Dump Graph Details

Before model conversion, set the following environment variables:

export DUMP_GE_GRAPH=1 # Determines the size of a dump graph.
export DUMP_GRAPH_LEVEL=1 # Determines 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 it 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.

Each of the preceding files corresponds to a step in the model build process, for example, the build begins with the execution of the ge_onnx_00000001_graph_0_PreRunBegin.pbtxt file and ends with the execution of the ge_onnx_00000078_graph_0_PreRunAfterBuild.pbtxt file. Each file contains all operators involved in the corresponding step. For details about the subgraph functions in each phase of the dump graph, see Table 1. (The dump subgraphs may vary between models, but the workflow is basically the same.)

**Table 1** Description of dump graph details
Subgraph	Component	Description
ge_proto_xxxx_FlowGraphPreRunBegin.txt	GE	Graph before FlowModelBuild
ge_proto_xxxx_AfterFlowGraphPartition.txt	GE	Graph after flow partitioning (flow partitioning: a partitioning method used in DataFlow)
ge_proto_xxxx_AfterParallelPartitioner.txt	GE	Graph after pipeline parallel partitioning (The "pipeline" refers to the PP in the backend inference scenario. Currently, it is not supported in commercial scenarios.)
ge_proto_xxxx_PreRunBegin.txt	GE	Graph structure obtained after custom optimization.
ge_proto_xxxx_RunCustomPassBegin.txt	GE	Custom pass input graph
ge_proto_xxxx_RunCustomPassEnd.txt	GE	Custom pass output graph
ge_proto_xxxx_PreRunAfterInitPreparation.txt	FE	Graph structure obtained after all initialization in the graph preparation phase.
ge_proto_xxxx_PrepareAfterCheckAndUpdateInput.txt	GE	Graph structure obtained after the graph input is checked and updated.
ge_proto_xxxx_PrepareAfterPropagateFormatIfNeed.txt	GE	Graph structure obtained after format inference on single-operator.
ge_proto_xxxx_OptimizeGraph_TagNoConstFoldingAfter.txt	FE	Used for quantization scenarios. The FE adds a tag for an operator to indicate no constant folding. When the GE checks that the tag exists, it does not perform constant folding.
ge_proto_xxxx_PreRunAfterOptimizeGraphPrepare.txt	GE	Graph structure obtained after the original graphs in the operator information library are prepared (by calling the OptimizeGraphPrepare API).
ge_proto_xxxx_PreRunAfterHandleSummaryOp.txt	GE	Graph structure obtained after Summary node processing.
ge_proto_xxxx_PrepareAfterGraphEquivalentTransformation.txt	GE	Graph structure obtained after the for loop is equivalently transformed into while loop.
ge_proto_xxxx_PrepareAfterProcessOutput.txt	GE	Graph structure obtained after graph data processing.
ge_proto_xxxx_PrepareAfterOptimizeAfterGraphNormalization.txt	GE	Output graph of graph optimization after graph normalization
ge_proto_xxxx_PrepareAfterProcessMultiBatch.txt	GE	Graph structure obtained after processing in the presence of dynamic batch size profiles.
ge_proto_xxxx_PrepareAfterInsertAipp.txt	GE	Graph structure obtained after AIPP processing.
ge_proto_xxxx_PrepareAfterProcessAippNodesDataFormat.txt	GE	Output graph of the AIPP node with format updated
ge_proto_xxxx_PreRunAfterNormalizeGraph.txt	GE	Output graph of graph normalization
ge_proto_xxxx_PreRunAfterOptimizeGraphInit.txt	GE	Output graph of graph optimization after graph initialization
ge_proto_xxxx_PrepareAfterProcessBeforeInfershape.txt	GE	Graph structure obtained after "dead edges"of the conditional operator are eliminated.
ge_proto_xxxx_after_first_inferformat.txt	GE	Graph structure obtained after format inference on the entire graph.
ge_proto_xxxx_after_infershape.txt	GE	Graph structure obtained after shape inference on the entire graph, with constant folding.
ge_proto_xxxx_PrepareAfterInferFormatAndShape.txt	GE	Graph structure obtained after format and shape inference. This graph has undergone the second format inference, compared with the preceding graph.
ge_proto_xxxx_PrepareAfterCtrlFlowPreProcess.txt	GE	Graph structure obtained after the conditional operator is preprocessed.
ge_proto_xxxx_PrepareAfterGetDynamicOutputShape.txt	GE	Graph structure obtained after graph output processing in the presence of dynamic batch size profiles.
ge_proto_xxxx_PrepareAfterProcessAippStage2.txt	GE	Graph structure obtained after graph input processing in AIPP mode.
ge_proto_xxxx_PrepareAfterPrepareOptimize.txt	GE	Graph structure obtained after optimization in the graph preparation phase.
ge_proto_xxxx_PreRunAfterPrepare.txt	GE	Graph structure obtained after graph preparation, the same as the preceding graph.
ge_proto_xxxx_OptimizeQuantGraph_FeGraphFusionAfter.txt	FE	Graph structure obtained after quantization in the graph optimization phase ends.
ge_proto_xxxx_OptimizeOriginalGraph_FeGraphFusionAfter.txt	FE	Graph structure obtained after graph fusion ends.
ge_proto_xxxx_OptimizeOriginalGraph_FeTopoSortingAfter.txt	FE	Graph structure obtained after graph fusion and topology sorting, for checking whether a ring is formed.
ge_proto_xxxx_PreRunAfterOptimizeOriginalGraph.txt	GE	Graph structure obtained after the original graphs in the operator information library are optimized (by calling the OptimizeOriginalGraph API).
ge_proto_xxxx_PrepareAfterUpdateInputOutputByUserOptions.txt	GE	Graph structure obtained after the graph input and output are processed based on the user's command-line options.
ge_proto_xxxx_PrepareAfterUpdateVariableFormats.txt	GE	Graph structure obtained after the variable formats are processed.
ge_proto_xxxx_PreRunAfterPrepareRunningFormatRefiner.txt	GE	Same as the preceding graph.
ge_proto_xxxx_BeforeOptimizeOriginalGraphJudgeInsert.txt	FE	Input graph of the op_judge process
ge_proto_xxxx_OptimizeOriginalGraph_FeOpJudgeAfter.txt	FE	Graph structure obtained after the operator judging process.
ge_proto_xxxx_OptimizeOriginalGraph_FeDistHeavyFormatAfter.txt	FE	Graph structure obtained after diffusion of heavy operators.
ge_proto_xxxx_OptimizeOriginalGraph_FeInsertTransNodeAfter.txt	FE	Graph structure obtained after the transform operator is inserted.
ge_proto_xxxx_PreRunAfterRefineRunningFormat.txt	GE	Graph structure obtained after each operator information library is optimized (by calling the OptimizeOriginalGraphJudgeInsert API).
ge_proto_xxxx_PreRunAfterSubexpressionMigration.txt	GE	Graph structure after the common subexpression is extracted in the scenario of dynamic dimension size profiles.
ge_proto_xxxx_before_SameTransdataBreadthFusionPass.txt	GE	Input graph of SameTransdataBreadthFusionPass
ge_proto_xxxx_after_SameTransdataBreadthFusionPass.txt	GE	Output graph of SameTransdataBreadthFusionPass
ge_proto_xxxx_OptimizeStage1_1.txt	GE	Graph structure obtained after graph optimization stage 1_1.
ge_proto_xxxx_OptimizeStage1_2.txt	GE	Graph structure obtained after graph optimization stage 1_2.
ge_proto_xxxx_PreRunAfterOptimize1.txt	GE	Graph structure obtained after optimization stage 1 for all graphs.
ge_proto_xxxx_PreRunAfterOptimizeAfterStage1.txt	GE	Graph structure obtained after each operator information library is optimized (by calling the OptimizeAfterStage1 API).
ge_proto_xxxx_PreRunAfterInferShape2.txt	GE	Graph structure obtained after the second shape inference.
ge_proto_xxxx_AfterPipelinePartition.txt	GE	Graph structure obtained after graph partitioning for the local pipeline, which is used in the helper scenario.
ge_proto_xxxx_BeforeStagePartition.txt	GE	Graph before stage partitioning (Stage partitioning is mainly used in the early embedding scenario. It is uncertain whether it is still used.)
ge_proto_xxxx_AfterStagePartition.txt	GE	Graph after stage partitioning
gge_proto_xxxx_AfterEnginePlacer.txt	GE	Graph after engine selection
ge_proto_xxxx_Before_DSP.txt	GE	Graph before dynamic and static model splitting
ge_proto_xxxx_After_DSP.txt	GE	Graph after dynamic and static model splitting
ge_proto_xxxx_AfterDynamicShapePartition.txt	GE	Graph structure obtained after graph partitioning with a dynamic shape.
ge_proto_xxxx_MergedComputeGraphAfterCompositeEnginePartition.txt	GE	Structure of the merged graph obtained after opposite subgraph partitioning and subgraph optimization.
ge_proto_xxxx_partition0_rank0_inputNodeGraph_AtomicEnginePartitioning.txt	GE	Structure of the input node subgraph obtained after graph partitioning based on the atomic engine rules.
ge_proto_xxxx_partition0_rank1_new_sub_graph1_AtomicEnginePartitioning.txt	GE	Structure of subgraph 1 obtained after graph partitioning based on the atomic engine rules.
ge_proto_xxxx_partition0_rank2_new_sub_graph110_AtomicEnginePartitioning.txt	GE	Structure of subgraph 110 obtained after graph partitioning based on the atomic engine rules.
ge_proto_xxxx_OptimizeSubgraphPreProc.txt	GE	Output graph of subgraph preprocessing optimization
ge_proto_xxxx_DNN_VM_RTS_OptimizeSubGraphBefore.txt	RTS	-
ge_proto_xxxx_DNN_VM_RTS_OptimizeSubGraphAfter.txt	RTS	-
ge_proto_xxxx_AIcoreEngine_OptimizeSubGraphBefore.txt	FE	-
ge_proto_xxxx_OptimizeSubGraphBefore.txt	GE	Subgraph structure obtained before optimization. Subgraphs have the same name but different sequence numbers, depending on the number of subgraphs.
ge_proto_xxxx_OptimizeSubGraphAfter.txt	GE	Subgraph structure obtained after optimization. Subgraphs have the same name but different sequence numbers, depending on the number of subgraphs.
ge_proto_xxxx_partition0_rank1_new_sub_graph1_lxfusion_input.txt	AOE	SGAT input graph in the ATC and AOE baseline scenarios.
ge_proto_xxxx_partition0_rank1_new_sub_graph1_after_rebuild.txt	AOE	UB fusion graph of the AOE SGAT internal process.
ge_proto_xxxx_AIcoreEngine_OptimizeSubGraphAfter.txt	FE	-
ge_proto_xxxx_OptimizeSubgraphPostProc.txt	GE	Output graph of subgraph postprocessing optimization
ge_proto_xxxx_mergedComputeGraph.txt	GE	Structure of the merged graph, the same as the preceding graph.
ge_proto_xxxx_MergedComputeGraphAfterAtomicEnginePartition.txt	GE	Structure of the merged graph obtained after opposite subgraph partitioning and subgraph optimization.
ge_proto_xxxx_PreRunAfterOptimizeSubgraph.txt	GE	Subgraph structure obtained after optimization.
ge_proto_xxxx_OptimizeWholeGraphaicpu_tf_optimizer.txt	GE	Graph information obtained after the original graph optimization API of each engine is called. OptimizeWholeGraph is followed by the engine name.
ge_proto_xxxx_OptimizeWholeGraphaicpu_ascend_optimizer.txt	GE	Graph information obtained after the original graph optimization API of each engine is called. OptimizeWholeGraph is followed by the engine name.
ge_proto_xxxx_OptimizeWholeGraphdvpp_graph_optimizer.txt	GE	Output graph after DVPP entire graph optimization
ge_proto_xxxx_OptimizeWholeGraphAIcoreEngine.txt	GE	Graph information obtained after the original graph optimization API of each engine is called. OptimizeWholeGraph is followed by the engine name.
ge_proto_xxxx_OptimizeWholeGraphDNN_VM_RTS_GRAPH_OPTIMIZER_STORE.txt	GE	Graph information obtained after the original graph optimization API of each engine is called. OptimizeWholeGraph is followed by the engine name.
ge_proto_xxxx_OptimizeWholeGraphDNN_VM_HOST_CPU_OPTIMIZER.txt	GE	Graph information obtained after the original graph optimization API of each engine is called. OptimizeWholeGraph is followed by the engine name.
ge_proto_xxxx_PreRunAfterOptimizeWholeGraph.txt	GE	Graph structure obtained after each operator information library is optimized (by calling the OptimizeWholeGraph API).
ge_proto_xxxx_BeforeHandleMemConflict.txt	GE	Graph before the memory conflict is handled
ge_proto_xxxx_BeforeHandleMemoryLayoutConflict.txt	GE	Input graph of memory allocation conflict handling
ge_proto_xxxx_PreRunAfterMemConflictProc.txt	GE	Output graph of memory allocation conflict handling
ge_proto_xxxx_PreRunAfterOptimize2.txt	GE	Graph structure obtained after optimization stage 2 for all graphs.
ge_proto_xxxx_PreRunAfterOptimizeGraphBeforeBuild.txt	GE	Graph structure obtained after each operator information library is optimized (by calling the OptimizeGraphBeforeBuild API).
ge_proto_xxxx_partition0_rank0_inputNodeGraph_SecondPartitioning.txt	GE	Structure of the input node graph obtained after seconding partitioning.
ge_proto_xxxx_partition0_rank1_new_sub_graph1_SecondPartitioning.txt	GE	Structure of subgraph 1 obtained after seconding partitioning.
ge_proto_xxxx_partition0_rank2_new_sub_graph110_SecondPartitioning.txt	GE	Structure of subgraph 2 obtained after seconding partitioning.
ge_proto_xxxx_BeforePreBuildModel.txt	GE	Graph structure obtained after second graph splitting and before graph building.
ge_proto_xxxx_AfterPreBuildModel.txt	GE	Graph structure obtained after pre-building.
ge_proto_xxxx_AfterCalcOpParam.txt	GE	Graph structure obtained after the tensor sizes of all nodes in the graph are computed
ge_proto_xxxx_BeforeAssignedLogicalStreams.txt	GE	Graph structure obtained before logic streams are assigned.
ge_proto_xxxx_AfterAssignedLogicalStreams.txt	GE	Graph structure obtained after logic streams are assigned.
ge_proto_xxxx_BeforeRefreshRealStream.txt	GE	Graph structure obtained before the stream sync activation relationship is processed. This graph has undergone memory allocation, compared with the preceding graph.
ge_proto_xxxx_AfterRefreshRealStream.txt	GE	Graph structure obtained after the stream sync activation relationship is processed.
ge_proto_xxxx_AfterBuildModel.txt	GE	Graph structure obtained after weight combination and generation of basic model data.
ge_proto_xxxx_AfterOptimizeStreamedSubGraph.txt	GE	Graph structure obtained after the stream allocation result is optimized.
ge_proto_xxxx_GenerateTaskBefore.txt	GE	Graph structure obtained before a task is generated on the node.
ge_proto_xxxx_GenerateTaskAfter.txt	GE	Graph structure obtained after a task is generated on the node. The GenerateTask API of each operator information library is called.
ge_proto_xxxx_AfterGetTask.txt	GE	Graph structure obtained after all tasks are generated on the node, the same as the preceding graph.
ge_proto_xxxx_Build.txt	GE	Graph structure obtained after graph building.
ge_proto_xxxx_PreRunAfterBuild.txt	GE	Same as the preceding graph.
ge_proto_xxxx_BeforeAttrsCompress.txt	GE	Graph before OM attribute compression
ge_proto_xxxx_AfterAttrsCompress.txt	GE	Graph after the OM attribute compression

Parent topic: References