Skip to content

Conversation

@spandantiwari
Copy link

Currently, constant folding pass during ONNX conversion removes all onnx::Constant nodes that are parents of nodes that are folded. In situations where the parent onnx::Constant node is other subscribers downstream this could be a problem. This change updates the removal logic to remove to only those onnx::Constant nodes that do not have other subscribers downstream

@pytorchbot pytorchbot added oncall: jit Add this issue/PR to JIT oncall triage queue module: onnx Related to torch.onnx labels May 3, 2019
@spandantiwari
Copy link
Author

cc: @houseroad

@spandantiwari
Copy link
Author

@houseroad - The failure doesn't seem related to this change.

@spandantiwari
Copy link
Author

@houseroad - It passed on rerun now.

@ezyang ezyang requested a review from houseroad May 6, 2019 14:39
Copy link
Member

@houseroad houseroad left a comment

Choose a reason for hiding this comment

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

Looks good.

Actually, we can add a constant fusion pass in onnx exporter. So the constants with the same value can be merged. :-)

Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

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

@houseroad has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

@spandantiwari
Copy link
Author

@houseroad - that sounds like a good idea.

@facebook-github-bot
Copy link
Contributor

@houseroad merged this pull request in 838ada3.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Merged module: onnx Related to torch.onnx oncall: jit Add this issue/PR to JIT oncall triage queue open source

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants