🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/137340

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

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

✅ No Failures

As of commit 65057d6 with merge base a063a82 ():
💚 Looks good so far! There are no failures yet. 💚

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

Is this intended to be merged in? If we can address the root cause that would be the best imo, since we have had issues with too many FX passes where we don’t know which one to look when an error is introduced. I can run a git bisect today to try to pin down the commit.

If we do decide to merge this as a temporary fix, I recommend moving the logic to _fx_passes, where all fx passes live.

The fix is just one pass on the fx graph. It is done inplace. It should be very fast. I can move the function to _fx_passes.py if that's what you mean.

The fix is just one pass on the fx graph. It is done inplace. It should be very fast. I can move the function to _fx_passes.py if that's what you mean.

I am more concerned about complexity and robustness over speed. But since this should be a temporary fix, it should be useful to unblock us for now. I would just make it very clear that it should be removed when the issue is resolved and should be removed before the 2.6 release

I tested that changing to contiguous works.

        with graph.inserting_after(node):
            new_node = graph.call_function(torch.ops.aten.contiguous.default, args=(node,))
            node.replace_all_uses_with(new_node)
            # new_node is replaced as well so we manually revert the replacement
            new_node.update_arg(0, node)
            node.users = {new_node: None}

@pytorchbot merge

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team