* fixed seg fault when using concrete shape
disable gradient as output
* fix evaluation hang issue for multiple gpu run
* Remove dead code, ORTModel and improve docstrings (#3814)
* Refine ORTTrainer docstring descriptions (#3907)
Add transformer glue test example to show how to use ORTTrainer to fine-tune a transformer model
Co-authored-by: liqun <liqun@OrtTrainingDev4.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
* online partition
* fix when multiple consumer nodes is in cut info
* fix windows build
* address feedback
* adding test
* feedback
* address feedback
* add parser for cut edge
* windows build
In this PR, we
1. create some APIs for creating NVTX objects
2. apply those APIs in pipeline-related operators and sequential executor.
As a result, we can explicitly see how a pipeline schedule is run by GPUs in
Nvidia's visual profiler. Note that these APIs are Linux only due to Nvidia's
limited support.
* Remove 'model_.' prefix for onnx model initializers in training
* fix test case remove redundant device test
* rename
* Fix state_dict/load_state_dict with frozen_weight
* nit
* Add monkey patch for pt opset 10
* remove pt patch in CI
* nit: newline
* gpt2 training perf
* gpt2 training perf
* debug
* debug
* debug
* fix bug
* minor
* on comments
* dynamic sql
* fix build
* minor
* linked hash
* on comments
* minor
* mem
* minor
Co-authored-by: Ethan Tao <ettao@microsoft.com>
* Initial update of readme
* Readme updates
* Review of consolidated README (#3930)
* Proposed updates for readme (#3953)
I found some of the information was duplicated within the doc, so attempted to streamline
* Fix links
* More updates
- fix build instructions
- nodejs doc reorganization
- roadmap update
- version fixes
* Update ORT Server build instructions
* More doc cleanup
* fix python dev notes name
* Update nodejs and some links
* sync eigen version back to master
* Minor fixes
* add nodsjs to sample table of content
* Update README.md
* Update README.md
* Update README.md
* Update README.md
* Update README.md
* Update README.md
* address PR feedback
* address PR feedback
* nodejs build instruction
* Update Java instructions to include gradle
* Roadmap refresh
Reformat some data, fix link, minor rewording
* Clarify Visual C++ runtime req
Co-authored-by: Nat Kershaw (MSFT) <nakersha@microsoft.com>
Co-authored-by: Prasanth Pulavarthi <prasantp@microsoft.com>
Co-authored-by: manashgoswami <magoswam@microsoft.com>
* dashboard integration - first phase
* change a field
* perf scripts
* addressing PR comments
* address comments and fix build
* minor
* make GetConfigFromData() const
* more update for comments
* addressing comments
* more on addressing comments
* minor
* fix build
* add condition check
* more on comments
* retrun status
* remove batch size
* on comments
* rename pkg path
* rename pkg path
* additional commentss
Co-authored-by: Ethan Tao <ettao@microsoft.com>
* Do not register Dropout(12) as training ONLY kernel.
* Move Dropout forward implementation in inference project.
* fix inference build test failures.
* remove fp16 test since its support is absent on CPU.
* build break.
* Fold Shape node in constant folding.
* bugfix
* Fix test failure.
* Bugfix for C++ frontend.
* Bugfix for C++ frontend.
Co-authored-by: Vincent Wang <weicwang@microsoft.com>
1. Parallel all the activations ops.
2. Parallel the performance critical path of the LRN op, which makes the ONNX model zoo googlenet model runs 60% faster(latency reduced from 21ms to 13ms).
3. Make the Gemm-Activation fusion support with all the activations ops. Before this change, it only supports LeakyRelu/Relu/Sigmoid/Tanh.
4. Delete onnxruntime/test/framework/op_kernel_test.cc because the file is almost empty.
5. Remove the loggings in KernelRegistry::TryFindKernel, return Status with error message instead.
Increase test comparison tolerance. Add output of random seed value for easier debugging later. Unify RandomValueGenerator::Uniform() to consistently use [min, max) interval.
* initial change to transformer.py
* prepare e2e transformer tests
* refactor transformer tests
* put test python files in a flat folder
* fix typo pip install transform(s)
* python 3.6
* python version to 3.6 in install_ubuntu.sh
* remove argparser
* to use opset ver 12
* workaround loss_scale naming patch in case of loss_fn_
* assign self.loss_fn_ so it can be checked
* skip a few un-needed post-process steps
* fix loss_scale_input_name, clean up post process steps
* skip non-frontend tests
* move cpu/cuda related files to coresponding cpu/cuda folder (#3668)
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* type cast for ratio is not necessary for dropout (#3682)
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* thrustallocator is not needed since cub is used directly for gather now. (#3683)
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* GatherND-12 Implementation (#3645)
* Renamed, UT passing
* Move GatherND CUDA Kerenl into onnxruntime
* Merge GatherNDOpTest
* Refactor Test code
* Merge CPU Kernel Impl
* Handle Negative Indice, Fix UT
* Improve CUDA kernel to handle negative index
* Minor Fixes
* Preserve GatherND-1 Cuda kernel
* Fix Mac build
* fix UT
* Fix Build
* fix GatherNDOpTest.double > CUDA error cudaErrorInvalidDeviceFunction:invalid device function
Co-authored-by: Sherlock Huang <bahuang@OrtTrainingDev3.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Co-authored-by: Peng Wang (pengwa) <pengwa@microsoft.com>
* update with reviewers' comments
* testBertTrainingGradientAccumulation was not using rtol and may fail occasionally with small (e-06) difference
* fix merge mistakes
Co-authored-by: liqun <liqun@OrtTrainingDev4.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Co-authored-by: Weixing Zhang <weixingzhang@users.noreply.github.com>
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
Co-authored-by: Sherlock <baihan.huang@gmail.com>
Co-authored-by: Sherlock Huang <bahuang@OrtTrainingDev3.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Co-authored-by: Peng Wang (pengwa) <pengwa@microsoft.com>
* Refactor GatherND CPU Kernel (Renaming & Simplify)
* Add batch_dim=1 or 2, negative slices tests
* Rename gather_nd_gard_impl.cu
* Use dispatcher to refactor CUDA GatherND/GatherNDGrad
* Change GatherNDBase::CommonComputeKernel --> GatherNDBase::PrepareCompute
* Use HasCudaEnvironment instead of __CUDA_ARCH__ for some double type tests
* Change naming of moments to Moment_x_<weight_name>
* Add checkpointing code and zero checkpoint aggregation
* Correct aggregation for LAMB, cleanup
* Add simple checkpointing test
* Add test for zero checkpoint aggregation
* Fix tests
* fix test
* Review changes
* Fix test after review comment fix
* Fix API, test
* Fix test after API change
* Decouple save load from ORTTrainer
* Add flag to not break checkpointing with ORTModel'
Co-authored-by: aishwarya bhandare <aibhanda@OrtTrainingDev3.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
* allow switching between eval and training modes dynamically
Co-authored-by: Tixxx <root@525204a066204ea794f942530b05ae7f000000.axlncovkyjne5caro2tmz3zryb.xx.internal.cloudapp.net>
* move cpu/cuda related files to coresponding cpu/cuda folder (#3668)
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* type cast for ratio is not necessary for dropout (#3682)
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* thrustallocator is not needed since cub is used directly for gather now. (#3683)
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* GatherND-12 Implementation (#3645)
* Renamed, UT passing
* Move GatherND CUDA Kerenl into onnxruntime
* Merge GatherNDOpTest
* Refactor Test code
* Merge CPU Kernel Impl
* Handle Negative Indice, Fix UT
* Improve CUDA kernel to handle negative index
* Minor Fixes
* Preserve GatherND-1 Cuda kernel
* Fix Mac build
* fix UT
* Fix Build
* fix GatherNDOpTest.double > CUDA error cudaErrorInvalidDeviceFunction:invalid device function
Co-authored-by: Sherlock Huang <bahuang@OrtTrainingDev3.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Co-authored-by: Peng Wang (pengwa) <pengwa@microsoft.com>
* Set gradient as output only for easy mode (#3694)
* Support GPU Event Operators (#3653)
* Add GPU event operators to support in-place updates in
gradient accumulator and optimizer for modifying the tensors
passing through those event operators.
* Address comment and polish code
* Merge shared code between CPU and GPU kernels
* Move event test to a new file
* Address comments
* Update onnxruntime/core/providers/cuda/gpu_data_transfer.cc
* fix path of cpu_featurizers_kernels.cc and cpu_featurizers_kernels.h
Co-authored-by: Weixing Zhang <weixingzhang@users.noreply.github.com>
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
Co-authored-by: Sherlock <baihan.huang@gmail.com>
Co-authored-by: Sherlock Huang <bahuang@OrtTrainingDev3.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Co-authored-by: Peng Wang (pengwa) <pengwa@microsoft.com>
Co-authored-by: ashbhandare <ash.bhandare@gmail.com>
Co-authored-by: Wei-Sheng Chin <wschin@outlook.com>
Co-authored-by: Ethan Tao <ettao@microsoft.com>
* Add GPU event operators to support in-place updates in
gradient accumulator and optimizer for modifying the tensors
passing through those event operators.
* Address comment and polish code
* Merge shared code between CPU and GPU kernels
* Move event test to a new file
* Address comments
* Update onnxruntime/core/providers/cuda/gpu_data_transfer.cc
* Renamed, UT passing
* Move GatherND CUDA Kerenl into onnxruntime
* Merge GatherNDOpTest
* Refactor Test code
* Merge CPU Kernel Impl
* Handle Negative Indice, Fix UT
* Improve CUDA kernel to handle negative index
* Minor Fixes
* Preserve GatherND-1 Cuda kernel
* Fix Mac build
* fix UT
* Fix Build
* fix GatherNDOpTest.double > CUDA error cudaErrorInvalidDeviceFunction:invalid device function
Co-authored-by: Sherlock Huang <bahuang@OrtTrainingDev3.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Co-authored-by: Peng Wang (pengwa) <pengwa@microsoft.com>
1. It is not necessary to include cudnn_common.h for kernels which are not implemented with CUDNN.
2. Minor change in layer norm kernel to simplify the code and resolve building warning.
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* pipeline transformer
* clean up
* address feedback
* add record/wait for first stage and updated split script
* address feedback
* make recv/send signal as initializer
* merge
* address feedback
* unify input and initializer
* address feedback and bug fix
* minor fix
* windows build
* fix
* Expand elmination and Expand gradient.
* Resolve comments.
* Fix test break.
* Check if graph can remove the node.
* Resolve comment.
Co-authored-by: Vincent Wang <weicwang@microsoft.com>
* fixes for ort_trainer.py to resume from checkpoint
* define self.state_dict_ during init
* add comment of explanation
* add unit test for restore from checkpoint
* fix file not found
Co-authored-by: suffian khan <sukha@microsoft.com>
1. Centralize its definition in common.cuh.
2. Rename it to GPU_WARP_SIZE which can be extended to AMD GPU later.
3. Centralize warp shuffle functions.
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* Remove Useless Cast during Transformer.
* Resolve comments.
* Check if graph can remove the node.
Co-authored-by: Vincent Wang <weicwang@OrtDevTest2v100.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
* add frontend minst test
* to use torch nightly with torchvision
* remove incorrect comment per reviewer's comment
* experiment torchvision import failure
* experiment install_deps.sh
* more experiment install_deps.sh
* experiment install_deps.sh with --upgrade
* Experiment with install_deps.sh.
* Experiment with install_ubuntu.sh.
* Use Ubuntu 18.04 and Python 3.6 for CI.
* Update cmake version for CI.
* Install MPI on Ubuntu 18.04 for CI.
* Increase tolerance for MNIST test.
* Go back to Ubuntu 16.04 for CI, fix installing from deadsnakes ppa.
* Clean-up.
* Update ort_trainer.py from ort_training.
* Get default Ubuntu Python ver back to 3.5.
* Add underscore to opset_version parameter name in ORTTrainer constructor.
* Move loss/model wrap before the call for sample output.
* Update expected values for MNIST test.
Co-authored-by: liqun <liqun@OrtTrainingDev4.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Co-authored-by: Sergii Dymchenko <sedymche@microsoft.com>
Fix training modification of Graph SetInputs() and SetOutputs(). Originally there were distinct code paths in Graph based on whether the graph was loaded from a GraphProto or created from scratch. The training modifications made that distinction a bit ambiguous - i.e., even though the Graph is loaded from a GraphProto for training, sometimes we rely on the other code path, e.g., to deduce the graph inputs after modifying it. Consequently, there was some odd behavior when using SetInputs(). For correctness, this change separates the cases where the graph is loaded from a GraphProto and where it is created from scratch.
* Changed internal loss scale to 1-D
* added test
Co-authored-by: root <root@525204a066204ea794f942530b05ae7f000000.axlncovkyjne5caro2tmz3zryb.xx.internal.cloudapp.net>
* Fixes for Expand, Where, ConcatGrad ReduceSumGrad.
* Roll back expand, fix, add tests for reduce grad.
* Roll back CPU Expand change.
* Fix after merge.
Co-authored-by: Vincent Wang <weicwang@microsoft.com>
Made some fixes to enable loss scale to be wired up to ORT from the Python frontend. In particular, now addition of loss scaling is done unconditionally if mixed precision is enabled. The generated loss scale input name is passed back to the frontend.
Also fixed how inputs were added during the training graph configuration. Graph::SetInputs() was causing some issues - it seems to not be working correctly.
Also added some mixed precision Python frontend tests.
Co-authored-by: liqun <liqun@OrtTrainingDev4.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
Implement pipeline event generator with OneFWOneBW schedule in timeline. Each stage of pipeline contains FW and BW of a subset of the model and are scheduled in one worker thread for each microbatch.
1. misaligned address in atomic_add()
2. GatherNDGradKernel to use atomic_add
3. enable/add UTs for GatherNDGrad and reduction_ops using half
- __CUDA_ARCH__ won't take effect on .cc code, leverage HasCudaEnvironment() instead
4. verified convergence graph and perf test
- p100 is much slower than v100 on fp16
- fp16/128 need to reduce batch size from 66 to 64 to avoid OOM issue
5. verify convergence test on Dev3/v100
TBD - broken UTs related to MatmulIntegerOpTest (works on v100/windows, though)
This is a draft of graph cut and wait/record to demonstrate cut and Wait/Record design. You may find sub models and profiling json under onnxruntime/test if you run "onnxruntime_test_all --gtest_filter=GradientGraphBuilderTest.TrainingSession_WithPipeline"
