[ThSSKJrJr /SQr/SQr\"SS5 \"SS/\Q5 \"SS /5 \"S S S /5 \"S S5 \"SSS/5 \"SS5 \"S\5 \"SS5 \"S\5 \"SSS/5 \"SSS/5 \"SS/5 \"SS S!/5 \"S"S#S$/5 \"S%S&5 \"S'SS(S)/\QS*PS+PS,PS-P5 \"S.S/S0S19 \"S2S3S0S49 \"S5S6S0S49 \"S7S85 \"S9S:5 \"S;S<5 \"S=S>5 \"S?S@5 \"SASB5 \"SCSDS0S19 \"SESF5 \"SGSHS0S19 \"SISJ5 \"SKSL5 \"SMSNS0S49 \"SOSPS0S49 \"SQSRS0S19 \"SSSTS0S49 \"SUSV5 \"SWSXS0S19 \"SYSZS0S0S[9 \"S\S]S0S19 \"S^S_5 \"S`SaS0S0Sb9 \"ScSdS0S0Sb9 \"SeSfS0S49 \"SgS3S0S49 \"ShS3S0S49 \"SiSjS0S49 \"SkSlS0S49 \"SmSnS0S49 \"SoSpS0S49 \"SqSrS0S49 \"SsStS0S49 \"SuSvS0S49 \"SwSxS0S49 \"SySzS0S49 \"S{S|S3S}9 g~))register_artifact register_log)z%torch.fx.experimental.symbolic_shapesztorch.fx.experimental.sym_nodeztorch.fx.experimental.recording)ztorch.distributed"torch._dynamo.backends.distributedtorch.nn.parallel.distributedcache)ztorch._inductor.remote_cacheztorch._inductor.fb.remote_cachedynamoz torch._dynamo fake_tensorztorch._subclasses.fake_tensoraotztorch._functorch.aot_autogradztorch._functorch._aot_autogradautogradztorch.autogradinductorztorch._inductorztorch._inductor.cudagraph_trees cudagraphszGLogs information from wrapping inductor generated code with cudagraphs.dynamictorch distributedc10dz"torch.distributed.distributed_c10dztorch.distributed.rendezvousddprrppztorch.distributed.pipeliningfsdpztorch.distributed.fsdpz"torch.distributed._composable.fsdpdtensorztorch.distributed._tensorztorch.distributed.tensoronnxz torch.onnxexportz torch.exportztorch.export.dynamic_shapesztorch._export.converterztorch._export.non_strict_utilsztorch._export.serde.serializez"torch.fx.experimental.proxy_tensorguardszhThis prints the guards for every compiled Dynamo frame. It does not tell you where the guards come from.T)visibleverbose_guards)off_by_defaultbytecodez{Prints the original and modified bytecode from Dynamo. Mostly useful if you're debugging our bytecode generation in Dynamo.graphzvPrints the dynamo traced graph (prior to AOTDispatch) in a table. If you prefer python code use `graph_code` instead. graph_codez4Like `graph`, but gives you the Python code instead. graph_sizesz5Prints the sizes of all FX nodes in the dynamo graph. trace_sourcezAs we execute bytecode, prints the file name / line number we are processing and the actual source code. Useful with `bytecode` trace_callzhLike trace_source, but it will give you the per-expression blow-by-blow if your Python is recent enough.trace_bytecodezCAs we trace bytecode, prints the instruction and the current stack. aot_graphszPrints the FX forward and backward graph generated by AOTDispatch, after partitioning. Useful to understand what's being given to Inductoraot_joint_graphz_Print FX joint graph from AOTAutograd, prior to partitioning. Useful for debugging partitioningaot_graphs_effectszkPrints the FX forward and backward graph generated by AOTDispatch, useful for debugging effects processing.pre_grad_graphsz{Prints the FX graph before inductor pre grad passes. Useful to understand what's being given to Inductor before grad passespost_grad_graphsz}Prints the FX graph generated by post grad passes. Useful to understand what's being given to Inductor after post grad passes ir_pre_fusionz,Prints the IR before inductor fusion passes.ir_post_fusionz+Prints the IR after inductor fusion passes.compiled_autogradzzPrints various logs in compiled_autograd, including but not limited to the graphs. Useful for debugging compiled_autograd.compiled_autograd_verbosezjWill affect performance. Prints compiled_autograd logs with C++ info e.g. autograd node -> fx node mapping ddp_graphszOnly relevant for compiling DDP. DDP splits into multiple graphs to trigger comms early. This will print each individual graph here. recompilesz?Prints the reason why we recompiled a graph. Very, very useful.recompiles_verbosezPrints all guard checks that fail during a recompilation. At runtime, Dynamo will stop at the first failed check for each failing guard. So not all logged failing checks are actually ran by Dynamo.)rr graph_breakszPrints whenever Dynamo decides that it needs to graph break (i.e. create a new graph). Useful for debugging why torch.compile has poor performancenot_implementedzPrints log messages whenever we return NotImplemented in a multi-dispatch, letting you trace through each object we attempted to dispatch to output_codez>Prints the code that Inductor generates (either Triton or C++))rr kernel_codez?Prints the code that Inductor generates (on a per-kernel basis)schedulezIInductor scheduler information. Useful if working on Inductor fusion algo perf_hintsonnx_diagnosticsfusionzADetailed Inductor fusion decisions. More detailed than 'schedule' loop_orderingzLogs related to loop orderingoverlapz0Detailed Inductor compute/comm overlap decisionssym_nodez.Logs extra info for various SymNode operationstrace_shape_eventszBLogs traces for every ShapeEnv operation that we record for replaycudagraph_static_inputsz:Logs static inputs handling in dynamo, AOT, and cudagraphs benchmarkingz+Detailed Inductor benchmarking information. autotuningzKAutotuning choice logs, such as kernel source, perf, and tuning parameters.graph_region_expansionzMLogs detailed steps of the duplicate graph region tracker expansion algorithmcustom_format_test_artifactz Testing only) log_formatN) _internalrrDYNAMIC DISTRIBUTEDU/var/www/auris/envauris/lib/python3.13/site-packages/torch/_logging/_registrations.pyrHs6     P X2'23 ]<=> U46VWX Z)* Z+-NOPM  Y Wg ]K(  13Q R  +-Q R T234 V.0TUV Y46PQR V\"  %    "  )  (  -   n  "Bt<B  |, VWJFnIQ  eq  BD2 1 A  p KE  C  Y  SD   E   O ,48$b> G #  6 4 H @ 1 Q S /BOrF