[Test] L4 complete diffusion feature test for Bagel models

[NumberWan]

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

Purpose

This PR adds L4 test for Bagel model(ByteDance-Seed/BAGEL-7B-MoT),
covering all acceleration features currently required by the Model×Feature RFC for Bagel.

Tests are picked up by nightly: test_*_expansion.py -m 'advanced_model and diffusion and H100'.
The tests are intended to be picked up by the L4 nightly diffusion pipeline described in RFC #1832.

Test Plan

The diffusion acceleration doc / #1217 states that Bagel support the following features:

• TeaCache
• Cache-DiT
• CFG-Parallel
• Tensor-Parallel

Test Coverage for Bagel (L4, online_serving)

New test file

• tests/e2e/online_serving/test_bagel_expansion.py

Model

• ByteDance-Seed/BAGEL-7B-MoT

Cases

Test ID	Feature	Server args (minimal for that feature)
single_card_teacache	TeaCache	--cache-backend tea_cache (1 GPU)
single_card_cache_dit	Cache-DiT	--cache-backend cache_dit (1 GPU)
parallel_cfg_2	CFG-Parallel size 2	--cache-backend tea_cache --cfg-parallel-size 2 (2 GPUs)
parallel_tp_2	Tensor-Parallel size 2	--cache-backend cache_dit --tensor-parallel-size 2 (2 GPUs)

Test Result

• pytest tests/e2e/online_serving/test_bagel_expansion.py --collect-only -m diffusion

---

tests/e2e/online_serving/test_bagel_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 4 items                                                                            

<Dir vllm-omni>
  <Package tests>
    <Package e2e>
      <Package online_serving>
        <Module test_bagel_expansion.py>
          <Function test_bagel[single_card_teacache]>
          <Function test_bagel[single_card_cache_dit]>
          <Function test_bagel[parallel_cfg_2]>
          <Function test_bagel[parallel_tp_2]>

Additional Test Result

Command:

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

collected 4 items
<Dir vllm-omni>
 <Package tests>
   <Package e2e>
     <Package online_serving>
       <Module test_bagel_expansion.py>
         L4 diffusion feature expansion tests for Bagel.
         Coverage:
         - TeaCache
         - Cache-DiT
         - CFG-Parallel
         - Tensor-Parallel
         <Function test_bagel[single_card_teacache]>
         <Function test_bagel[single_card_cache_dit]>
         <Function test_bagel[parallel_cfg_2]>
         <Function test_bagel[parallel_tp_2]>

CI Test Result

CI Successful in both 4 test

Running test: test_bagel[single_card_teacache]
Running test: test_bagel[single_card_cache_dit]
Running test: test_bagel[parallel_cfg_2]
Running test: test_bagel[parallel_tp_2]
 

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 Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1ca862ce99

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".

[NumberWan]

@congw729 @yenuo26 PTAL

[hsliuustc0106]

@princepride PTAL

[hsliuustc0106]

what's the expected result for these tests?

[NumberWan]

The expected result is that for each configuration (TeaCache / Cache-DiT / CFG-Parallel / TP2), Bagel can successfully generate images via online serving, and assert_diffusion_response verifies that:

• images are generated without errors
• the number of images equals num_outputs_per_prompt
• the resolution matches (height=512, width=512) in extra_body.
• At the L4 level we also use these cases in the nightly pipeline to monitor e2e latency under each feature combination.

An additional comment was added at the top of the file to clarify the use of the file.

[princepride]

We need compare some specific pixel value to ensure these feature can output expect result, so you can refer to bagel's e2e test, and I also think only use OmniDiffusion is enough to test these feature.

[NumberWan]

Thanks for the suggestion!
But according to the internal agreement that:

• If a diffusion feature is supported in online serving, prefer to test it via online serving unless it only support offline serving.
• At the L4 e2e level we currently only validate that images/videos are generated with the expected shape, and record e2e latency.

This Bagel L4 suite follows exactly the same pattern/infra as test_qwen_image_edit_expansion.py and #1682 to keep the behavior consistent across models.

[congw729]

#1832 This is the RFC @princepride

[princepride]

Thank you, I will check it later

[yenuo26]

Thank you, I will check it later

I think maybe you can unify the bagel test cases in the current test-ready.yml into test_bagel.py, and use the current code style

[NumberWan]

@princepride Thanks! This PR follows the current L4 diffusion e2e scope in RFC #1832 / template #1682 (online serving, shape checks). . If this is acceptable, could you please dismiss the “changes requested” so this PR can merge?

[yenuo26]

I think you should use pytest -s -v tests/e2e/online_serving/test_*_expansion.py -m "advanced_model and diffusion and H100" --collect-only to verify whether the test cases can be collected.
And，could you attach the local execution results?

[NumberWan]

I think you should use pytest -s -v tests/e2e/online_serving/test_*_expansion.py -m "advanced_model and diffusion and H100" --collect-only to verify whether the test cases can be collected. And，could you attach the local execution results?

Thank you, I just added the test result according to the command you provided. The test result in the "Additional Test Result" part.

[NumberWan]

@hsliuustc0106 @Gaohan123 All requested CI test have been passed(The corresponding result in the "Test Result" section.) Could you please help take a look for merge?