CCjgSSKJr SSKrSSKrSSKrSSKrSSKrSSKrSSKrSSK J r J r SSK J r SSKJr SSKJrJrJrJr SSKJrJrJr SSKrSSKJr S S KJr S S KJr \(a SS K J!r! SS KJ"r" S SK#J$r$ S SK%J&r& S SKJ'r'J(r( SSjr)\RT"\+5r,\RZR]\+S5r/\0"S5r1SSjr2"SS\35r4SSjr5SSjr6SSjr7"SS\35r8"SS\85r9"SS\85r:"S S!\85r;"S"S#\;5r<"S$S%\;5r="S&S'\;5r>"S(S)\;5r?"S*S+\;5r@"S,S-\;5rA"S.S/\85rB"S0S1\85rC"S2S3\85rD"S4S5\85rE"S6S7\85rF"S8S9\F5rG"S:S;\85rH"S<S=\85rI"S>S?\85rJ"S@SA\85rK"SBSC\85rL"SDSE\ 5rM"SFSG\85rN"SHSI\85rO"SJSK\85rP"SLSM\85rQ"SNSO\R5rS"SPSQ\85rT"SRSS\85rU"STSU\U5rV"SVSW\U5rW"SXSY\W5rX"SZS[\W5rY"S\S]\U5rZ"S^S_\U5r["S`Sa\U5r\"SbSc\U5r]"SdSe\U5r^"SfSg\U5r_\`\V\a\W\b\X\c\Z\d\Y\e\[\3\\\f\]\g\^0 rh"ShSi\f5ri"SjSk\35rjSSljrkSSSm.SSnjjrlSSojrmSSpjrnSSqjro\RRR\RRR\RRR\RRR\RRR4rwSSrjrxSSsjry\ "S St9SSuj5rzSSvjr{\|"5r}\}SwSwSx.SSyjjr~"SzS{5rSS|jrSSS}jjrSS~jrSSjrSSjrSSSjjrSSjrSSjrSSjrSSSjjrSSjrSSjrSSSjjrg)) annotationsN)autoEnum) lru_cache)Path) format_exc format_list FrameSummary StackSummary)AnyNoReturn TYPE_CHECKING)get_file_path_2)config)counters)VariableTracker) CompileId)DynamoTracerOutput)InstructionTranslatorBase)DynamoFrameTypeFrameExecStrategyc,SURSS5-$)NzhFor more information about this error, see: https://pytorch.org/docs/main/generated/exportdb/index.html#_-)replace) case_names ڟC:\Users\johnn\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages\torch/_dynamo/exc.pyexportdb_error_messagerTs" I   C % & graph_breaks)_torch_dynamo_tracer_outputfirst_useful_frameframe_exec_strategycU$Nargss r_safe_exception_argsr*is Kr c8^\rSrSr%S\S'SU4SjjrSrU=r$)_SerializedExceptionmstr_dynamo_original_exception_typec>[TU][U55 [U5RS[U5R 3Ulg)N.)super__init__r.type __module__ __qualname__r/)selfexc __class__s rr3_SerializedException.__init__ps@ S"Cy##$Ad3i&<&<%= > ,r )r/)r8 BaseExceptionreturnNone__name__r5r6__firstlineno____annotations__r3__static_attributes__ __classcell__r9s@rr,r,ms%((  r r,c[U5$r&)r,)r8s r_safe_inner_exceptionrFws  $$r c0nUR5Hbup#U[;d[U[R5(aSX'M4US:Xa$[U[ 5(a[ U5X'M^X1U'Md U$)Ninner_exception)items_EXCEPTION_STATE_ATTRS_TO_DROP isinstancetypes FrameTyper;rF)stateresultnamevalues r_safe_exception_staterR{shF{{}  1 1Zu5W5WFL & &:e]+K+K07FL 4L % Mr TorchDynamoExceptioncVURU5n[R"U/UQ76 U$r&)__new__r;r3)exc_typer)r8s r#_reconstruct_torch_dynamo_exceptionrWs,   8 $C3&& Jr cN^\rSrSrSrSU4SjjrS SjrS SjrS SjrSr U=r $) rSaBase exception class for all TorchDynamo-specific exceptions. Attributes: _torch_dynamo_tracer_output: Optional tracer output attached to the exception frame_exec_strategy: Optional frame execution strategy to control how convert_frame should handle this exception. When set, convert_frame will use this strategy instead of the default behavior. This allows exceptions to signal specific execution strategies (e.g., SKIP, RUN_ONLY) without requiring separate exception types for control flow. cB>[TU]"U0UD6 SUlSUlgr&)r2r3r"r$)r7r)kwargsr9s rr3TorchDynamoException.__init__s% $)&)FJ(=A r cl[[U5[UR54UR 54$r&)rWr4r*r) __getstate__r7s r __reduce__TorchDynamoException.__reduce__s1 / $Z-dii8 9      r c,[UR5$r&)rR__dict__r_s rr^!TorchDynamoException.__getstate__s$T]]33r cBUbURRU5 ggr&)rcupdate)r7rNs r __setstate__!TorchDynamoException.__setstate__s   MM  ' r )r"r$)r)r r[r r<r=)r<tuple[Any, ...]r<dict[str, Any])rNzdict[str, Any] | Noner<r=) r?r5r6r@__doc__r3r`r^rgrBrCrDs@rrSrSs" B  4((r c\rSrSrSrg)InternalTorchDynamoErrorr'Nr?r5r6r@rBr'r rrnrnr rnc\rSrSrSrg)ResumePrologueTracingErrorr'Nrpr'r rrsrsrqr rsc@^\rSrSr%S\S'SS.SU4SjjjrSrU=r$)RestartAnalysis str | Nonerestart_reasonN)ryc,>Xl[TU]"U6 gr&)ryr2r3)r7ryr)r9s rr3RestartAnalysis.__init__s, $r )r)r ryrxr<r=r>rDs@rrvrvs@D   r rvc\rSrSrSrg)SpeculationRestartAnalysisr'Nrpr'r rr}r}rqr r}c\rSrSrSrSrg)AutogradGradRestartAnalysiszRaised when autograd.grad consumed grad_fns that are returned. On restart, autograd.grad will graph break instead of being traced. r'Nr?r5r6r@rlrBr'r rrrsr rc\rSrSrSrSrg)RequiresGradRestartAnalysiszRaised when a source-less requires_grad_() intermediate leaks as output. On restart, requires_grad_() will graph break instead of being traced, preserving partial acceleration for code before the call. r'Nrr'r rrrsr rc\rSrSrSrg)UnspecializeRestartAnalysisr'Nrpr'r rrrrqr rc\rSrSrSrg) CompileCollectiveRestartAnalysisr'Nrpr'r rrrrqr rc\rSrSrSrg)TensorifyScalarRestartAnalysisr'Nrpr'r rrrrqr rc\rSrSrSrg) SkipFramer'Nrpr'r rrrrqr rc0^\rSrSrSSU4SjjjrSrU=r$)TorchRuntimeErrorc>[TU]U5 XlUbUUlg[RR R 5Ulgr&r2r3msgtorch_guardsTracingContext extract_stack real_stackr7rrr9s rr3TorchRuntimeError.__init__H %  --;;= r rrr&rr.rStackSummary | Noner<r=r?r5r6r@r3rBrCrDs@rrr   r rc0^\rSrSrSSU4SjjjrSrU=r$)InvalidBackendc>SU<3nUU(a#SSR[[U55S3OS- nUS- n[TU]U5 g)NzInvalid backend: z, did you mean: z, ?r1z< See `torch._dynamo.list_backends()` for available backends.)joinmapreprr2r3)r7rP suggestionsrr9s rr3InvalidBackend.__init__sX!$* tyyT;)?@A C  MM r r'r&)rPr.rzlist[str] | Noner<r=rrDs@rrrs r rc,^\rSrSrSU4SjjrSrU=r$) ResetRequiredcL>[TU][R"S55 g)Nz Must call `torch._dynamo.reset()` before changing backends. Detected two calls to `torch.compile()` with a different backend compiler arguments. )r2r3textwrapdedent)r7r9s rr3ResetRequired.__init__s!  OO  r r'r<r=rrDs@rrrs   r rcF^\rSrSrSU4SjjrSSjrSrU=r$)ShortenTracebacki c2>[TU]"U0UD6 Xlgr&)r2r3r#)r7r#r)r[r9s rr3ShortenTraceback.__init__ s $)&)"4r c&URnURbUb[R(aU$URURLa5UR nUc [ S5eURURLaM5URU5$)Nz#internal error, please report a bug) __traceback__r#rverbosetb_frametb_nextAssertionErrorwith_traceback)r7tbs rremove_dynamo_frames%ShortenTraceback.remove_dynamo_framess{     " " *bjFNNKkk!8!88Bz$%JKKkk!8!88""2&&r r#)r)r r#types.FrameType | Noner[r r<r=)r<z typing.Self)r?r5r6r@r3rrBrCrDs@rrr s255.D5PS5 5 ''r rc<^\rSrSrSU4SjjrSrU=r$)BackendCompilerFailedic>[USS5UlX lSUR<S[U5RSU3n[ TU]XCS9 g)Nr?rzbackend=z raised: z: r)getattr backend_namerHr4r?r2r3)r7 backend_fnrHr#rr9s rr3BackendCompilerFailed.__init__sZ $J C@.**-Z_8M8V8V7WWYZiYjk Dr )rrH)rr rH Exceptionr#rr<r=rrDs@rrrs9 E E# E3 E  E Er rcl^\rSrSrSSSS.S U4SjjjjrS SjrS S SjjrSrU=r$) Unsupportedi-N)rrc>[TU]U5 U(d([RRR 5nXPlXlX0lSUl UR5 X l SUl gNF) r2r3rrrrrr skip_framecategory add_to_statsgb_typelogged)r7rrrrrr9s rr3Unsupported.__init__.s[ 55CCEJ$$$(  #*  r c URc [S5e[URUR==S-ss'[URURS::a[URUR gg)Nz/category must be set before removing from statsrr)rrrrr_s rremove_from_statsUnsupported.remove_from_statsCsi ==  !RS S)Q.) DMM "488 , 1'1 2r cLXl[UUR==S- ss'g)Nr)rrr)r7rs rrUnsupported.add_to_statsJs  488$)$r )rrrrrr)F) rr.rr.rboolrrxrrr<r=r) unimplemented)rr.r<r=) r?r5r6r@r3rrrBrCrDs@rrr-sk  !%*.    ( *2**r rc\rSrSrSrg)$UnknownPropertiesDuringBackwardTraceiOr'Nrpr'r rrrOrqr rc\rSrSrSrg)RecompileErroriSr'Nrpr'r rrrSrqr rc\rSrSrSrg)InfiniteGeneratorErroriWr'Nrpr'r rrrWr rc\rSrSrSrSrg)CondOpArgsMismatchErrori\z7 Internal error from cond() due to arguments mismatch. r'Nrr'r rrr\sr rc\rSrSr\"5r\"5r\"5r\"5r\"5r \"5r \"5r \"5r Sr g) UserErrorTypeibr'N)r?r5r6r@rDYNAMIC_CONTROL_FLOW ANTI_PATTERNSTANDARD_LIBRARYCONSTRAINT_VIOLATION DYNAMIC_DIM INVALID_INPUTINVALID_OUTPUT*UNSUPPORTED_ALIASED_MUTATED_DYNAMIC_INPUTSrBr'r rrrbs>66Lv6&KFMVN15.r rc@^\rSrSrSSU4SjjjrSrU=r$) UserErrorimcz>Ub[[U[5(d[S[U535eUR S5(aUS- nOUS- nU[ U5- n[ TU]U5 [RRR5Ul SUl SUlXlX lX lg)a Type of errors that would be valid in Eager, but not supported in TorchDynamo. The error message should tell user about next actions. error_type: Type of user error msg: Actionable error message case_name: (Optional) Unique name (snake case) for the usage example in exportdb. Nzcase_name must be a str, got r1  F)rKr.rr4endswithrr2r3rrrrrrr error_typermessage)r7rrrr9s rr3UserError.__init__ns  i--$'DT)_DU%VWW||C  s t  ))4 4C --66DDF $ r )rrrrrrr&)rrrr.rrxr<r=rrDs@rrrms0KO'.1>H r rc0^\rSrSrSSU4SjjjrSrU=r$)StepUnsupportedic>[TU]U5 XlU(d([RR R 5nX lSUlgr) r2r3rrrrrrrrs rr3StepUnsupported.__init__s< 55CCEJ$ r )rrrr&rrrDs@rrrs r rc\rSrSrSrg)UnsafeScriptObjectErrorir'Nrpr'r rrrrqr rc0^\rSrSrSSU4SjjjrSrU=r$)UncapturedHigherOrderOpErroric>[TU]U5 XlUbUUlg[RR R 5Ulgr&rrs rr3%UncapturedHigherOrderOpError.__init__rr rr&rrrDs@rrrrr rc\rSrSrSrg)FailOnRecompileLimitHitir'Nrpr'r rrrrqr rc\rSrSrSrg) PackageErrorir'Nrpr'r rr r rqr r cB^\rSrSrSS.SU4SjjjrSrU=r$)ObservedExceptioniNrc>[TU]"U0UD6 UbUUlg[RRR 5Ulgr&)r2r3rrrrrr7rr)r[r9s rr3ObservedException.__init__sJ $)&)%  --;;= r r)r rrr[r r<r=rrDs@rr r s3=A  &9 LO   r r cN^\rSrSr%S\S'SS.SU4SjjjrSrU=r$) ObservedUserStopIterationiz Any | NonerQNr cd>[TU]SUS9 [U5S:a USUlgSUlg)Nzunhandled `raise StopIteration`r r)r2r3lenrQrs rr3"ObservedUserStopIteration.__init__s5 :zR t9q=aDJDJr rQrr>rDs@rrrs9  =A&9LO r rc\rSrSrSrg)ObservedLookupErrorir'Nrpr'r rrrrr rc\rSrSrSrg)ObservedIndexErrorir'Nrpr'r rrrrr rc\rSrSrSrg)ObservedKeyErrorir'Nrpr'r rrrrr rc\rSrSrSrg)ObservedGeneratorExitir'Nrpr'r rrrrqr rc\rSrSrSrg)ObservedAttributeErrorir'Nrpr'r rr!r!rr r!c\rSrSrSrg)ObservedRuntimeErrorir'Nrpr'r rr#r#rr r#c\rSrSrSrg)ObservedNotImplementedErrorir'Nrpr'r rr%r%rqr r%c\rSrSrSrg)ObservedTypeErrorir'Nrpr'r rr'r'rr r'c\rSrSrSrg)FakeTensorObservedExceptionir'Nrpr'r rr)r)rqr r)c\rSrSrSrSrg)UnhandledDescriptorErroriacRaised by object_generic_getattr when a descriptor type is not recognized by _resolve_descriptor_get. Subclasses NotImplementedError so callers that catch NotImplementedError (e.g., generic_getattr's graph-break fallback) still work, but callers that want to distinguish unhandled descriptors from other NotImplementedErrors can catch this specifically. r'Nrr'r rr+r+sr r+c\rSrSrSrSrg)TritonUnavailableErrori a Raised by DeviceInterface.raise_if_triton_unavailable to signal that a device cannot run Triton (e.g. no Triton backend was built for it). Subclasses RuntimeError so existing callers that catch RuntimeError keep working, while callers that only want to react to Triton unavailability - such as has_triton() - can catch this specific type instead of swallowing every RuntimeError, which would hide unrelated bugs. r'Nrr'r rr-r- sr r-cU[;a3[US[U55n[SUS3[405[U'[U$)Nr?ObservedError)observed_exception_maprr.r4r )rVrPs rget_dynamo_observed_exceptionr2sN--xS];+/tfE "%6$8", x( "( ++r )r)r[cSSKJn U(a9UVs/sH+n[U[5(aUR X5OUPM- nnO/nUR X5R XU=(d 05nUR US5 gs snf)Nr)SourcelessBuilder)variables.builderr4rKr.create call_functiondo_raise)rVtxr)r[r4argargs_ exception_vts rraise_observed_exceptionr=#s5  2,unimplemented_with_warning..s /  r c>T$r&r')graph_break_msgsrrOrPs?r ) metadata_fn payload_fnz%sT)rcontext explanationhintsfrom_exc log_warning)format_error_msg_verboser_loggingtrace_structuredgraph_breaks_logdebugr)ecoderrUrVrW_unimplementedrRs @runimplemented_with_warningrblsd /q7O NN## + $41"N r c[R"US5R5nSRSU55n[R"US5R5nUSUSUSU3n[ U5nU(aUSU3- nU$)N rc3r# UH-nS[R"US5R5-v M/ g7f)z Hint: rdN)rindentlstrip).0hints r -format_graph_break_message..s,HM X__T6299;;s57z Explanation: z Developer debug context: z: For more details about this graph break, please visit: )rrfrgrget_gbid_documentation_link)rrUrVrW hints_strrdocumentation_links rformat_graph_break_messageros //+v6==?K HMIoogv.557G  }  #9  (C 5W= MN`Mabb Jr )maxsizec[[5R5Rn[ S[ U5S5n[ U5n[R"U5nSSS5 0nWR5HupVUH nXTUS'M M U$!,(df  N;=f![a [RS5 0nNaf=f)z Loads the gb_type to gb_id map from the graph break registry from JSON file with caching. Includes historical gb_type (mapping behavior of duplicate gb_types with different gb_ids is undefined). rzgraph_break_registry.jsonNz!Error accessing the registry fileGb_type) r__file__resolveparentrr.openjsonloadrlog exceptionrI) script_dir registry_pathfregistrymappingkventrys r_load_gb_type_to_gb_id_maprs (^++-44 ' J!< - Ayy|H!G E()E)$ %! N!  9:s0AB' B!B' B$ B'$B''!C  C cXSn[5nX;aUSX RS5S3$g)z Retrieves the GBID documentation link for a given graph break type. Args: gb_type: The graph break type to look up. Returns: A string containing the documentation URL if found, otherwise None. z;https://meta-pytorch.github.io/compile-graph-break-site/gb/gbGBz.htmlN)rrg)rGRAPH_BREAK_SITE_URLgb_type_to_gb_id_maps rrlrlsJ F67&#$B';'D'K'KD'Q&RRW X  r F)rXrYrc0[XX#5nU(a[RU5 U[LaZSn[ US5(a UR n[ U[5(aURSU3n[XpXhS9e[XpXhS9Ue[XpU5e)a Called within dynamo to cause a graph break. Args: gb_type: Context-free graph break type. It should be a short string without any information specific to the tracing context (i.e. no dynamically-generated strings) context: Developer context for the graph break. It can contain tracing context/dynamic strings. explanation: User-facing context-dependent explanation for the graph break. Can be dynamic. hints: List of user-facing hints for the graph break. NrzJ *** While handling this graph break, another graph break occurred: *** r ) rorywarning_NOTHINGhasattrrrKrr) rrUrVrWrXrYrrpast_real_stacks rrrs( %W{ JC Cx 8\ * *&11O h , ,\\N"pqtpuvCcJS S *   cJ //r c2\rSrSrSSjrSSjrSSjrSrg) KeyErrorMsgi cXlgr&r)r7rQs rr3KeyErrorMsg.__init__ s r c,[UR5$r&)r.rQr_s r__str__KeyErrorMsg.__str__s4::r c"UR5$r&)rr_s r__repr__KeyErrorMsg.__repr__s||~r rN)rQr r<r=)r<r.)r?r5r6r@r3rrrBr'r rrr sr rc[US5(a/URS5nUSSURUSUS3nU$)N _hop_namez Explanation:rz Higher Order Operator: r)r partitionr)r8rliness r!augment_exc_message_with_hop_namersTsK   01Qxj3CMM?58*USTXJ W  Jr cSSKnSUl[U5nUb?[U5S:a0USUlUSSR UR U553- n[ R(a1[US5(a USURSURS3- n[ R(d[US 5(aUS - n[US 5(a[URS 5(am[URS 5(a5USURRSURRS3- nOUSURRS3- n[UR5S:XaSO[URS5n[!X5n[#U[$5(a$['XQ-54URSS-UlgXQ-nU4URSS-Ulg)Nrz from user code: rrecord_filenamez! Last frame execution written to zD. To run only this frame while debugging, run torch._dynamo.replay('z'). rz Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo" rH minifier_path buck_commandz Minifier script written to zW. Run this buck command to find the smallest traced graph which reproduces this error: rzQ. Run this script to find the smallest traced graph which reproduces this error. r) tracebackinnermost_user_frame_summaryget_real_stackrrr rreplay_record_enabledrrrrHrrr)r.rrKKeyErrorr)r8rexportrrold_msgnew_msgs raugment_exc_messager#s'+C$$J#j/A"5+5b>( %bggi.C.CJ.O&P%QRR ##5F(G(G 01D1D0EF++,E 3  >>gc<88  J  s%&&7 _,, 3&& 7 7 /0C0C0Q0Q/RS0030C0C0P0P/QQSU C /0C0C0Q0Q/RS__ C MQ&bC ,> !  enn ,   5>> ) JJ4 B %  r cvSSKJn [R"US3U5nU(aUR S5$U$)Nr)TORCH_DYNAMO_RESUME_IN_PREFIXz _(\w+)_at_\d+)resume_executionrrematchgroup)rPrrs rremove_resume_prefixrs4? HH56oF ME {{1~ Kr c[5nUHknURcM[UR5nU(a&U(aUSRU:Xa X!S'X2lMTX2lUR U5 Mm U$)a When we graph break, we create a resume function and make a regular Python call to it, which gets intercepted by Dynamo. This behavior is normally shown in the traceback, which can be confusing to a user. So we can filter out resume frames for better traceback clarity. Example: File "..." line 3, in f File "..." line 5, in torch_dynamo_resume_in_f_at_80 File "..." line 10, in torch_dynamo_resume_in_f_at_120 becomes File "..." line 10, in f r)r rrrPr)r new_stackrrPs rcollapse_resume_framesrsm(I >> ! #EJJ/ )B-"4"4"<!bMJJ   U # r cSURSURSURS3nUS- nU[5- n[ X5nUb,US- nUSR [ U55- nUS- nUS- nU$) NWON'T CONVERT r line rz.========== TorchDynamo Stack Trace ========== zY ========== The above exception occurred while processing the following code ========== rz ==========)rrco_firstlinenorrrr )r8r`rrrrs rrZrZs a(8(8'9@S@S?TTVW CCC:<C+J    rww{:.// t  x Jr c Z[USS5S[USS5S[USS5S3$) Nrz z (rrrr))r)r`s rformat_frame_inforsF 4K 0 12 D- 5 67.231 6r c8Ub[U5nSUSU3$SU3$)Nz6torch.compile intentionally decided to skip the frame z! and fall back to eager. Reason: zVtorch.compile intentionally decided to skip the frame and fall back to eager. Reason: )r)r`reason frame_infos rformat_skip_frame_messagersB &t, DZLQh  h  r c [R(a [XX#5$SURSURSUR S[ 53$)Nrrrz due to: )rrrZrrrr)r8r`rrs rformat_error_msgrsS ~~'?JJ DLL>4+;+;*<= = 88r )rr.r<r.)r)rir<ri)r8r;r<r,)rNrkr<rk)rVztype[TorchDynamoException]r)rir<rS)rVtype[Exception]r<ztype[ObservedException]) rVrr9rr)z(list[VariableTracker] | list[str] | Noner[z!dict[str, VariableTracker] | Noner<r )r9rrr.r<r )r9r r<r=)r_rr`types.CodeTyperr.rUr.rVr.rW list[str]r<r ) rr.rUr.rVr.rWrr<r.rj)rr.r<rx)rr.rUr.rVr.rWrrXr rYrrrr<r )r8rrr.r<r.)rF)r8rrr.rrr<r=)r_rrrr<ztuple[str | None, int | None])r<r r&)r8rrDynamoFrameType | Noner<r)rr r<r )rPr.r<r.)rz!StackSummary | list[FrameSummary]r<r )NN) r8rr`rrrxrrr<r.)r`rr<r.)r`ztypes.CodeType | Nonerr.r<r.) __future__rrwloggingrrrrLtypingenumrr functoolsrpathlibrrrr r r r r r torch._guardsrtorch._utils_internalrrrutilsrtorch._dynamo.variablesrr output_graphrsymbolic_convertrrrr getLoggerr?ryr[getArtifactLoggerr] frozensetrJr* RuntimeErrorr,rFrRrWrSrnrsrvr}rrrrrrrrrrrrrrrrrrrrrrrr r rrrrrr!r#r%r'r) StopIteration LookupError IndexError GeneratorExitrAttributeErrorNotImplementedErrorr?r1r+r-r2r=rArDrH _subclasses fake_tensorDataDependentOutputExceptionDynamicOutputShapeExceptionUnsupportedOperatorExceptionUnsupportedFakeTensorException$UnsupportedMutationAliasingException!exceptions_allowed_to_be_fallbackrbrorrlobjectrrrrrrrrrrrrrZrrrr'r rrs"1f   II//17'0;9!>>33HnM"+" < % (0?(<(@ 3  !5  *   // /    _  $  ,  )   (  '+'$ E, E$*&*D +?  )  1 2 8D8$<* 2  #7  i  '   ,    1   +  ,  *  -  .  ,  "3  )  "3 ,$"( *&4   2 \ ,6:04 $$!$ 3 $ . $  $28 9 .< !!>> !!== !!>> !!@@ !!FF %!## # #  #  # ##L      4 140 8%0 %0%0 %0  %0  %0%0%0%0T .-b  ' " >9 59C C1CCB   L#'$(     "   6  "#'$( 8  8  8  8 " 8  8r