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

Autograd Function Fallback bug fix - moe support #8105

Merged
merged 11 commits into from
Jul 7, 2021
Merged

Autograd Function Fallback bug fix - moe support #8105

merged 11 commits into from
Jul 7, 2021

Conversation

pengwa
Copy link
Contributor

@pengwa pengwa commented Jun 20, 2021 •
edited
Loading

Description: Autograd Function Fallback bug fix - moe support

  • Support forward inputs orders like "Non_tensor/Tensor/Non_tensor". Correspondingly, support "None/Tensor_Grad/None" for backward outputs.
  • Report RuntimeError when PythonOp detected but _enable_custom_autograd_function is NOT enabled.
  • Simplify the attributed used by PythonOpGrad. Renaming some attribute to reflect whether it is used for tensor inputs or all (tensor + non-tensor) inputs.

Attached the PythonOP/PythonOpGrad schemas (and how those attributes used) after the change:

image

Motivation and Context

  • Why is this change required? What problem does it solve?
  • If it fixes an open issue, please link to the issue here.

@pengwa
Support forward inputs orders like "Non_tensor/Tensor/Non_tensor". Co…
cfeaad9 
…rrespondingly, support "None/Tensor_Grad/None" fpr backward outputs.
@pengwa
Report RuntimeError when PythonOp detected but _enable_custom_autogra…
446d66b 
…d_function is enabled.
@pengwa
Fix "PoliCheck ] - Defect : Term "hang", Component : orttraining\ortt…
3ed9e6f 
…raining\python\training\ortmodule\__init__.py (1 issue)"
@pengwa pengwa requested a review from tlh20 June 20, 2021 12:22
@pengwa pengwa requested a review from nbcsm June 20, 2021 12:22
tlh20
tlh20 requested changes Jul 2, 2021
View reviewed changes
Copy link
Contributor

@tlh20 tlh20 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A few minor comments on places we could extend assertion checks. Adjacent to some of the changes in this PR there is a use of "static" that looks suspicious -- could you check if that is correct?

onnxruntime/core/language_interop_ops/torch/torch_proxy.cc Outdated Show resolved Hide resolved
onnxruntime/core/language_interop_ops/torch/torch_proxy.cc Outdated Show resolved Hide resolved
orttraining/orttraining/core/graph/gradient_builder.cc Show resolved Hide resolved
orttraining/orttraining/core/graph/gradient_builder.cc Outdated Show resolved Hide resolved
orttraining/orttraining/core/graph/training_op_defs.cc Outdated Show resolved Hide resolved
orttraining/orttraining/core/graph/training_op_defs.cc Outdated Show resolved Hide resolved
orttraining/orttraining/core/graph/training_op_defs.cc Outdated Show resolved Hide resolved
orttraining/orttraining/core/graph/training_op_defs.cc Outdated Show resolved Hide resolved
orttraining/orttraining/training_ops/cpu/torch/torch_custom_function_kernel_base.cc Outdated Show resolved Hide resolved
orttraining/orttraining/training_ops/cpu/torch/torch_custom_function_kernel_base.cc Outdated Show resolved Hide resolved
@pengwa @tlh20
Update orttraining/orttraining/core/graph/training_op_defs.cc
7ed72bc 
Co-authored-by: Tim Harris <tiharr@microsoft.com>
@SherlockNoMad SherlockNoMad added the training issues related to ONNX Runtime training; typically submitted using template label Jul 2, 2021
SherlockNoMad
SherlockNoMad previously approved these changes Jul 2, 2021
View reviewed changes
pengwa and others added 3 commits July 3, 2021 10:45
@pengwa @tlh20
Apply suggestions from code review
50dba46 
Refine the schema description

Co-authored-by: Tim Harris <tiharr@microsoft.com>
@pengwa
Resolve review comments
8070c04
@pengwa
Merge branch 'pengwa/autograd_moe' of https://github.com/microsoft/on…
dc94b4b 
…nxruntime into pengwa/autograd_moe
@pengwa
Copy link
Contributor Author

pengwa commented Jul 3, 2021

A few minor comments on places we could extend assertion checks. Adjacent to some of the changes in this PR there is a use of "static" that looks suspicious -- could you check if that is correct?

Thanks @tlh20 for your time reviewing , I have addressed all of comments . :)

@pengwa pengwa merged commit 2347a0a into master Jul 7, 2021
@pengwa pengwa deleted the pengwa/autograd_moe branch July 7, 2021 00:58
@garymm garymm mentioned this pull request Jun 16, 2022
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@wschin wschin wschin left review comments

@thiagocrepaldi thiagocrepaldi thiagocrepaldi left review comments

@baijumeswani baijumeswani baijumeswani approved these changes

@tlh20 tlh20 tlh20 approved these changes

@SherlockNoMad SherlockNoMad SherlockNoMad left review comments

@BowenBao BowenBao Awaiting requested review from BowenBao

@liqunfu liqunfu Awaiting requested review from liqunfu

@mrry mrry Awaiting requested review from mrry

@spandantiwari spandantiwari Awaiting requested review from spandantiwari

@nbcsm nbcsm Awaiting requested review from nbcsm

Assignees
No one assigned
Labels
training issues related to ONNX Runtime training; typically submitted using template
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
6 participants
@pengwa @wschin @thiagocrepaldi @SherlockNoMad @baijumeswani @tlh20