Support specifying a torch range

[oraluben]

Motivation: we're using a custom pytorch build and have patch release more recently than upstream. While those updates do not change any abi, we want to allow torchvision/torchaudio (pytorch/audio#3886) to tolerance the changed pytorch version when we update pytorch, otherwise pip will raise a warning.

In this PR, we add two new envionment variable to generate a version range for pytorch.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/8936

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 4 New Failures, 8 Pending, 1 Unrelated Failure

As of commit 72b8fc4 with merge base d28001e ():

NEW FAILURES - The following jobs have failed:

• Tests / unittests-linux (3.10, linux.12xlarge, cpu) / linux-job (gh)
  RuntimeError: Command docker exec -t bf2caf615ed1d9fee58b2c90d49bbd949ca33c1175f6c203219acaf90c0f0e09 /exec failed with exit code 1
• Tests / unittests-linux (3.11, linux.12xlarge, cpu) / linux-job (gh)
  RuntimeError: Command docker exec -t e8dc7bc46c27085b3a4bce001532b26f1ece127aab6d979d1aee4c37779b380e /exec failed with exit code 1
• Tests / unittests-linux (3.12, linux.12xlarge, cpu) / linux-job (gh)
  RuntimeError: Command docker exec -t f5fa85c4ac0109908a14a05c531db0719c26cf6d03cc33f5a2d56de276235c1d /exec failed with exit code 1
• Tests / unittests-linux (3.9, linux.12xlarge, cpu) / linux-job (gh)
  RuntimeError: Command docker exec -t caae84ae187f05c5b1eca38ccc58438073cadd011aa6843269e2be817057596e /exec failed with exit code 1

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• Tests / unittests-windows (3.9, windows.g5.4xlarge.nvidia.gpu, cuda, 11.8) / windows-job (gh) (trunk failure)
  AttributeError: '_OpNamespace' 'image' object has no attribute '_jpeg_version'

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[NicolasHug]

Hi @oraluben , is it an option for you to just disable / ignore the pip warning?

I sympathize with the release challenges, I wouldn't be able to guarantee that the PYTORCH_VERSION_GE and PYTORCH_VERSION_LT survive future refacftorings

[oraluben]

@NicolasHug Thanks for your reply!

For us we want to keep pip dependency resolving working (for example, torchvision==0.21 -> torch>=2.6,<2.7 -> torch==2.6.0.x where x is our minor version. If we want to live with pip warning, the only way to install torchvision with latest torch is to install torchvision (with torch 2.6) first, than install torch 2.6.0.x with a second pip install, which is not we want.

I wouldn't be able to guarantee that the PYTORCH_VERSION_GE and PYTORCH_VERSION_LT survive future refacftorings

Ofc it would be best to get this merged in upstream, I think our scenerio would also be a common use case for some other users/companies. Otherwise we might have to maintain a small patch locally...

[NicolasHug]

Alright. I'll merge this because maintenance cost is hopefully minimal, but as I wrote in c8b2ccf, we won't necessarily be able to support this forever - this is only on a best-effort basis. Thanks for the PR @oraluben