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.

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

Maxpool_With_Mask #267

Merged
merged 7 commits into from
Dec 30, 2018
Merged

Maxpool_With_Mask #267

merged 7 commits into from
Dec 30, 2018

Conversation

duli2012
Copy link
Member

A custom op that that does maxpool on variable length of input data. For a model we're working on, this op substantially reduces both graph complexity and latency.

@duli2012 duli2012 requested a review from a team as a code owner December 29, 2018 23:07
@tracysh
Copy link
Contributor

tracysh commented Dec 30, 2018

How do we keep these operators from increasing the winml image size? Does this have an application beyond your custom model? I would have thought this is why we have external custom operators.

@duli2012
Copy link
Member Author

duli2012 commented Dec 30, 2018

@tracysh I shared the same concern with you during the discussion for PR229(https://github.com/Microsoft/onnxruntime/pull/229/). At the end, the decision was to add a compilation flag to decide whether the build include contrib ops or not. By external custom op, do you mean wrap custom op in dll? We did consider it, but it's not ready yet. Anyway, the current bar for checking in a custom op is relatively low. But if we change our strategy in the future, I have no problem to undo these check-ins.

@duli2012 duli2012 merged commit 4f49a4a into master Dec 30, 2018
@duli2012 duli2012 deleted the duli/maxpool_with_mask branch December 30, 2018 04:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants