[Test] L4 complete diffusion feature test for LongCat Image Edit models

[NumberWan]

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

Purpose

This PR adds an L4 diffusion feature expansion e2e test for LongCat-Image-Edit (meituan-longcat/LongCat-Image-Edit), covering the recommended combination acceleration features for LongCat-Image-Edit.

Tests are picked up by nightly: test_*_expansion.py -m 'advanced_model and diffusion and H100'.

The test is intended to be picked up by the L4 nightly diffusion pipeline described in RFC #1832.

Test Plan

The diffusion acceleration doc / RFC #1217 states that LongCat-Image-Edit supports the following features according the recommended combination:

• Module-wise CPU offloading
• Cache-DiT
• Ulysses-SP

Test Coverage for LongCat-Image-Edit (L4, online_serving)

New test file

• tests/e2e/online_serving/test_longcat_image_edit_expansion.py

Model

• meituan-longcat/LongCat-Image-Edit

Cases

Test ID	Feature	Server args (minimal for that feature)
single_card_001	CPU offloading	--enable-cpu-offload (1 GPU)
parallel_001	Ulysses-SP (size 2)	--cache-backend cache_dit --ulysses-degree 2 (2 GPUs)

Test Result [Passed]

• Run:
  • pytest tests/e2e/online_serving/test_longcat_image_edit_expansion.py --collect-only -m diffusion

=============================================================== test session starts ================================================================
platform linux -- Python 3.12.3, pytest-9.0.2, pluggy-1.6.0
rootdir: /home/w00917303/vllm-omni
configfile: pyproject.toml
plugins: anyio-4.12.1
collected 2 items                                                                                                                                  

<Dir vllm-omni>
  <Package tests>
    <Package e2e>
      <Package online_serving>
        <Module test_longcat_image_edit_expansion.py>
          <Function test_longcat_image_edit[single_card_001]>
          <Function test_longcat_image_edit[parallel_001]>

---

Additional Test Result [Passed]

Command :

pytest -s -v tests/e2e/online_serving/test_longcat_image_edit_expansion.py \
  -m "diffusion and advanced_model and H100" \
  --collect-only

=============================================================== test session starts ================================================================
platform linux -- Python 3.12.3, pytest-9.0.2, pluggy-1.6.0 -- /home/w00917303/.venv/bin/python3
cachedir: .pytest_cache
rootdir: /home/w00917303/vllm-omni
configfile: pyproject.toml
plugins: anyio-4.12.1
collected 2 items                                                                                                                                  

<Dir vllm-omni>
  <Package tests>
    <Package e2e>
      <Package online_serving>
        <Module test_longcat_image_edit_expansion.py>
          Comprehensive tests of diffusion features that are available in online serving mode
          and are supported by the following model:
          - LongCat-Image-Edit: image-to-image edit with a single image + single edit prompt input
          Coverage:
          - CPU offloading (model-level sequential offload via --enable-cpu-offload)
          - Cache-DiT
          - SP (Ulysses)
          
          This validates:
           - Successful image generation at the expected 1024x1024 resolution with recommended feature combinations
          <Function test_longcat_image_edit[single_card_001]>
            Test the recommended feature combinations for LongCat-Image-Edit.
          <Function test_longcat_image_edit[parallel_001]>
            Test the recommended feature combinations for LongCat-Image-Edit.

CI Test Result [Passed]

CI Successful in both 2 tests

• Test_longcat_image_edit[single_card_001]
• Test_longcat_image_edit[parallel_001]

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d89dacc452

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[lishunyang12]

Left a few comments. Main concern is the gap between what the docstring claims and what's actually tested.

[NumberWan]

@lcukyfuture PTAL

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d89dacc452

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[lishunyang12]

LGTM

[NumberWan]

@congw729 @yenuo26 @lcukyfuture @fhfuih @wtomin PTAL

All the local test and CI test has been passed.

[fhfuih]

LGTM. It is not a high-priority model, so these test cases shall be good enough for now.

[Gaohan123]

LGTM. Thanks