Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

[v1.x] ONNX export support for RNN and sum_axis #20226

Merged
merged 6 commits into from
Apr 30, 2021
Merged

[v1.x] ONNX export support for RNN and sum_axis #20226

merged 6 commits into from
Apr 30, 2021

Conversation

waytrue17
Copy link
Contributor

Description

Add onnx conversion logic for RNN and sum_axis. Add unittest for the two.

Checklist

Essentials

  • PR's title starts with a category (e.g. [BUGFIX], [MODEL], [TUTORIAL], [FEATURE], [DOC], etc)
  • Changes are complete (i.e. I finished coding on this PR)
  • All changes have test coverage
  • Code is well-documented

Changes

  • Feature1, tests, (and when applicable, API doc)
  • Feature2, tests, (and when applicable, API doc)

Comments

  • If this change is a backward incompatible change, why must this change be made.
  • Interesting edge cases to note here

@waytrue17 waytrue17 requested a review from szha as a code owner April 28, 2021 21:49
@mxnet-bot
Copy link

Hey @waytrue17 , Thanks for submitting the PR
All tests are already queued to run once. If tests fail, you can trigger one or more tests again with the following commands:

  • To trigger all jobs: @mxnet-bot run ci [all]
  • To trigger specific jobs: @mxnet-bot run ci [job1, job2]

CI supported jobs: [edge, unix-gpu, centos-cpu, unix-cpu, clang, centos-gpu, windows-gpu, windows-cpu, miscellaneous, sanity, website]

Note:
Only following 3 categories can trigger CI :PR Author, MXNet Committer, Jenkins Admin.
All CI tests must pass before the PR can be merged.

@mseth10 mseth10 added the pr-awaiting-testing PR is reviewed and waiting CI build and test label Apr 28, 2021
@szha szha requested a review from Zha0q1 April 28, 2021 21:50
@mseth10 mseth10 added pr-work-in-progress PR is still work in progress and removed pr-awaiting-testing PR is reviewed and waiting CI build and test labels Apr 28, 2021
Zha0q1
Zha0q1 reviewed Apr 28, 2021
View reviewed changes
tests/python-pytest/onnx/test_operators.py
@pytest.mark.parametrize('input_size', [16, 32, 64])
@pytest.mark.parametrize('num_layers', [1, 2])
@pytest.mark.parametrize('batch_size', [1, 2, 4])
@pytest.mark.parametrize('seq_length', [16, 32])
Copy link
Contributor

Choose a reason for hiding this comment

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

why removing 32?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Saw the same assertion issue with large seq_length/state_size/input_size. Will need to further decrease one of them to pass rnn, just like what we did before for lstm and gru.

tests/python-pytest/onnx/test_operators.py
else:
op_export_test('rnn', M, [x, param, state], tmp_path)
op_export_test('rnn', M, [x, param, state], tmp_path, atol=1e-2)
Copy link
Contributor

Choose a reason for hiding this comment

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

Do we know how large was the difference?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

From my local runs, the diff is between 1e-3 to 1e-2.

@mseth10 mseth10 added pr-awaiting-testing PR is reviewed and waiting CI build and test pr-work-in-progress PR is still work in progress and removed pr-work-in-progress PR is still work in progress pr-awaiting-testing PR is reviewed and waiting CI build and test labels Apr 28, 2021
@mseth10 mseth10 added pr-awaiting-testing PR is reviewed and waiting CI build and test pr-work-in-progress PR is still work in progress and removed pr-work-in-progress PR is still work in progress pr-awaiting-testing PR is reviewed and waiting CI build and test labels Apr 29, 2021
@mseth10 mseth10 added pr-awaiting-testing PR is reviewed and waiting CI build and test pr-work-in-progress PR is still work in progress pr-awaiting-review PR is waiting for code review and removed pr-work-in-progress PR is still work in progress pr-awaiting-testing PR is reviewed and waiting CI build and test labels Apr 29, 2021
josephevans
josephevans reviewed Apr 29, 2021
View reviewed changes
python/mxnet/onnx/mx2onnx/_op_translations/_op_translations_opset12.py Outdated
"""Map MXNet's sum_axis operator.
sum_axis is equivalent to sum in MXNet
"""
return convert_sum(node, **kwargs)
Copy link
Contributor

Choose a reason for hiding this comment

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

Instead of wrapping convert_sum() here, couldn't we just add the decorator to register "sum_axis" to convert_sum() ?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Registered it under convert_sum. Thanks for the good point!

@mseth10 mseth10 added pr-awaiting-testing PR is reviewed and waiting CI build and test and removed pr-awaiting-review PR is waiting for code review labels Apr 29, 2021
josephevans
josephevans approved these changes Apr 29, 2021
View reviewed changes
Copy link
Contributor

@josephevans josephevans left a comment

Choose a reason for hiding this comment

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

LGTM, thanks!

@mseth10 mseth10 added pr-awaiting-review PR is waiting for code review and removed pr-awaiting-testing PR is reviewed and waiting CI build and test labels Apr 30, 2021
@Zha0q1 Zha0q1 merged commit 2127c3e into apache:v1.x Apr 30, 2021
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Reviewers

@Zha0q1 Zha0q1 Zha0q1 left review comments

@josephevans josephevans josephevans approved these changes

@szha szha Awaiting requested review from szha szha is a code owner

Assignees
No one assigned
Labels
pr-awaiting-review PR is waiting for code review
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
5 participants
@waytrue17 @mxnet-bot @josephevans @Zha0q1 @mseth10