🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/140996

• 📄 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.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• [DomainsOnly] Jobs fail with GLIBC version not found

❌ 1 New Failure

As of commit 5a54d8e with merge base eff2217 ():

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

Should we force them to be contiguous or force them to follow the FX strides?

Should we force them to be contiguous or force them to follow the FX strides?

The problem is we don't have the real eager strides. The strides stored in meta['val'] may be the one after optimization for the bwd graph.

Richard also mentioned that having the real eager strides is needed for custom-op. That gives us more motivation to pass the real eager strides to Inductor.

@shunting314 If that's true then we need to avoid changing eager strides I think haha.

If that's true then we need to avoid changing eager strides I think haha.

But we should normally be free to change the strides of saved activations since they are not user visible. Keeping them following eager strides may be too restrictive and leaves little padding/layout-optimization opportunities.

@pytorchbot merge -f

Line error for nested tensors is unrelated.

❌ 🤖 pytorchbot command failed:

@pytorchbot merge: error: argument -f/--force: expected one argument
usage: @pytorchbot merge [-f MESSAGE | -i] [-ic] [-r [{viable/strict,main}]]

Try @pytorchbot --help for more info.

@pytorchbot merge -f "Line error for nested tensors is unrelated."

Merge started

Your change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use -f as last resort and instead consider -i/--ignore-current to continue the merge ignoring current failures. This will allow currently pending tests to finish and report signal before the merge.

Learn more about merging in the wiki.

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

Btw, by default we want custom operators to match their strides in eager mode. Forcing .contiguous() is not the best. This unfortunately broke the vllm-compile integration but at least there's a workaround (setting the tags on the custom ops)