[Bugfix] Update Flux2-dev & Dynin_omni L4 e2e test

[wtomin]

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

Purpose

1. Solving [CI Failure]: Diffusion · Other · Function Test with L4, test_flux_2_dev_expansion.py, Model loading out of memory #2722. The main reason is that Flux.2-dev model parameters are too large to fit in L4 (24GB).

This PR:

• changes the hardware from L4 to H100;
• remove cache-dit argument, because Flux.2-dev does not support cache-dit yet.

2. Solving [CI Failure]: Function Test with L4, test_dynin_omni_expansion.py, Model loading out of memory #2721. The major reason is the Dynin_omni model weights are too large to fit in L4 card.

This PR:

• changes the hardware from L4 to H100;

3. Additional edit: enlarge diffusion perf test baseline threshold

In buildkite test , it indicates the threshold for test_qwen_image_ulysses2_cfg2_vae_patch4 is a little too tight, which can cause failure occasionally.

This PR enlarges the baseline threshold from 2.34 to 2.7.

Test Plan

Run it in nightly CI.

pytest -s -v tests/e2e/online_serving/test_*_expansion.py -m "advanced_model and diffusion and L4" --run-level "advanced_model"

Test Result

Nightly tests were all passed. See results in this url.

---

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. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
• (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]

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.

[yenuo26]

If this requires H100, I think tests/e2e/offline_inference/test_dynin_omni.py needs to be modified accordingly.

[wtomin]

Will update it in a new PR.

[hsliuustc0106]

Blocker scan:

Category	Result
Correctness	PASS
Reliability/Safety	PASS
Breaking Changes	PASS
Test Coverage	PASS
Documentation	PASS
Security	PASS

OVERALL: NO BLOCKERS

VERDICT: COMMENT

Test-only PR: hardware changes (L4→H100) for Flux2-dev & Dynin_omni due to OOM, removed cache-dit arg, perf baseline relaxation 2.34→2.7. Gates pass. Buildkite PENDING (expected - this PR changes test hardware).

Non-blocking: Please paste nightly CI results in Test Result section once buildkite completes.

[wtomin]

Nightly tests were all passed. See results in url.

[gcanlin]

LGTM