Ei'WSrSSKrSSKrSSKrSSKrSSKrSSKrSSKrSSKrSSK J r J r SSK J r SSKJrJrJr SSKrSSKJr SSKJrJrJrJrJrJrJrJrJrJrJ r J!r!J"r"J#r#J$r$ SSK%J&r& SSK'J(r( S S K)J*r* S S K+J,r,J-r-J.r. S S K/J0r0 \Rb"\25r3\ "S 5r4\4Rk5r6S\R(RnS\ \S\ \R(Rn\8\/\R(Rn4S\94Sjr:"SS5r;S\,S\\<S\;4Sjr=S6SSSS.S\R(RnS\ \S\\<S\9S\9S\\<S\S6S\R(RnS\ \S\\<S\9SS4 S!jjr?S6S\R(RnS\ \S\\<S\9SS4 S"jjr@S\R(RnS\ \S\\<SS4S#jrA\.S\RnS\ \S\\<S\Rn4S$j5rB\.S\RnS\ \S\\<S\Rn4S%j5rCS\RnS\ \S\,S&\ \S\94 S'jrDS(\S)\RRS*\S\8\4S+jrGS(\S)\RRS*\SS4S,jrHS(\S)\RRS*\SS4S-jrISS.SSS/S0.S)\RRS*\S\.add_pathsisW$&GGLLAW$X!8'*xx'+C,=,=>LLMN(C$r,zXAccuracy failed for the TorchDynamo produced graph. Creating script to minify the error.zBad accuracy detected.DCompiled Fx GraphModule failed. Creating script to minify the error.r1r)r5partialr7r repro_after Exception repro_leveldump_to_minify_after_dynamor8copydeepcopyr+logwarningfx GraphModulegraphr rdump_backend_state) r:r r!r=r"rK compiled_gmrA dump_state_fns r*__call__WrapBackendDebug.__call__as  ''(F(FQ&Q !!%<<<<    ) y T !!Q&+B@S@ST!!Q&)$--*;^L "2{CCKKr0r4==+BC&++ ((@ACcNIDL3"-dmmB.?"PK%kB.&b9K-!KK^))Q.(1(9(9.d>Q>Q) &NN2t}}RXX/FG ++q03NN2t}}RXX/FG* // cN%s>(E22 I1>& ! ! FT* VyRXX^^1D0EY%OPIKK0#bhhnn2Ey i   +-&    77<< 3LKK2ILx&,,. OOI,  s F F%cT[R"U5(de[XX#5$)z Dumps the dynamo graph to repro the issue. 1) It tries to convert Fx GraphModule to a string. If we can, it writes to a repro.py file. 2) If we can't convert Fx GraphModule to a string, we use to_folder to save the module and save a tar file. )rcan_convert_to_stringr)r rqr1rrs r*r^r^ s(  1 1" 5 55 5 %b NNr,c [RR[5S5n[RR U5(d[R "USS9 [ [UUU[RS:HUSS95 g)NrTrrNminify)rrrorp) rDrErFrrrrrrrU)r rqr1rs r*rVrV$shWW\\,.- 8F 77>>& ! ! FT*'   !--2   r,c ZSSKJn [U5nUVs/sH:n[U[R 5(aUR ROUPM< nnU"X5n[Xa5 [S5es snf![an[U5n[RS5 [R"[ US9n U "["R$"U[&R("UR*55U5 [R"[,UUS9n U"UUU U S9 SnAU$SnAff=f)NrminifierzNo issue was detectedrOrQ)r" orig_failure module_fails dump_state)functorch.compilerrrxrgrznodehintr ValueErrorrTrfrYrZr5rRr^r[r\rWrXr] backend_fails) r r!r1rr"ir_rArr`fails_fns r*dynamo_minifier_backendr<s+ /K DRCQaz!U\\22 9> !"5 +6011  3x  R ")) m  bnnRrxx)@A>R$$ #%    !$  I) sAA8A== D*BD%%D*cSSKJn [U5nUR5 [ XU5(a[ R S5 [R"[USS9n[R"[US9nU"[R"U[R"UR55U5 U"UUUUS9 U$[ RS5 U$) Nrrz2Accuracy failed for the TorchDynamo produced graphT)r1rr)r"rz*Input graph does not fail accuracy testing)rrrevalr+rYrZr5rRr^r[r\rWrXr]error)r r!r1rr"r`rs r* dynamo_accuracy_minifier_backendrfs+ /KGGIr;77 HI!)) mD $$ #  bnnRrxx)@A>R  !$   I >? Ir,rcSSKJn [U[U55 U"X5n[U[U55 g![a5n[ U5nU"SX75R 5S:aSnAgSnAgSnAff=f)a  Minifier uses this function to identify if the minified graph module fails with the same error. One caveat is that minifier can potentially go into a wrong direction when the resulting graph module fails for a different reason. To avoid this, we save the string for the original exception and check similarity between new and old exception. They can be somewhat different in some cases, when the exception string depends on the failing node information. So, we have a loose similarity metric to guide the minifier path. r)SequenceMatcherNg?TF)difflibrrrrTrfratio)r r!r"rrr_e new_failures r*rrsz"("=nMN!"5 +'F~'VW  !f 4 ; A A Cc I J  s2; A:&A55A:optionsmodrc[US5(d[RS5 O0URS:a [RSUR5 [ 5nU"U5 [ SUR S9n[URUS9nU"U5 URnSSS5 U$!,(df  W$=f)N_versionzzload_args does not have a _version attribute, please file a bug to PyTorch and describe how you generate this repro scriptrzload_args is version %s, but this version of PyTorch only supports version 0. We will try to run it anyway but there may be an incompatibility; if so, try upgrading your version of PyTorch.zLoading inputs)desctotal)ropbar) r9rYrZrrrrrrorq)rrr nop_readerr input_readerrqs r* run_load_argsrs 9j ) )  >    ! KK@""   !J j #:+;+; <"G,<,<4H ,   = K = < Ks >)B11 Cc[XU5nUR(d [S5nO [S5nURc [ S5e[ R "UURS9n[RRU5"U5n[RRSURS9 U"U6 SSS5 g!,(df  g=f)NrrzCompiler name is None - this likely means that a custom compiler was called by torchdynamo. Please remove this error, import your custom compiler function, and replace the backend=None line in run_repro to backend=rQcudaenabled) raccuracyrbackend RuntimeErrorr5rRrg_dynamooptimizeampautocast)rrrrqr"ropt_mods r* repro_minifyrs y 1D   $%>? $%GH  I  (//oo mm$$%<=cBG   FG,<,<  = > = =s 9C Cc [RRUR5"U5nURS:waUR 5 UR 5 [R RSURS9 [XU5n[XU5(dS5e[UUU[R[RS9(d [S5eSSS5 g[R RSURS9 [XU5n[XURSS9 A[XU5n[UUURSS9 SSS5 g!,(df  g=f!,(df  g=f) NrrzEager itself failedr%z Dynamo failedT)r& disable_clone)rgrrrrrrrrrrr(r)r rr&)rrrrrqs r* repro_runrsCmm$$W__5c:G2    YY  0@0@  A y9D"3T22 I4I I2"22$88 $O44 B AYY  0@0@  A y9D c'2B2BRV W y9D  ))"   B AB AB AsAE&AE7& E47 Frinductor)rprrorrrrrr=c |^^^^UHn[RSU5 M TSLaSmOTSLaSm[R"SUSUST<S T<S 3 [RS 9n S [RS S4UUUU4Sjjn U R SSSS9n U R SSS9n U "U 5 U RSSSS9 U R SSS9n U "U 5 Sn[[R5S::aU/[RSSQnU RU5n[[S.nUUR"XU5 g)NzPUnrecognized kwarg %s; perhaps this repro was made on a newer version of PyTorchTrFrzAn after_dynamo repro script, typically triggering a bug in Dynamo or AOTAutograd. When run with no arguments, this script defaults to running 'z9'. Extra flags may be available; to find out more, try 'zs --help'. There are also alternate subcommands available, see below. default settings on this script: accuracy=z save_dir=ru) descriptionformatter_classparserr#c d>UR5nURSSSSTSS9 URSSSTSS 9 URS [TS S S 9 URSSSSSS9 URSSSSSS9 URSTSSS9 URSSSSS9 URS [TS!S"S 9 g)#Nz --no-accuracyr store_constrz>do not test accuracy, just run the module and see if it errors)destactionconstdefaulthelpz --accuracyz test accuracy)rrrrz --save-dirDIRz!directory where saved inputs live)typermetavarrz --no-save-dirroz(don't use any directory for saved inputs)rrrrz --no-isolateisolate store_falseFz1no isolate (doesn't do anything for after_dynamo))rrrrz --autocast store_truezuse torch.cuda.amp.autocast)rrrz --no-autocastrz!don't use torch.cuda.amp.autocast)rrrz --backendBACKENDztorch.compile backend to use)add_mutually_exclusive_group add_argumentrf)raccuracy_grouprrrros r* common_flagsrun_repro..common_flags(s,<<>##  Q $  ##   $   4     ;     D    .    4   /  r,rpz {run,minify})rrrequiredrmzjust run the repro)rz --only-fwdrz+don't run backwards compilation for testing)rrrzrun the minifier on the reprorP)rrm)rYrZargparseArgumentParserRawTextHelpFormatteradd_subparsers add_parserrrsysargv parse_argsrrrp)rrrprrorrr=krr subparsers parser_run parser_minifyrqr COMMAND_FNSs ```` r* run_reprors ^   4 U   $ $  CG9M  +  + !55 F8 X448 8 8 t&& 'J&& !'J : ))6*M D 388}'#((12,'%GK y9r,)F)K__doc__rrWr5loggingrDrrrcollections.abcrr importlibrtypingrrrrgtorch.fxr[torch._dynamo.debug_utilsr r r r r rrrrrrrrrr%torch.fx.experimental.symbolic_shapesr torch.hubrrrbackends.registryrrr debug_utilsr getLoggerr3rYinductor_config is_fbcoderHr\rhboolr+r.rfrlrrr^rVrrrnnModulerrrrrjr,r*rsE$  .#'' "IRR9! 89  $ $ &   SM 588//c;UXX=Q=QQR  KK\ E( E9A# E E.! 5  "5 5 3-5C=5 5  5sm55 5x! - - 3--C=- -  -J! O O 3-OC=O O  O. $,SMBJ3- 0& &(0 &FNsm&^^&&R (0 FNsm^^@ SM3-   H3UXX__c4#EHHOO8 s  S T N!#"{: {:{: {: D#I {: sm {:{:{:{: {:r,