI—®öIü—®öIϲöIϲöIË×öIpË×öIi÷Iìi÷I¡÷Iô¡÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIñxùIp ñxùI+eúIt +eúIûûIT ûûIËØûI˜ËØûIsëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L“ïõI{öI�{öIEöI EöI©KöI©KöIMnöI€MnöIë®öIüë®öI{µöI{µöIØöIpØöI½÷Iì½÷IM÷IôM÷Ia_÷I`a_÷IÁƒ÷IÌÁƒ÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIëxùIp ëxùI%eúIt %eúIéûIT éûI¹ØûI˜¹ØûIaëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI'öI�'öIÅDöI ÅDöIIöIIöIùmöI€ùmöI—®öIü—®öIá²öIá²öIË×öIpË×öIi÷Iìi÷I³÷Iô³÷Ií`÷I`í`÷IM…÷IÌM…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúI=ûIT =ûI ÙûI˜ ÙûIµëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI�öI��öIEöI EöIsKöIsKöIMnöI€MnöIë®öIüë®öIEµöIEµöIØöIpØöI½÷Iì½÷I÷Iô÷Iç`÷I`ç`÷IG…÷IÌG…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIßxùIp ßxùIeúIt eúIÝûIT ÝûI­ØûI˜­ØûIUëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L�ïõI!öI�!öI¿DöI ¿DöI IöI IöIómöI€ómöI‘®öIü‘®öIÛ²öIÛ²öIÅ×öIpÅ×öIc÷Iìc÷I­÷Iô­÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Iy¶÷Iìy¶÷IEøIL EøI£øI¬ £øIßxùIp ßxùIeúIt eúIÅûIT ÅûI•ØûI˜•ØûI=ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI!öI�!öI¿DöI ¿DöI÷HöI÷HöIímöI€ímöI‹®öIü‹®öIɲöIɲöI¿×öIp¿×öI]÷Iì]÷I›÷Iô›÷Ií`÷I`í`÷IM…÷IÌM…÷Is¶÷Iìs¶÷I?øIL ?øI+£øI¬ +£øIyùIp yùI=eúIt =eúI…ûIT …ûIUÙûI˜UÙûIýëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI‡öI�‡öI%EöI %EöI�KöI�KöISnöI€SnöIñ®öIüñ®öIoµöIoµöI%ØöIp%ØöIÃ÷IìÃ÷IA÷IôA÷Iç`÷I`ç`÷IG…÷IÌG…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõIöI�öI¹DöI ¹DöI÷HöI÷HöIímöI€ímöI‹®öIü‹®öIɲöIɲöI¿×öIp¿×öI]÷Iì]÷I›÷Iô›÷Ia÷I`a÷Ie…÷IÌe…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIãØûI˜ãØûI‹ëûIœ€} 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 })outputElements576input0Elements640input1Elements6272invocations17L‡ïõI‡öI�‡öI%EöI %EöI�KöI�KöIYnöI€YnöI÷®öIü÷®öIoµöIoµöI+ØöIp+ØöIÉ÷IìÉ÷IA÷IôA÷Iÿ`÷I`ÿ`÷I_…÷IÌ_…÷Is¶÷Iìs¶÷I?øIL ?øI£øI¬ £øIßxùIp ßxùIeúIt eúI×ûIT ×ûI§ØûI˜§ØûIOëûIœ€} 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‡ïõIöI�öI¹DöI ¹DöIIöIIöIçmöI€çmöI…®öIü…®öIÕ²öIÕ²öI¹×öIp¹×öIW÷IìW÷I§÷Iô§÷Ió`÷I`ó`÷IS…÷IÌS…÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIýxùIp ýxùI7eúIt 7eúI‘ûIT ‘ûIaÙûI˜aÙûI ìûIœ€} 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‡ïõI‡öI�‡öI%EöI %EöIgKöIgKöIYnöI€YnöI÷®öIü÷®öI9µöI9µöI+ØöIp+ØöIÉ÷IìÉ÷I ÷Iô ÷Ií`÷I`í`÷IM…÷IÌM…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øI÷xùIp ÷xùI1eúIt 1eúIIûIT IûIÙûI˜ÙûIÁëûIœ€} 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‡ïõIöI�öI³DöI ³DöIIöIIöIçmöI€çmöI…®öIü…®öIÕ²öIÕ²öI¹×öIp¹×öIW÷IìW÷I§÷Iô§÷Ia÷I`a÷Iq…÷IÌq…÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIëxùIp ëxùI%eúIt %eúIûIT ûIéØûI˜éØûI‘ëûIœ€} 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‡ïõI�öI��öI+EöI +EöIgKöIgKöIYnöI€YnöI÷®öIü÷®öI9µöI9µöI+ØöIp+ØöIÉ÷IìÉ÷I ÷Iô ÷Ia÷I`a÷Ie…÷IÌe…÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øIåxùIp åxùIeúIt eúIïûIT ïûI¿ØûI˜¿ØûIgëûIœ€} 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�ïõIöI�öI­DöI ­DöIñHöIñHöIámöI€ámöI®öIü®öIòöIòöI³×öIp³×öIQ÷IìQ÷I•÷Iô•÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI£øI¬ £øIGwùIp GwùI%GúIt %GúIUûIT UûIK×ûI˜K×ûIóéûIœ€} 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�ïõI�öI��öI+EöI +EöI‘KöI‘KöI_nöI€_nöIý®öIüý®öIcµöIcµöI1ØöIp1ØöIÏ÷IìÏ÷I5÷Iô5÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI%£øI¬ %£øIMwùIp MwùI+GúIt +GúI[ûIT [ûIQ×ûI˜Q×ûIùéûIœ€} 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�ïõIöI�öI­DöI ­DöIñHöIñHöIÛmöI€ÛmöIy®öIüy®öIòöIòöI­×öIp­×öIK÷IìK÷I•÷Iô•÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øI5wùIp 5wùIGúIt GúIïûIT ïûIåÖûI˜åÖûI�éûIœ€} 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�ïõI“öI�“öI1EöI 1EöI—KöI—KöI_nöI€_nöIý®öIüý®öIiµöIiµöI1ØöIp1ØöIÏ÷IìÏ÷I;÷Iô;÷I+_÷I`+_÷I‹ƒ÷IÌ‹ƒ÷Im¶÷Iìm¶÷I9øIL 9øI £øI¬  £øI5wùIp 5wùIGúIt GúIãûIT ãûIÙÖûI˜ÙÖûI�éûIœ€} 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