[hacZ17]

Summary

• ROCm HunyuanImage process-kill coverage uses hunyuan_image3_dit_rocm_reliability.yaml, which keeps TP=4 and sets enable_expert_parallel: false.
• CUDA reliability still uses hunyuan_image3_dit.yaml with expert parallel on.
• A separate non-blocking weekly job runs one fault-free 512x512 request on that original expert-parallel deploy. The nightly pixel-accuracy job is unchanged.

Why this is not a duplicate

No open PR splits the HunyuanImage reliability deploy this way. This does not change the Qwen OOM recovery server set in #9425.

Contributor intent

1. What outcome do you want to achieve? Stop the weekly HunyuanImage process-kill matrix from failing in the healthy warmup, while keeping the full-model expert-parallel failure visible.
2. What is the highest-leverage variable for that outcome? Expert parallel on the TP=4 DiT deploy. The weekly cases died in warmup before any process was killed.
3. How will this work move that variable? Point ROCm fault injection at an overlay with expert parallel off, and add one non-blocking weekly baseline that still serves one request with expert parallel on.
4. Whose customer or user needs does this address, and what are those needs? AMD weekly CI owners need process-kill coverage to test fault handling, and they still need a lane that shows whether the expert-parallel runtime can finish a full-model request.

AI assistance disclosure

Cursor, using Grok 4.7, drafted the deploy overlay, the weekly baseline job, and this description. The human contributor directed this containment for the case where only expert-parallel-on fails. The four-case vLLM 0.31/0.32 comparison was not rerun on this machine. Local contract tests passed. The MI300 jobs have not been run at this head.

Test plan

• .venv/bin/python -m pytest tests/buildkite/test_amd_pipeline.py::test_hunyuan_rocm_reliability_disables_expert_parallel tests/buildkite/test_amd_pipeline.py::test_weekly_reliability_is_distinct_from_nightly tests/buildkite/test_amd_suite_selection.py::test_debug_weekly_selects_four_unconditional_jobs -p no:shard --override-ini='addopts=' — 3 passed
• Changed-file pre-commit hooks passed, including Buildkite schema, Ruff, mypy 3.10, and test marks
• On the exact head, run the weekly Hunyuan reliability job and the expert-parallel baseline on MI300

Base: c84638f3208f0fffd872200adfc5860103a6244b
Head: f9efc69a0415547ce9396dd094ab8cf2804309ee

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-10-26T17:40:13.028960Z We couldn’t complete this review. Comment @codex review to try again.	f9efc69	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[vllm-omni-review-bot]

This PR appears to belong to: docs/design/module/vllm_omni_config.md, docs/design/module/entrypoints.md.

Module owners: @alex-jw-brooks @lishunyang12 @NickCao

Routing: @alex-jw-brooks via module of the changed files, CODEOWNERS; @lishunyang12 via module of the changed files, CODEOWNERS; @NickCao via module of the changed files, CODEOWNERS

@hacZ17, please review your own changes and leave a short self-review comment describing what you checked. PRs without author self-review may not be assigned a reviewer.

Please take a look when you have a chance. If you would like an automated review, mention @vllm-omni-review-bot in a comment.

[andyluo7]

Exact-head qualification for a311af7 (rebased onto main 4c51b23):

• CUDA READY #18345 passed with all 18 jobs terminal and no nonzero exit.
• AMD #14888 targeted Hunyuan coverage passed:
  • Reliability job 01a1804c-bc2b-4294-acf4-94c464c81c56: 16 passed; the JUnit guard enforces zero skipped tests; four artifacts uploaded.
  • TP=4 expert-parallel startup job 01a1804c-bc32-4e2c-98ca-1c33b3fd7770: /health returned HTTP 200, 1 passed; the JUnit guard enforces zero skipped tests; four artifacts uploaded.
• Changed-file pre-commit and Buildkite pipeline validation passed locally.
• Full local pytest was not runnable on this Mac because Python 3.14 lacks torch; the matching-head CUDA and ROCm jobs above provide the runtime evidence.

The rebase preserved #9435's AMD pipeline additions and resolved the shared test constant without changing #9428's behavior. No actionable findings remain.

[andyluo7]

Approved at exact head a311af7 after review, required CUDA READY qualification, and targeted AMD Hunyuan reliability plus expert-parallel startup qualification.