PyTorch fixes CI bug: doc previews now upload even on partial failure

PyTorch, the popular deep learning framework with over 100,000 GitHub stars, has merged a critical fix for its continuous integration pipeline. The commit, tagged as trunk/17d734d460dfaf83a98b6874b8d7fe7cc9196c99, addresses a condition in the "upload-docs-preview" job within the repository's workflow. Previously, the job would skip uploading documentation preview artifacts if any other job in the same workflow had failed. This behavior prevented developers from accessing generated previews of documentation changes when unrelated tests or builds had issues.

The fix, contributed by developer malfet and approved by PyTorch maintainers ngimel and huydhn, modifies the condition to always attempt uploading documentation artifacts regardless of the overall workflow status. This change ensures that even when some jobs fail (e.g., due to flaky tests or hardware timeouts), the documentation preview remains accessible. The update was made via pull request #185685 and tagged on May 30. While seemingly minor, this fix improves developer productivity and CI reliability for one of the largest open-source AI projects.