[Fix] Fix gemma CI test failing on main

[tdoublep]

Purpose

The optional CI job "Language Models Test (Extended Generation)" is currently failing on main due to one of the Gemma tests that is using out-dated API calls. This PR adapts that failing test to work with latest vLLM API.

This CI job is currently the only one we have that tests hybrid SSM/attention models and so it is pretty useful to have it up and running.

Test Plan

n/a

Test Result

Test is now passing (will trigger it through CI)

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Summary of Changes

Hello @tdoublep, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request addresses a failing CI job related to Gemma model tests. The primary goal is to update an outdated test to be compatible with the current vLLM API, ensuring the continued validation of hybrid SSM/attention models. The changes involve adapting API calls within the test and configuring the environment for proper execution.

Highlights

• Test Fix: Updated the test_dummy_loader for Gemma models to align with the latest vLLM API, specifically by adjusting collective_rpc calls and accessing model configurations.
• Environment Configuration: Introduced a monkeypatch to temporarily set the VLLM_ALLOW_INSECURE_SERIALIZATION environment variable, which is necessary for the updated test to run correctly.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in issue comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request correctly updates the Gemma test to align with recent API changes in vLLM. The use of monkeypatch to set VLLM_ALLOW_INSECURE_SERIALIZATION is appropriate for this test context, enabling the use of a lambda function with collective_rpc. The modifications to the collective_rpc call and attribute access paths for the model engine configuration appear correct and reflect the API refactoring mentioned in the description. The change is straightforward and effectively addresses the failing CI job.

[houseroad]

wondering why this cause failure?

[DarkLight1337]

I think it's because you're passing a function into collective_rpc which can only be serialized via pickle (insecure)

[tdoublep]

The language-models-test-extended-generation test still fails but different failures now:

[2025-06-26T17:43:39Z] FAILED models/language/generation/test_common.py::test_models[False-5-32-TitanML/tiny-mixtral] - AssertionError: Test0:
--
  | [2025-06-26T17:43:39Z] Matched tokens:	[]
  | [2025-06-26T17:43:39Z] hf:	'ottdone бли施 Besodigd De Account me merc De me ге酒ott Bes酒ottott酒наль酒ott Potavirus酒酒酒酒酒 Davisavirus'	{1562: -7.970005989074707, 29191: -8.063755989074707, 13654: -8.188755989074707, 30658: -8.235630989074707, 20671: -8.360630989074707}
  | [2025-06-26T17:43:39Z] vllm:	' orth surg блиDIMemulumodigdría denom Васи Васи scattered whispsi黄tiestiestiesacon crienthenthenthenthSingriegpsi"},iface chefarn寸'	{17396: Logprob(logprob=-8.14117431640625, rank=1, decoded_token='▁orth'), 17358: Logprob(logprob=-8.21929931640625, rank=2, decoded_token='▁conflic'), 15630: Logprob(logprob=-8.25054931640625, rank=3, decoded_token='odigd'), 8393: Logprob(logprob=-8.25054931640625, rank=4, decoded_token='▁Bes'), 18018: Logprob(logprob=-8.31304931640625, rank=5, decoded_token='▁preserve')}
  | [2025-06-26T17:43:39Z] 

FAILED models/language/generation/test_granite.py::test_models[5-64-bfloat16-ibm/PowerMoE-3b] - AssertionError: Test0:
  | [2025-06-26T17:43:39Z] Matched tokens:	[203]
  | [2025-06-26T17:43:39Z] hf:	'\n## Installation\n\n### Requirements\n\n- Python 3.7+\n- PyTorch 1.8+\n- Transformers 4.10+\n- Hugging Face Transformers 4.10+\n- PyTorch-Lightning 1.4+\n- PyTorch-Lightning Data'	{433: -0.1428334265947342, 1318: -2.1428334712982178, 1011: -4.892833232879639, 1482: -5.142833232879639, 383: -6.642833232879639}
  | [2025-06-26T17:43:39Z] vllm:	'\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nThe 1\n\n\n\n\n\n\n\nThe 1\n\n\n\n\nThe 1\n\n\n\n\nThe 1\n\n\n\n\nThe 1\n\n\n\nThe 1\n'	{203: Logprob(logprob=-1.6709017753601074, rank=1, decoded_token='Ċ'), 31: Logprob(logprob=-2.5459017753601074, rank=2, decoded_token='-'), 1318: Logprob(logprob=-3.6396517753601074, rank=3, decoded_token='The'), 51: Logprob(logprob=-4.077151775360107, rank=4, decoded_token='A'), 15698: Logprob(logprob=-4.577151775360107, rank=5, decoded_token='âĢ¢')}

This extended generation job is seriously flaky (which I guess is to be expected since it is not running by default).

The gemma tests are now passing, so I think this PR can be merged.

[tdoublep]

@mgoin @DarkLight1337 @houseroad Could I get a force-merge here?

There are two unrelated tests that are failing in the extended generation job due to another issue (see comment here). Looking into that separately, but might as well get these fixes in.