[CI][ROCm] Add LingBot nightly coverage

[haic0]

Purpose

Add a dedicated non-blocking AMD nightly job for the dense robbyant/lingbot-video-dense-1.3b CFG-off smoke tests/e2e/online_serving/test_lingbot_video_expansion.py::test_cfg_off.

The job uses one MI300 (mi300_1), a 90-minute timeout, the exact full_model and diffusion and rocm and MI325 and cards_1 selector, fail-closed collect-only preflight, and checkout-relative JUnit staging under artifacts/rocm-lingbot-nightly/. The shared AMD MiniJinja template supplies depends_on: amd-build.

Why this is not a duplicate

Open and merged PR searches found no existing LingBot AMD nightly coverage. #7892 adds complementary CUDA routing, while #7933 and #7934 cover different model families.

Final rebase and validation

• Target base: aad30908772a8188c17997341d62768dc0bab3b7
• Exact head: 277182032a650d7a9e7f2d0f0522f841b14bf3e2
• Linear history: one commit directly on the target base, with no merge commit.
• The old and final net diffs have identical stable patch IDs; no source behavior changed during the rebase.
• Isolated AMD pipeline and suite-selection tests: 27 passed.
• Changed-file pre-commit: passed, including Ruff, mypy, test-mark checks, YAML, and Buildkite schema validation.
• git diff --check, Unicode format-control scan, and DCO/trailer audit: passed.
• Commit trailers include Assisted-by: Cursor and Signed-off-by: haic0 <149741444+haic0@users.noreply.github.com>.

ROCm eligibility remains limited to test_cfg_off; test_batch_cfg_extra_params remains CUDA-only. Exact model collection and MI300 execution were not rerun because this lightweight environment lacks vLLM. No exact-head GPU/model run is claimed.

[vllm-omni-review-bot]

This PR was classified as CI work.

CI owner: @yenuo26 @congw729 @NickCao

Routing: @yenuo26 via semantic router, CI owner, CODEOWNERS; @congw729 via CODEOWNERS; @NickCao via CODEOWNERS

@haic0, 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.

[haic0]

Self-review

Updated after the true rebase. I reviewed every final changed line on exact head 277182032a650d7a9e7f2d0f0522f841b14bf3e2 against frozen upstream main target aad30908772a8188c17997341d62768dc0bab3b7.

• History is linear: one commit, no merge commits, and the commit parent is the exact target base.
• The old and final net diffs have identical stable patch IDs; no source behavior changed during the rebase.
• The LingBot job remains one-GPU MI300, non-blocking, 90 minutes, exact-node/fail-closed collection, and checkout-relative JUnit staging.
• ROCm eligibility remains limited to test_cfg_off; test_batch_cfg_extra_params remains CUDA-only.
• Validation: isolated AMD pipeline/suite tests 27 passed; changed-file pre-commit passed, including Buildkite schema; diff check, Unicode format-control scan, and exact DCO/AI trailers passed.
• No new model-specific example, code-quality fragility pattern, or evidence-backed simplification issue was found in the diff.

Residual validation is an exact-head MI300 checkpoint run. The local environment lacks vLLM, so exact model collection and GPU execution were not rerun.

[vllm-omni-review-bot]

Omni ReviewBot: no human activity for 7 days

@haic0 this pull request has had no human commit, comment or review since 2026-09-22. Please confirm the current plan and next step. The author or a maintainer decides whether to change the PR state.

To keep it moving, any one of these is enough: push an update, reply to the open blocker, or post the current plan and timeline.

[haic0]

Addressed all four review comments in 6b2581e5 and rebased onto current main (5af3a165).

• Removed source depends_on; MiniJinja render confirms the template still injects amd-build.
• Matched the direct native AMD command/artifact pattern from [CI][ROCm] Add Cosmos3 nightly coverage #7933/[CI][ROCm] Add and stabilize HunyuanImage3 nightly coverage #7934 and retained exact fail-closed collection.
• Removed the LingBot job-specific pipeline unit test without adding a replacement.
• Consolidated CUDA+ROCm marks for test_cfg_off at file scope while keeping the sibling test CUDA-only.

Validation: generic AMD tests 26 passed; whole-file marker collection produced ROCm=1 (test_cfg_off only), CUDA H100=2, CUDA B200=2; MiniJinja render/YAML semantic checks passed; changed-file pre-commit and git diff --check passed. No MI300 was available for a real checkpoint run.

[yenuo26]

LGTM