PyTorch 2.5.1

[jeongseok-meta]

Checklist

• Used a personal fork of the feedstock to propose changes
• Bumped the build number (if the version is unchanged)
• Reset the build number to 0 (if the version changed)
• Re-rendered with the latest conda-smithy (Use the phrase @conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• Ensured the license file is being packaged.

[jeongseok-meta]

@conda-forge-admin, please rerender

[conda-forge-admin]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found it was in an excellent condition.

[conda-forge-admin]

Hi! This is the friendly automated conda-forge-webservice.

I tried to rerender for you, but it looks like there was nothing to do.

_{This message was generated by GitHub actions workflow run https://github.com/conda-forge/conda-forge-webservices/actions/runs/11393752191.}

[hmaarrfk]

You might need to update my patch to help find numpy, my suggestion didn't seem like it went through upstream....

[jeongseok-meta]

Submitted pytorch/pytorch#138287 for the nvtx patch

[jeongseok-meta]

Woohoo, it passed the cmake configuration stage and is now in building stage. By the way, I don't have permissions to run several CI jobs. Could I get those permissions or would someone like to take over this PR or create a new one?

[hmaarrfk]

just use azure for now, that CI gets cloggled anyway, then you wait for ever ;)

[hmaarrfk]

if you can get to the 6 hour timeout, we can switch back to the larger runner

[jeongseok-meta]

@conda-forge-admin, please rerender

[jeongseok-meta]

@conda-forge-admin, please rerender

[jeongseok-meta]

@conda-forge-admin, please rerender

[hmaarrfk]

There are other things you have to patch out.

torch 2.5.0.post100 has requirement sympy==1.13.1; python_version >= "3.9", but you have sympy 1.13.3.

[jeongseok-meta]

but that takes ownership away from you, and centralizes it into "me".

This is totally fine to me as long as we can move forward!

[hmaarrfk]

@conda-forge-admin please rerender

[hmaarrfk]

Honestly, feel free to clean up the commigts and force push:

1. Drop all MNT commits from rerendering
2. Squash all the commits you made (and me) into one of your own
3. Leave the commits of other contributors intact (if any)
4. Rerender
5. Force push

Then if you have a few powerful linux machines, you can run the builds yourself.

[jeongseok-meta]

Done! I've squashed all the commits into one with co-authoring.

Also, I've initiated a local build for Linux. Once it's complete, how can I upload the files (if needed)? They will be too large to attach to this PR comment thread.

[Tobias-Fischer]

Should we create an issue in https://github.com/Quansight/open-gpu-server/issues to request an image with more disk space, to avoid having to do CFEP03 in the future?

[jeongseok-meta]

@Tobias-Fischer sounds good! Quansight/open-gpu-server#47

[hmaarrfk]

You should upload them to your own anaconda channel as outlined in CFPE03

[hmaarrfk]

The logs should be uploaded here for me and others to review

[hmaarrfk]

the linker script strikes again!

[jslee02]

Alright, I just finished building Linux packages locally with the linker script patch (which I'm not 100% sure if it's a proper fix, though):

• build log: log_files.zip
• Packages:
  • https://anaconda.org/jslee02/libtorch
  • https://anaconda.org/jslee02/pytorch
  • https://anaconda.org/jslee02/pytorch-cpu
  • https://anaconda.org/jslee02/pytorch-gpu

[hmaarrfk]

I was about to upload but then i realized that the pytorch-cpu package only has a single build....

That said this seems to be a problem with the 2.4.1 package as well and likely for many of the megabuild packages....
https://conda-metadata-app.streamlit.app/?q=conda-forge%2Flinux-64%2Fpytorch-cpu-2.4.1-cpu_mkl_py39h09a6fac_103.conda

lets merge this and find a longer term solution later.

[jeongseok-meta]

OMG, this got merged! 🎉 🎉 Thank you so much for all your help!

For the pytorch-cpu package, what would be the desired outcome?

[hmaarrfk]

That users can install it with ay python version.

Right now it locks them to a single one.

[danpetry]

yeah, if you use pin_subpackage with exact=True, it'll pin to the python version.
you could do

- name: pytorch-{{ "cpu" if cuda_major ==  0 else "gpu" }}
   requirements:
      run:
        - pytorch ={{ version }}={{ "cpu" if cuda_major == 0 else "cuda" }}*

[jeongseok-meta]

@danpetry, thank you for the suggestion! Could you please take a look #282 if it also makes sense to you?

[danpetry]

I will have a look as soon as I can :)