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

File Limit Request: torch - 1000 MB #2341

Closed
3 tasks done
seemethere opened this issue Oct 27, 2022 · 3 comments
Closed
3 tasks done

File Limit Request: torch - 1000 MB #2341

seemethere opened this issue Oct 27, 2022 · 3 comments

Comments

@seemethere
Copy link

Project URL

https://pypi.org/project/torch

Does this project already exist?

  • Yes

New Limit

1000

Update issue title

  • I have updated the title.

Which indexes

PyPI

About the project

https://pytorch.org

Reasons for the request

We have upgraded CUDA versions to 11.7 and with that the binary size of torch has grown to 850+MB, in order to continue to support releases on PyPI We're requesting a limit increase to 1000MB in order to avoid having to request size increases in the future.

We're anticipating that CUDA support will continue to be important to multiple users so the size increase should be a net benefit to consumers on the PyPI platform for those who would like to use PyPI as the dependency platform for their machine learning systems.

Related issues: pytorch/pytorch#53336

Code of Conduct

  • I agree to follow the PSF Code of Conduct
@seemethere
Copy link
Author

cc @di

@rgommers
Copy link

The PyTorch 1.13.0 release came out 3 days ago and had manylinux1 wheels of ~890 MB, so it looks like someone bumped the limit (thank you!).

Let me add a bit of context for the request:

  • The 1 GB requested here is what was declared the max file size limit that PyPI maintainers are willing to support in https://discuss.python.org/t/what-to-do-about-gpus-and-the-built-distributions-that-support-them/7125
  • For a long time PyTorch was stuck on CUDA 10.2, because the CUDA 11.x wheels ended up being in the 1.2-1.3 GB size range
  • The manylinux standard and auditwheel didn't allow not bundling in some large shared libraries
    • An auditwheel --exclude option to exclude libraries (so package authors can put those in separate wheels if they really have to) was finally merged last week: Add --exclude option to auditwheel repair pypa/auditwheel#368, so it may be possible in the future to unbundle the PyTorch wheels a bit. This is not recommended still, and technically does break the manylinux spec - but the understanding is that package authors are allowed to take that responsibility.

It seems like a change was made; it'd be great to have that made "official" by closing this issue and commenting on the new limit.

@cmaureir
Copy link
Member

Hey @seemethere and @rgommers thanks for the detailed explanation.
I've set the upload limit for torch to 1000 MB on PyPI. Please be mindful of the frequency of releases at that size.

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

No branches or pull requests

3 participants