[CI] Add Qwen3-omni offline UT

[R2-Y]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Add offline UT for qwen3-omni

Test Plan

pytest -sv tests/multi_stages/test_qwen3_omni.py

Test Result

Pass

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[chatgpt-codex-connector]

The account who enabled Codex for this repo no longer has access to Codex. Please contact the admins of this repo to enable Codex again.

[R2-Y]

@ywang96 Hi Roger, I’m running pytest on an H200 with a tweaked qwen3‑omni config. I’m worried it may not work on A100 or other GPUs. Could you let me know which device types are available on Buildkite? So I can update the CI yaml to test again. Thanks!

[gcanlin]

Maybe we could reuse the conftest.py in #168.

[R2-Y]

Maybe we could reuse the conftest.py in #168.

Sure, I refer your PR, can add one single modality test to conftest.py later after your's merged.

[ywang96]

Maybe it's a good idea to use 2xH100 for qwen3-omni e2e test for now unless we run into resource issues. Feel free to take a look at #221

[R2-Y]

Maybe it's a good idea to use 2xH100 for qwen3-omni e2e test for now unless we run into resource issues. Feel free to take a look at #221

Thanks your suggestion, CI passed

[Gaohan123]

LGTM, thanks!

[hsliuustc0106]

please check with other test PRs and make all of them aligned with each other including the naming, test cases and util functions