Manually update vllm CI pin.

[zhxchen17]

Summary:

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

Fixes #ISSUE_NUMBER

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/166494

• 📄 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

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:

• Replace all macOS instances with nextjs due CVE-2025-55182

✅ You can merge normally! (1 Unrelated Failure)

As of commit 55c770c with merge base 4710fd9 ():

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• inductor / inductor-test / test (inductor_torchbench, 2, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (similar failure)
  shufflenet_v2_x1_0

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

[huydhn]

Thank you for the fix!

[zou3519]

Looks like we got problems. the vllm_basic_correctness_test appears to be a compile failure. Given it succeeds in vLLM CI I suspect a compile issue

[desertfire]

Should we make the pin update automatic?

[huydhn]

Should we make the pin update automatic?

It's automated, e.g. #165274, but that PR has been blocked for 2 weeks now. So, we end up with this manual update to coordinate the fix from vLLM and PyTorch

[zhxchen17]

Update: I tried to repro the test failures locally and seems 80% of tests (e.g. lora ones) stuck on my local and never finish (maybe I setup something wrong).

For the remaining 20% tests I can repro the failures, and it's still there when I turn off VLLM_USE_AOT_COMPILE, indicating it's more likely general compiler/vllm issues on the trunk. I can keep an eye on this but maybe we should get someone else to take a look for now.

[zhxchen17]

Need the following fixes to pytorch/vllm landed to restore trunk CI to green:

• [aot_compile]change VLLM backend to read fake args from example_value vllm-project/vllm#29104
  • For fixing https://github.com/pytorch/pytorch/actions/runs/19786099631/job/56695169795?pr=166494
• [aot_compile] Pass additional globals to callable #169070 & [bugfix] Pass globals to aot_compiled function vllm-project/vllm#29428
  • For fixing https://github.com/pytorch/pytorch/actions/runs/19794165965/job/56713564450?pr=166494
• [multimodal][test] Reduce memory utilization for test_siglip to avoid OOM  vllm-project/vllm#29504
  • For fixing https://github.com/pytorch/pytorch/actions/runs/19794165965/job/56713564429?pr=166494

[huydhn]

I have temporarily marked vLLM trunk jobs as unstable in #169298 because we now need this change from vLLM vllm-project/vllm#29588 to fix an upstream issue from HuggingFace. Let's aim to update the pinned commit to latest vLLM main in the next couple of days

[huydhn]

@pytorchbot merge -f 'All vLLM tests are passing now, no need to run trunk jobs'

[pytorchmergebot]

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

Advanced Debugging

Check the merge workflow status
here