pytorch - 💡(How to fix) Fix Update install matrix on pytorch.org to reflect CUDA 12.8 deprecation [2 comments, 3 participants]

TL;DR

Update the install matrix on the PyTorch website to remove CUDA 12.8 (cu128) as a supported option for nightly builds.

Guidance

• Verify the current status of CUDA 12.8 nightly builds to confirm they are indeed stale or deprecated.
• Check the build pipeline configuration to determine why cu128 nightly builds have stopped.
• Update the install matrix on the PyTorch website to reflect the current supported CUDA versions.
• Consider adding a note or comment to the install matrix indicating that cu128 is no longer supported.

Example

No code snippet is necessary for this issue, as it involves updating documentation rather than code.

Notes

The solution assumes that the cu128 nightly builds are indeed stale or deprecated, as evidenced by the lack of new builds in over a week. However, it's possible that the build pipeline is temporarily down or experiencing issues, so verification is necessary before making changes to the install matrix.

Recommendation

Apply workaround: Update the install matrix to remove cu128 as a supported option, as it appears to be no longer maintained. This will help prevent confusion and ensure that users are directed to supported versions of CUDA.