Dump Graph Details

Before model conversion, set the following environment variables to dump the model graphs generated during build. You can observe model changes by viewing the dump graphs.

export DUMP_GE_GRAPH=1      # Set the content level of the dump graph.
export DUMP_GRAPH_LEVEL=1   # Set the number of dump graphs.

The following files are generated in the current path where the atc command is executed. For details about the environment variables, see 2.

ge_onnx*.pbtxt: model description structure based on ONNX. You can open this file using visualizer software such as Netron.
ge_proto*.txt: text file stored in Protobuf format. You can convert it into a JSON file to facilitate fault locating. This file appears in pair with the ge_onnx*.pbtxt file, but has more attributes of the string type than the ge_onnx*.pbtxt file, making the ge_proto*.txt file more comprehensive. You can open either of them.
Compared with the ge_onnx*.pbtxt file, the ge_proto*.txt file has a smaller size. Therefore, setting DUMP_GE_GRAPH to 2 or 3 has the same effect on the ge_proto*.txt file, that is, dumping information without data such as weights.
ge_readable*.txt: (optional) highly readable text file in Dynamo FX graph style. This type of file is generated only when the DUMP_GRAPH_FORMAT environment variable is set and the value contains readable.

Each of the preceding files corresponds to a step in the model build process, for example, the build begins with the ge_onnx_xxxx_PreRunBegin.pbtxt file and ends with the ge_onnx_xxxx_PreRunAfterBuild.pbtxt file. Each file contains all operators involved in the corresponding step. For details about the subgraphs at each dump phase, see Table 1. (The dump subgraphs may vary with models, but the main workflow remains consistent.)

**Table 1** Description of dump graph details
No.	Subgraph	Component	Phase	Trustlisted or Not
1	ge_proto_xxxx_GraphPreRunBegin.txt	GE	Graph before build.	-
2	ge_proto_xxxx_AfterFlowGraphPartition.txt	GE	Graph after flow partitioning (flow partitioning is a partitioning mode used in DataFlow).	-
3	ge_proto_xxxx_AfterParallelPartitioner.txt	GE	Graph after pipeline parallel partitioning (pipeline here refers to PP in the backend inference scenario).	-
4	ge_proto_xxxx_PreRunBegin.txt	GE	Original graph structure.	Trustlisted
5	ge_proto_xxxx_RunCustomPassBeforeInfershape.txt	GE	Exit graph of the custom pass before InferShape.	Trustlisted
6	ge_proto_xxxx_PreRunAfterInitPreparation.txt	FE	Graph structure after all initialization processing in the graph preparation phase.	-
7	ge_proto_xxxx_PreRunAfterHandleSummaryOp.txt	GE	Graph structure after Summary node processing.	-
8	ge_proto_xxxx_PrepareAfterCheckAndUpdateInput.txt	GE	Graph structure after the graph input is checked and updated.	-
9	ge_proto_xxxx_PrepareAfterGraphEquivalentTransformation.txt	GE	Graph structure after equivalent replacement of For-loop graph structures with While-loop graph structures.	-
10	ge_proto_xxxx_PrepareAfterProcessOutput.txt	GE	Graph structure after graph data processing.	-
11	ge_proto_xxxx_PrepareAfterOptimizeAfterGraphNormalization.txt	GE	Exit graph after graph optimization following graph normalization.	-
12	ge_proto_xxxx_PrepareAfterInsertAipp.txt	GE	Graph structure after AIPP processing (when AIPP parameters are configured).	-
13	ge_proto_xxxx_PrepareAfterProcessAippNodesDataFormat.txt	GE	Exit graph after AIPP node format update.	-
14	ge_proto_xxxx_PreRunAfterNormalizeGraph.txt	GE	Exit graph after graph normalization.	Trustlisted
15	ge_proto_xxxx_PreRunAfterOptimizeGraphInit.txt	GE	Exit graph after graph optimization and initialization.	-
16	ge_proto_xxxx_OptimizeGraph_TagNoConstFoldingAfter.txt	FE	Used in quantization scenarios. The FE tags operators that should skip constant folding. During constant folding, the GE checks this tag and skips constant folding if the tag exists.	-
17	ge_proto_xxxx_HcclAfterOptimizeGraphPrepare.txt	HCCL	Graph after optimization in the HCCL graph preparation phase.	-
18	ge_proto_xxxx_PreRunAfterOptimizeGraphPrepare.txt	GE	Graph structure after original-graph preparation via each operator information library (by calling the OptimizeGraphPrepare API).	-
19	ge_proto_xxxx_PrepareAfterProcessBeforeInfershape.txt	GE	Graph structure after dead-edge elimination for conditional operators.	-
20	ge_proto_xxxx_AfterFirstInferformat.txt	GE	Graph structure after full-graph format inference.	-
21	ge_proto_xxxx_AfterInfershape.txt	GE	Graph structure after full-graph shape inference, accompanied by constant folding.	Trustlisted
22	ge_proto_xxxx_PrepareAfterInferFormatAndShape.txt	GE	Graph structure after all InferFormat and InferShape processing. A second full-graph InferFormat occurs between this and the previous graph.	-
23	ge_proto_xxxx_RunCustomPass_AfterInferShape.txt	GE	Exit graph of the custom pass after InferShape.	Trustlisted
24	ge_proto_xxxx_AfterSecondInferformat.txt	GE	Graph after the second format inference.	-
25	ge_proto_xxxx_PrepareAfterCtrlFlowPreProcess.txt	GE	Graph structure after conditional operator preprocessing.	-
26	ge_proto_xxxx_PrepareAfterGetDynamicOutputShape.txt	GE	Graph structure after graph output processing in dynamic-profile scenarios.	-
27	ge_proto_xxxx_PrepareAfterProcessAippStage2.txt	GE	Graph structure after graph input node processing in AIPP mode.	-
28	ge_proto_xxxx_PrepareAfterPrepareOptimize.txt	GE	Graph structure after optimization in the graph preparation phase.	-
29	ge_proto_xxxx_PreRunAfterPrepare.txt	GE	Graph structure after all graph preparation processing (same as the previous graph).	Trustlisted
30	ge_proto_xxxx_OptimizeQuantGraph_FeGraphFusionAfter.txt	FE	Graph structure after quantization in the graph optimization phase.	-
31	ge_proto_xxxx_OptimizeOriginalGraph_FeGraphFusionAfter.txt	FE	Graph structure after graph fusion.	-
32	ge_proto_xxxx_OptimizeOriginalGraph_FeTopoSortingAfter.txt	FE	Graph structure after topological sorting following graph fusion, for detecting cycles.	-
33	ge_proto_xxxx_OptimizeOriginalGraph_DSAFeOpJudgeAfter.txt	FE	Graph after DataType and Format selection through dynamic shape analysis.	-
34	ge_proto_xxxx_HcclBeforeOptimizeOriginalGraph.txt	HCCL	Graph before HCCL-based original-graph optimization.	-
35	ge_proto_xxxx_HcclAfterOptimizeOriginalGraph.txt	HCCL	Graph after HCCL-based original-graph optimization.	-
36	ge_proto_xxxx_RunCustomPassAfterBuiltinFusionPass.txt	GE	Exit graph of the custom pass after built-in pass execution.	Trustlisted
37	ge_proto_xxxx_PreRunAfterOptimizeOriginalGraph.txt	GE	Graph structure after original-graph optimization via each operator information library (by calling the OptimizeOriginalGraph API).	Trustlisted
38	ge_proto_xxxx_PrepareAfterUpdateInputOutputByUserOptions.txt	GE	Graph structure after the graph inputs and outputs are processed based on user-specified options.	-
39	ge_proto_xxxx_PrepareAfterUpdateVariableFormats.txt	GE	Graph structure after variable Format processing.	-
40	ge_proto_xxxx_PreRunAfterPrepareRunningFormatRefiner.txt	GE	Same as the previous graph.	-
41	ge_proto_xxxx_BeforeOptimizeOriginalGraphJudgeInsert.txt	FE	Entry graph of the op_judge workflow.	-
42	ge_proto_xxxx_OptimizeOriginalGraph_FeOpDtypeJudgeAfter.txt	FE	Graph after precision mode selection.	-
43	ge_proto_xxxx_PreRunAfterRefineRunningPrecision.txt	GE	Graph after precision selection.	-
44	ge_proto_xxxx_AfterPrecisionRefine.txt	GE	Graph generated after conversion-operator fusion following precision selection.	-
45	ge_proto_xxxx_PreRunAfterAfterPrecisionRefine.txt	GE	Graph generated after conversion-operator fusion and automatic fusion following precision selection. If automatic fusion is not enabled, this subgraph is identical to ge_proto_xxxx_AfterPrecisionRefine.txt.	-
46	ge_proto_xxxx_OptimizeOriginalGraph_FeOpFormatJudgeAfter.txt	FE	Full op_judge graph after format selection.	-
47	ge_proto_xxxx_OptimizeOriginalGraph_FeDistHeavyFormatAfter.txt	FE	Graph structure after format propagation for heavy operators.	-
48	ge_proto_xxxx_OptimizeOriginalGraph_FeInsertTransNodeAfter.txt	FE	Graph structure after conversion operator insertion.	-
49	ge_proto_xxxx_PreRunAfterRefineRunningFormat.txt	GE	Graph structure after optimization via each operator information library (by calling the OptimizeOriginalGraphJudgeInsert API).	-
50	ge_proto_xxxx_PreRunAfterSubexpressionMigration.txt	GE	Graph structure after common subexpression extraction in dynamic-profile scenarios.	-
51	ge_proto_xxxx_before_SameTransdataBreadthFusionPass.txt	GE	Entry graph of SameTransdataBreadthFusionPass.	-
52	ge_proto_xxxx_after_SameTransdataBreadthFusionPass.txt	GE	Exit graph of SameTransdataBreadthFusionPass.	-
53	ge_proto_xxxx_OptimizeStage1_1.txt	GE	Graph structure after graph optimization stage 1_1.	-
54	ge_proto_xxxx_OptimizeStage1_2.txt	GE	Graph structure after graph optimization stage 1_2.	-
55	ge_proto_xxxx_PreRunAfterOptimize1.txt	GE	Graph structure after all graph optimization stage 1 processing.	-
56	ge_proto_xxxx_PreRunAfterOptimizeAfterStage1.txt	GE	Graph structure after optimization via each operator information library (by calling the OptimizeAfterStage1 API).	Trustlisted
57	ge_proto_xxxx_RunCustomPassAfterOriginGraphOptimize.txt	GE	Custom pass executed after original graph optimization.	Trustlisted
58	ge_proto_xxxx_PreRunAfterInferShape2.txt	GE	Graph structure after the second InferShape processing.	-
59	ge_proto_xxxx_BeforeStagePartition.txt	GE	Graph before stage partitioning.	-
60	ge_proto_xxxx_AfterStagePartition.txt	GE	Graph after stage partitioning.	-
61	ge_proto_xxxx_AfterEnginePlacer.txt	GE	Graph after engine selection.	-
62	ge_proto_xxxx_Before_DSP.txt	GE	Graph before static-dynamic partitioning.	-
63	ge_proto_xxxx_After_DSP.txt	GE	Graph after static-dynamic model partitioning.	-
64	ge_proto_xxxx_AfterDynamicShapePartition.txt	GE	Graph structure after dynamic-shape graph partitioning.	-
65	ge_proto_xxxx_MergedComputeGraphAfterCompositeEnginePartition.txt	GE	Structure of the merged graph after opposite subgraph partitioning and subgraph optimization.	-
66	ge_proto_xxxx_partition0_rank0_inputNodeGraph_AtomicEnginePartitioning.txt	GE	Structure of the input node subgraph after graph partitioning based on the atomic engine rules.	-
67	ge_proto_xxxx_partition0_rank1_new_sub_graph1_AtomicEnginePartitioning.txt	GE	Structure of subgraph 1 after graph partitioning based on the atomic engine rules.	-
68	ge_proto_xxxx_partition0_rank2_new_sub_graph110_AtomicEnginePartitioning.txt	GE	Structure of subgraph 110 after graph partitioning based on the atomic engine rules.	-
69	ge_proto_xxxx_OptimizeSubgraphPreProc.txt	GE	Exit graph after subgraph optimization preprocessing.	-
70	ge_proto_xxxx_DNN_VM_RTS_OptimizeSubGraphBefore.txt	RTS	-	-
71	ge_proto_xxxx_DNN_VM_RTS_OptimizeSubGraphAfter.txt	RTS	-	-
72	ge_proto_xxxx_AIcoreEngine_OptimizeSubGraphBefore.txt	FE	Entry graph before AI Core subgraph optimization.	-
73	ge_proto_xxxx_OptimizeSubGraphBefore.txt	GE	Subgraph structure before subgraph optimization. Each subgraph has a copy with the same name but different sequence numbers; the total count depends on the number of subgraphs.	-
74	ge_proto_xxxx_OptimizeSubGraphAfter.txt	GE	Subgraph structure after subgraph optimization. Each subgraph has a copy with the same name but different sequence numbers; the total count depends on the number of subgraphs.	-
75	ge_proto_xxxx_partition0_rank1_new_sub_graph1_lxfusion_input.txt	AOE(lxfusion)	SGAT input graph for ATC and AOE baseline scenarios.	-
76	ge_proto_xxxx_partition0_rank1_new_sub_graph1_after_rebuild.txt	AOE(lxfusion)	UB fusion graph within the AOE SGAT internal workflow.	-
77	ge_proto_xxxx_AIcoreEngine_OptimizeSubGraphAfter.txt	FE	Exit graph after AI Core subgraph optimization.	-
78	ge_proto_xxxx_OptimizeSubgraphPostProc.txt	GE	Exit graph after subgraph optimization postprocessing.	-
79	ge_proto_xxxx_mergedComputeGraph.txt	GE	Structure of the merged graph (same as the previous graph).	-
80	ge_proto_xxxx_MergedComputeGraphAfterAtomicEnginePartition.txt	GE	Structure of the merged graph after opposite atomic engine partitioning and subgraph optimization.	-
81	ge_proto_xxxx_PreRunAfterOptimizeSubgraph.txt	GE	Graph structure after subgraph optimization.	Trustlisted
82	ge_proto_xxxx_OptimizeWholeGraphaicpu_tf_optimizer.txt	GE	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
83	ge_proto_xxxx_OptimizeWholeGraphaicpu_ascend_optimizer.txt	GE	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
84	ge_proto_xxxx_OptimizeWholeGraphdvpp_graph_optimizer.txt	GE	Exit graph after DVPP optimization in full-graph optimization.	-
85	ge_proto_xxxx_OptimizeWholeGraphhccl_alltoallvc_fusion_optimizer.txt	HCCL	Graph after fusion optimization in the HCCL-based original-graph optimization phase.	-
86	ge_proto_xxxx_OptimizeWholeGraphAIcoreEngine.txt	GE	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
87	ge_proto_xxxx_OptimizeWholeGraphDSAEngine.txt	FE	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
88	ge_proto_xxxx_OptimizeWholeGraphhccl_graph_optimizer.txt	HCCL	Graph optimized in the HCCL-based original-graph optimization phase.	-
89	ge_proto_xxxx_OptimizeWholeGraphDNN_VM_RTS_GRAPH_OPTIMIZER_STORE.txt	GE	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
90	ge_proto_xxxx_OptimizeWholeGraphDNN_VM_RTS_FFTS_PLUS_GRAPH_OPTIMIZER_STORE.txt	RTS	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
91	ge_proto_xxxx_OptimizeWholeGraphDNN_VM_HOST_CPU_OPTIMIZER.txt	GE	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
92	ge_proto_xxxx_OptimizeWholeGraphge_local_graph_optimizer.txt	GE	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
93	ge_proto_xxxx_OptimizeWholeGraphffts_plus.txt	FE	Graph information after original-graph optimization API calls by each engine (OptimizeWholeGraph followed by the engine name).	-
94	ge_proto_xxxx_PreRunAfterOptimizeWholeGraph.txt	GE	Graph structure after optimization via each operator information library (by calling the OptimizeWholeGraph API).	-
95	ge_proto_xxxx_PreRunAfterOptimize2.txt	GE	Graph structure after all graph optimization stage 2 processing.	-
96	ge_proto_xxxx_PreRunAfterOptimizeGraphBeforeBuild.txt	GE	Entry graph for model build.	Trustlisted
97	ge_proto_xxxx_PreRunAfterOptimizeTensorMove.txt	GE	Graph after redundant TensorMove node optimization.	-
98	ge_proto_xxxx_BeforeHandleMemConflict.txt	GE	Graph before memory conflict resolution.	-
99	ge_proto_xxxx_AfterHandleMemConflict.txt	GE	Graph after memory conflict resolution.	-
100	ge_proto_xxxx_BeforeHandleMemoryLayoutConflict.txt	GE	Entry graph before memory layout conflict resolution.	-
101	ge_proto_xxxx_PreRunAfterMemConflictProc.txt	GE	Exit graph after memory read/write conflict resolution.	-
102	ge_proto_xxxx_partition0_rank0_inputNodeGraph_SecondPartitioning.txt	GE	Structure of the input node subgraph after second partitioning.	-
103	ge_proto_xxxx_partition0_rank1_new_sub_graph1_SecondPartitioning.txt	GE	Structure of subgraph 1 after second partitioning.	-
104	ge_proto_xxxx_partition0_rank2_new_sub_graph110_SecondPartitioning.txt	GE	Structure of subgraph 110 after second partitioning.	-
105	ge_proto_xxxx_BeforeUnfoldSubgraphs.txt	GE	Graph before dynamic-shape graph unfolding.	-
106	ge_proto_xxxx_AfterUnfoldSubgraphs.txt	GE	Graph after dynamic-shape graph unfolding.	-
107	ge_proto_xxxx_RunCustomPass_BeforeAssignLogicStream{pass_name}.txt	GE	Graph before user-defined stream allocation pass.	Trustlisted
108	ge_proto_xxxx_RunCustomPass_AfterAssignLogicStream{pass_name}.txt	GE	Graph after user-defined stream allocation pass.	Trustlisted
109	ge_proto_xxxx_AfterAssignResource.txt	GE	Graph after stream allocation, memory allocation, and GenTask.	-
110	ge_proto_xxxx_Build.txt	GE	Exit graph for model build.	Trustlisted
111	ge_proto_xxxx_PreRunAfterBuild.txt	GE	Same as the previous graph.	-
112	ge_proto_xxxx_BeforeAttrsCompress.txt	GE	Graph before offline model compression.	-
113	ge_proto_xxxx_AfterAttrsCompress.txt	GE	Graph after offline model compression.	-
114	ge_proto_xxxx_BeforeAttrsDecompress.txt	GE	Graph before offline model decompression.	-
115	ge_proto_xxxx_AfterAttrsDecompress.txt	GE	Graph after offline model decompression.	-
116	ge_proto_xxxx_ComputeGraphBeforeLowering.txt	GE	Computational graph before lowering.	Trustlisted
117	ge_proto_xxxx_Before_MultiStream_LoweringFirstLastEventSync.txt	GE	Execution graph before multi-stream event insertion.	-
118	ge_proto_xxxx_ExeGraphBeforeOptimize.txt	GE	Execution graph before graph optimization.	Trustlisted
119	ge_proto_xxxx_AfterZeroCopy.txt	GE	Execution graph after zero-copy optimization.	-
120	ge_proto_xxxx_AfterCEM.txt	GE	Execution graph after CEM optimization.	-
121	ge_proto_xxxx_AfterCopyFlowLaunch.txt	GE	Execution graph after inline copy optimization.	-
122	ge_proto_xxxx_TrustOutTensorAfter.txt	GE	Execution graph after TrustOutTensor optimization.	-
123	ge_proto_xxxx_AfterAicpuFuseHostInputs.txt	GE	Graph after AicpuFuseHostInputs optimization.	-
124	ge_proto_xxxx_ExecuteGraphAfterSplit.txt	GE	Final execution graph after dynamic-shape graph splitting.	Trustlisted

Parent topic: References