saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-22 19:23:30 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Disable MatmulTransposeFusion for CPU EP (#7135)

Browse source 
It causes convergence issue in BERT on CPU
This commit is contained in:
KeDengMS 2021-03-25 17:16:58 -07:00 committed by GitHub
parent 2bf54bcaa2
commit c9b29fbd06
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 1 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
orttraining/orttraining/core/optimizer/graph_transformer_utils.cc
View file

@ -171,7 +171,7 @@ std::vector<std::unique_ptr<GraphTransformer>> GenerateTransformers(
//transformers.emplace_back(onnxruntime::make_unique<ConstantFolding>(l1_execution_providers));
transformers.emplace_back(onnxruntime::make_unique<MatMulAddFusion>(l1_execution_providers));
transformers.emplace_back(onnxruntime::make_unique<FreeDimensionOverrideTransformer>(free_dimension_overrides));
transformers.emplace_back(onnxruntime::make_unique<MatmulTransposeFusion>(l1_execution_providers));
transformers.emplace_back(onnxruntime::make_unique<MatmulTransposeFusion>(cuda_rocm_execution_providers));
transformers.emplace_back(onnxruntime::make_unique<BiasDropoutFusion>(cuda_rocm_execution_providers));
transformers.emplace_back(onnxruntime::make_unique<BiasSoftmaxFusion>(l1_execution_providers));
transformers.emplace_back(onnxruntime::make_unique<MatMulScaleFusion>(l1_execution_providers, weights_to_train));