mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-06-29 03:30:52 +00:00
4b5b5f7101
### Description This PR is to update the win-ort-main branch to the tip main branch as of 2025-01-23. ### PR List ddf0d377a7 [QNN EP] Add LoggingManager::HasDefaultLogger() to provider bridge API (#23467) 05fbbdf91f [QNN EP] Make QNN EP a shared library (#23120) 1336566d7f Add custom vcpkg ports (#23456) 2e1173c411 Update the compile flags for vcpkg packages (#23455) 1f628a9858 [Mobile] Add BrowserStack Android MAUI Test (#23383) 009cae0ec8 [js/webgpu] Optimize ConvTranspose (Continue) (#23429) 04a4a694cb Use onnx_protobuf.h to suppress some GCC warnings (#23453) 2e3b62b4b0 Suppress some strict-aliasing related warnings in WebGPU EP (#23454) b708f9b1dc Bump ruff from 0.9.1 to 0.9.2 (#23427) c0afc66b2a [WebNN] Remove workarounds for TFLite backend (#23406) 8a821ff7f9 Bump vite from 6.0.7 to 6.0.11 in /js/web/test/e2e/exports/testcases/vite-default (#23446) 220c1a203e Make ORT and Dawn use the same protobuf/abseil source code (#23447) b7b5792147 Change MacOS-13 to ubuntu on for android-java-api-aar-test.yml. (#23444) 19d0d2a30f WIP: Dp4MatMulNBits accuracy level 4 matmul for WebGPU EP (#23365) 95b8effbc4 [QNN EP]: Clean up QNN logging resources if an error occurs during initialization (#23435) 626134c5b5 Bump clang-format from 19.1.6 to 19.1.7 (#23428) 0cf975301f Fix eigen external deps (#23439) f9440aedce Moving RN_CI Android Testing to Linux (#23422) 1aa5902ff4 [QNN EP] workaround for QNN validation bug for Tanh with uint16 quantized output (#23432) 7f5582a0e2 Seperate RN andriod and IOS into 2 separated Stages. (#23400) 73deac2e7f Implement some missing element wise Add/Sub/Mul/Div/Neg operations for CPU and CUDA EPs (#23090) 949fe42af4 Upgrade Java version from react-native/android to Java 17 (#23066) 0892c23463 Update Qnn SDK default version to 2.30 (#23411) 94c099bcec Fix type cast build error (#23423) d633e571d1 [WebNN EP] Fix AddInitializersToSkip issues (#23354) e988ef00e2 [QNN EP] Fix regression for MatMul with two quantized/dynamic uint16 inputs (#23419) 7538795f6b Update onnxruntime binary size checks ci pipeline's docker image (#23405) 6c5ea41cad Revert "[QNN EP] Clean up correctly from a partial setup (#23320)" (#23420) e866804bbe Enable comprehension simplification in ruff rules (#23414) 0a5f1f392c bugfix: string_view of invalid memory (#23417) 4cc38e0277 fix crash when first input of BatchNormalization is 1-D (#23387) 033441487f Target py310 and modernize codebase with ruff (#23401) 87341ac010 [QNN EP] Fix segfault when unregistering HTP shared memory handles (#23402) ### Motivation and Context This update includes the change to make QNN-EP a shared library. --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: Adrian Lizarraga <adlizarraga@microsoft.com> Co-authored-by: Justin Chu <justinchuby@users.noreply.github.com> Co-authored-by: Yulong Wang <7679871+fs-eire@users.noreply.github.com> Co-authored-by: Edward Chen <18449977+edgchen1@users.noreply.github.com> Co-authored-by: Changming Sun <chasun@microsoft.com> Co-authored-by: Peishen Yan <peishen.yan@intel.com> Co-authored-by: Tianlei Wu <tlwu@microsoft.com> Co-authored-by: Hector Li <hecli@microsoft.com> Co-authored-by: Jian Chen <cjian@microsoft.com> Co-authored-by: Alexis Tsogias <1114095+Zyrin@users.noreply.github.com> Co-authored-by: junchao-zhao <68935141+junchao-loongson@users.noreply.github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: sushraja-msft <44513542+sushraja-msft@users.noreply.github.com> Co-authored-by: Wanming Lin <wanming.lin@intel.com> Co-authored-by: Jiajia Qin <jiajiaqin@microsoft.com> Co-authored-by: Caroline Zhu <wolfivyaura@gmail.com> |
||
|---|---|---|
| .. | ||
| experiment.py | ||
| gpt2_model_transform.py | ||
| layer_norm_transform.py | ||
| model_transform.py | ||
| nv_run_pretraining.py | ||
| opset12_model_transform.py | ||
| performance_investigation.py | ||
| pipeline_model_split.py | ||
| README.txt | ||
| single_node_perf.sh | ||
| sqldb_to_tensors.py | ||
| train.py | ||
| watch_experiment.py | ||
Procedure to export NV's pytorch model to ONNX. 1. cd into BERT in DeepLearningExamples and launch the docker. 2. run nv_run_pretraining.py using the same parameter you run run_pretraining.py. It will produce a model in your checkpoint directory. 3. Assume that the exported model's name is 'bert.onnx'. Run model_transform.py bert.onnx 4. Then run layer_norm_transform.py bert_optimized.onnx. The final model name would be bert_optimized_layer_norm.onnx 5. Now, you can run training with the newly created model. Note that if you want to change model's configuration, you can edit bert_config.json in the BERT directory. Example commands: Step 2 (inside docker): python3 /workspace/bert/nv_run_pretraining.py --input_dir=data/bookcorpus/hdf5_shards/ --output_dir=/results/checkpoints1 --config_file=bert_config.json --bert_model=bert-large-uncased --warmup_proportion=0 --num_steps_per_checkpoint=2000 --learning_rate=0.875e-4 --seed=42 --do_train --phase2 --max_seq_length=512 --max_predictions_per_seq=80 --max_steps=200 --train_batch_size=2 Step 3 (inside onnxruntime/build/Linux/RelWithDeb): sudo /data/anaconda/envs/py35/bin/python /bert_ort/wechi/DeepLearningExamples/PyTorch/LanguageModeling/BERT/model_transform.py /bert_ort/wechi/DeepLearningExamples/PyTorch/LanguageModeling/BERT/results/checkpoints1/bert_for_pretraining_without_loss_vocab_30528_hidden_1024_maxpos_512.onnx Step 4 (inside onnxruntime/build/Linux/RelWithDeb): sudo /data/anaconda/envs/py35/bin/python /bert_ort/wechi/DeepLearningExamples/PyTorch/LanguageModeling/BERT/layer_norm_transform.py /bert_ort/wechi/DeepLearningExamples/PyTorch/LanguageModeling/BERT/results/checkpoints1/bert_for_pretraining_without_loss_vocab_30528_hidden_1024_maxpos_512_optimized.onnx Step 5 (inside onnxruntime/build/Linux/RelWithDeb): ./onnxruntime_training_bert --num_of_perf_samples=100 --train_batch_size=1 --mode=perf --model_name /bert_ort/wechi/DeepLearningExamples/PyTorch/LanguageModeling/BERT/results/checkpoints1/bert_for_pretraining_without_loss_vocab_30528_hidden_1024_maxpos_512_optimized_layer_norm