Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[WIP] New PyTorch frontend API #4176

Closed
thiagocrepaldi opened this issue Jun 9, 2020 · 4 comments · Fixed by #5045
Closed

[WIP] New PyTorch frontend API #4176

thiagocrepaldi opened this issue Jun 9, 2020 · 4 comments · Fixed by #5045
Assignees
@thiagocrepaldi @rayankrish @spandantiwari
Labels
training issues related to ONNX Runtime training; typically submitted using template

Comments

@thiagocrepaldi
Copy link
Contributor

This issue will track all tasks related to the development of the new PyTorch frontend API. Once this issue is merged, the new frontend will be available on master branch.

Implementation will follow the MS internal design that has been reviewed internally
@thiagocrepaldi thiagocrepaldi added training issues related to ONNX Runtime training; typically submitted using template Frontend labels Jun 9, 2020
@thiagocrepaldi
Copy link
Contributor Author

#4220 introduces ORTTrainerOptions, optim and amp submodules, CI tests for the new frontend

@thiagocrepaldi thiagocrepaldi mentioned this issue Jun 16, 2020
Merged
This was linked to pull requests Jun 16, 2020
Add PytorchTrainerOptions class for the new pytorch frontend #4220
Closed
Add TrainStepInfo used on the new frontend API #4256
Merged
This was linked to pull requests Jun 18, 2020
Add Optimizer classes to the new frontend #4280
Merged
Add Loss Scaler classes to the new frontend #4306
Merged
@thiagocrepaldi thiagocrepaldi linked a pull request Jun 27, 2020 that will close this issue
Add LRScheduler implementation #4357
Merged
@spandantiwari
Copy link

There are multiple PRs in flight for new PyTorch frontend API. Is there an order in which these PRs have to be merged?

@thiagocrepaldi
Copy link
Contributor Author

There are multiple PRs in flight for new PyTorch frontend API. Is there an order in which these PRs have to be merged?

Yes, from the "Linked pull requests" list on the right side of this issue page, we should merge the first 3 from top to bottom. The last 2 can be merged in any order after that

This was linked to pull requests Jun 30, 2020
Add ORTTrainerOptions class for the new pytorch frontend #4382
Merged
Add _ORTTrainerModelDesc to perform validation for model description #4416
Merged
Add basic ORTTrainer API #4435
Merged
@thiagocrepaldi thiagocrepaldi pinned this issue Jul 17, 2020
@tracysh tracysh unpinned this issue Jul 18, 2020
This was linked to pull requests Jul 30, 2020
Add ONNX Model Export to New Frontend #4612
Merged
Create training session + minor improvements #4668
Merged
@thiagocrepaldi thiagocrepaldi removed a link to a pull request Jul 30, 2020
Add PytorchTrainerOptions class for the new pytorch frontend #4220
Closed
@thiagocrepaldi thiagocrepaldi linked a pull request Jul 30, 2020 that will close this issue
Save ONNX model in file #4671
Merged
@thiagocrepaldi
Copy link
Contributor Author

thiagocrepaldi commented Jul 31, 2020 •
edited
Loading

PRs already merged:
#4256
#4280
#4306
#4357
#4382
#4416
#4435
#4592
#4605
#4612
#4668
#4671
#4674
#4677
#4694
#4716
#4727
#4739
#4757
#4758
#4772
#4777
#4793
#4795
#4807
#4814
#4826
#4827
#4828
#4833
#4842
#4855
#4859
#4880
#4885
#4877
#4887
#4919
#4950
#4963
#4965
#4966
#4968
#5002
#5037
#5052

This was unlinked from pull requests Jul 31, 2020
Add TrainStepInfo used on the new frontend API #4256
Merged
Add Optimizer classes to the new frontend #4280
Merged
Add Loss Scaler classes to the new frontend #4306
Merged
Add LRScheduler implementation #4357
Merged
Add ORTTrainerOptions class for the new pytorch frontend #4382
Merged
Add _ORTTrainerModelDesc to perform validation for model description #4416
Merged
This was unlinked from pull requests Aug 28, 2020
Resolve Optim Params Issues #4877
Merged
Fix checkpoint API on new frontend #4887
Merged
Remove cerberus from wheel package #4919
Merged
This was linked to pull requests Aug 28, 2020
to allow parallel training with mpi4py #4942
Merged
Debuggability #4959
Closed
Port #4920 into the new pytorch frontend #4965
Merged
Fix fetches when eval_step's input is a subset of train_step's input #4966
Merged
@thiagocrepaldi thiagocrepaldi removed a link to a pull request Aug 31, 2020
Add code owners for pytorch frontend #4963
Merged
@thiagocrepaldi thiagocrepaldi linked a pull request Sep 1, 2020 that will close this issue
Update CODEOWNERS and minor docstring fix #5002
Merged
@thiagocrepaldi thiagocrepaldi removed a link to a pull request Sep 1, 2020
to allow parallel training with mpi4py #4942
Merged
This was linked to pull requests Sep 1, 2020
Add flag to _internal_use to control export of contrib ops in ort trainer #4968
Merged
Re-enable CI tests for the new PyTorch frontend #5017
Merged
This was unlinked from pull requests Sep 2, 2020
Fix checkpoint API and improve loss scaler handling #4950
Merged
Port #4920 into the new pytorch frontend #4965
Merged
Fix fetches when eval_step's input is a subset of train_step's input #4966
Merged
This was linked to pull requests Sep 2, 2020
Add warning to non pickable models #5037
Merged
Move onnxruntime.experiment to onnxruntime.training namespace #5045
Merged
@thiagocrepaldi thiagocrepaldi reopened this Sep 3, 2020
@thiagocrepaldi thiagocrepaldi linked a pull request Sep 3, 2020 that will close this issue
Fix unregister pytorch export for contrib ops #5052
Merged
This was unlinked from pull requests Sep 4, 2020
Update CODEOWNERS and minor docstring fix #5002
Merged
Add flag to _internal_use to control export of contrib ops in ort trainer #4968
Merged
Re-enable CI tests for the new PyTorch frontend #5017
Merged
Add warning to non pickable models #5037
Merged
Fix unregister pytorch export for contrib ops #5052
Merged
@thiagocrepaldi thiagocrepaldi reopened this Sep 4, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@thiagocrepaldi thiagocrepaldi

@rayankrish rayankrish

@spandantiwari spandantiwari

Labels
training issues related to ONNX Runtime training; typically submitted using template
Projects
None yet
Milestone
No milestone
3 participants
@thiagocrepaldi @rayankrish @spandantiwari