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

Does torchsparse supports 16-bit training ? #17

Closed
tchaton opened this issue Jan 2, 2021 · 2 comments · Fixed by #69
Closed

Does torchsparse supports 16-bit training ? #17

tchaton opened this issue Jan 2, 2021 · 2 comments · Fixed by #69
Assignees

Comments

@tchaton
Copy link

tchaton commented Jan 2, 2021

Dear All,

I am currently adding Pytorch Lightning support to TorchPoints3d.

If TorchSparse was to support 16-bit and be scriptable, we could have a pretty decent speed up and people would have strong reason to use torchsparse for production.

Here is the PR: https://github.com/nicolas-chaulet/torch-points3d/pull/528

Also, here are updates from our previous release. We support ME and torchsparse backends + possible conversion between them.

Screenshot 2021-01-02 at 13 00 37

Best regards,
Thomas Chaton.

@tchaton tchaton changed the title Does torchsparseu supports 16-bit training ? Does torchsparse supports 16-bit training ? Jan 2, 2021
@zhijian-liu zhijian-liu self-assigned this Jan 3, 2021
@zhijian-liu
Copy link
Contributor

Thanks for the great work! The current version does not support FP16, but we are actively working on incorporating FP16 into the library. We will let you know once we have some results.

@YoushaaMurhij
Copy link

YoushaaMurhij commented Apr 30, 2021

@zhijian-liu. Thanks for your amazing work! Any updates on FP16 support in the library? It would be much faster if we can use model.half() and inputs.half() like in Pytorch.

@zhijian-liu zhijian-liu linked a pull request May 30, 2021 that will close this issue
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 a pull request may close this issue.

3 participants