Skip to content

[CI] Add DeepGEMM warmup to stage-c-test-deepep-4-gpu#19806

Merged
Kangyan-Zhou merged 1 commit intomainfrom
ci/deepep-4gpu-warmup
Mar 4, 2026
Merged

[CI] Add DeepGEMM warmup to stage-c-test-deepep-4-gpu#19806
Kangyan-Zhou merged 1 commit intomainfrom
ci/deepep-4gpu-warmup

Conversation

@alisonshao
Copy link
Collaborator

Summary

  • The stage-c-test-deepep-4-gpu job times out (20 min) because DeepGEMM JIT compilation consumes ~10 min across 3 server launches during the test
  • The 8-GPU DeepEP test already has warmup steps that avoid this — this PR adds the same pattern to the 4-GPU test
  • Adds two warmup steps before "Run test": warmup_deep_gemm.py and warmup_server.py with lmsys/sglang-ci-dsv3-test:4

Failure example: https://github.com/sgl-project/sglang/actions/runs/22595582756/job/65626168622?pr=19470

Test plan

  • stage-c-test-deepep-4-gpu passes within 20 min timeout
  • Warmup steps complete successfully (should be fast with SGLANG_JIT_DEEPGEMM_FAST_WARMUP=true)

The 4-GPU DeepEP test times out because DeepGEMM JIT compilation
consumes ~10 of the 20-minute test budget across 3 server launches.

Add the same warmup steps already used by stage-c-test-deepep-8-gpu-h200:
- Warmup DeepGEMM JIT Compilation (lmsys/sglang-ci-dsv3-test:4)
- Warmup Server CUDA Graphs (lmsys/sglang-ci-dsv3-test:4)
@gemini-code-assist
Copy link
Contributor

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

@alisonshao
Copy link
Collaborator Author

/rerun-stage stage-c-test-deepep-4-gpu

@github-actions
Copy link
Contributor

github-actions bot commented Mar 3, 2026

✅ Triggered stage-c-test-deepep-4-gpu to run independently (skipping dependencies).

@github-actions
Copy link
Contributor

github-actions bot commented Mar 3, 2026

🔗 View workflow run

@alisonshao
Copy link
Collaborator Author

/rerun-stage stage-c-test-deepep-4-gpu

@github-actions
Copy link
Contributor

github-actions bot commented Mar 3, 2026

✅ Triggered stage-c-test-deepep-4-gpu to run independently (skipping dependencies).

@github-actions
Copy link
Contributor

github-actions bot commented Mar 3, 2026

🔗 View workflow run

@Kangyan-Zhou Kangyan-Zhou merged commit c18cff4 into main Mar 4, 2026
92 of 94 checks passed
@Kangyan-Zhou Kangyan-Zhou deleted the ci/deepep-4gpu-warmup branch March 4, 2026 00:18
Kangyan-Zhou pushed a commit to Kangyan-Zhou/sglang that referenced this pull request Mar 4, 2026
)

Co-authored-by: Alison Shao <alisonshao@Mac.attlocal.net>
qeternity pushed a commit to qeternity/sglang that referenced this pull request Mar 6, 2026
)

Co-authored-by: Alison Shao <alisonshao@Mac.attlocal.net>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants