Skip to content

Add Split-18: new num_outputs attribute #4389

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.

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

Merged
merged 0 commits into from
Aug 18, 2022
Merged

Conversation

p-wysocki
Copy link
Contributor

Signed-off-by: p-wysocki przemyslaw.wysocki@intel.com

Description

  • New optional Split attribute num_outputs determining the number of equally sized output tensors.

Motivation and Context

@postrational
Copy link
Contributor

Previously discussed in #4358

@p-wysocki p-wysocki requested a review from postrational July 28, 2022 16:21
@@ -17,7 +17,8 @@ def export_1d() -> None:
'Split',
inputs=['input'],
outputs=['output_1', 'output_2', 'output_3'],
axis=0
axis=0,
num_outputs=3
Copy link
Contributor

Choose a reason for hiding this comment

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

I think we should perhaps have test-cases for both the earlier version and newer version? @jcwchen ?

Copy link
Member

@jcwchen jcwchen Aug 2, 2022

Choose a reason for hiding this comment

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

After this PR: #3855, going forward all of node test cases for the same operator will be updated to the latest opset version if there is an opset bump. If people would like to keep some old test cases for old opset versions, they need to specify the opset_version while creating node test cases. Take Loop op as an example, see this line.

Copy link
Contributor

Choose a reason for hiding this comment

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

I believe that we will need to copy the old test-case, but add the opset version to be the old one (and the new one can be as in the current PR)?

Copy link
Member

Choose a reason for hiding this comment

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

Correct. In that case we can keep the same test cases for both newer and older opset versions.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I applied the changes you suggested. Opset13 tests were reverted, new (opset18) tests stayed the same.

@jcwchen
Copy link
Member

jcwchen commented Aug 18, 2022

For unknown reason (might be a GitHub server issue), this PR hasn't been merged into the main branch even after 1 hrs... Let's see whether it will be merged toady.

@gramalingam
Copy link
Contributor

As per @p-wysocki , it is unclear why this (empty) merge even happened. He says he did nothing to trigger it. And it seems to be an empty merge with no commits. What's happening? Seems mysterious to me.

@p-wysocki
Copy link
Contributor Author

I was 1 commit ahead of onnx:main, it contained all my changes, squashed with a sign off. Unfortunately some other commits got in and there were ~350 files changed. Long story short, I reverted this one, single commit from p-wysocki:add_split18 and pushed it to my fork, essentially syncing onnx:main with p-wysocki:add_split18, what can be seen as "Files changed: 0". I think this may be just a weird behavior from GitHub, I did not enable merging (merge should not be possible at all anyway), my changes are not on master and it would be just a "visual" glitch.

We had this happen in OpenVINO too, where an external contributor seemingly merged his changes, but they were not on master and the PR showed 0 changed files.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
Status: Done
Development

Successfully merging this pull request may close these issues.

5 participants