fix: adapt to upstream vLLM changes (2026-03-02)

[Meihan-chen]

Summary

Fixes CI failures in schedule_test_vllm_main caused by upstream vLLM changes.

Commit range: 15d76f74e2fdb12a95ea00f0ca283acf6219a2b7..6290470843c131681e3e1318ae71070a34f33225

Issues Fixed

• CudagraphDispatcher.dispatch() — disable_full param replaced with valid_modes/invalid_modes API (PR #34102)
• compile_or_warm_up_model() — now must return float compilation_time (PR #35503)
• AscendMMEncoderAttention.forward_oot() — new sequence_lengths param added (PR #35564)
• rms_norm for SP — auto-forcing of +rms_norm removed upstream, re-added in platform.py (PR #35410)

Issues Skipped (Environment Flakes)

• None detected in this run

Test plan

• Verify all 4 previously failing jobs pass with the new vLLM commit
• Confirm v0.16.0 compatibility is not broken

🤖 Generated with Claude Code

• vLLM version: v0.16.0
• vLLM main: vllm-project/vllm@15d76f7

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Summary of Changes

Hello, 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 and resolves CI failures in the schedule_test_vllm_main pipeline by adapting the vllm_ascend codebase to recent breaking API changes introduced in the upstream vLLM library. The changes ensure continued compatibility and correct execution of vLLM on Ascend devices, particularly concerning CUDAGraph dispatching, model compilation, attention mechanisms, and RMS normalization for sparse attention.

Highlights

• CudagraphDispatcher.dispatch() API Update: Adapted to the change where disable_full parameter was replaced with valid_modes and invalid_modes in the upstream vLLM.
• compile_or_warm_up_model() Return Type: Updated the method signature and return value to align with the upstream change requiring it to return a float for compilation time.
• AscendMMEncoderAttention.forward_oot() Parameter: Incorporated the new sequence_lengths parameter into the forward_oot method.
• RMS Norm for SP Re-enforcement: Re-added the auto-forcing of +rms_norm for SP on Ascend, which was removed upstream, to ensure correct functionality.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

• vllm_ascend/ops/mm_encoder_attention.py
  • Added sequence_lengths as an optional parameter to the forward_oot function signature.
• vllm_ascend/platform.py
  • Introduced logic to append +rms_norm to compilation_config.custom_ops if SP is enabled and +rms_norm is not already present, compensating for an upstream change.
• vllm_ascend/worker/model_runner_v1.py
  • Modified the dispatch_cudagraph lambda function to accept valid_modes and invalid_modes instead of disable_full, and updated its internal call to self.cudagraph_dispatcher.dispatch.
  • Updated a subsequent call to dispatch_cudagraph to pass valid_modes instead of disable_full.
• vllm_ascend/worker/worker.py
  • Updated the type hint for the compile_or_warm_up_model method to indicate it returns a float.
  • Added a return statement to compile_or_warm_up_model to return the compilation_time from the configuration.

Ignored Files

• Ignored by pattern: .github/workflows/** (4)
  • .github/workflows/bot_pr_create.yaml
  • .github/workflows/pr_test_full.yaml
  • .github/workflows/pr_test_light.yaml
  • .github/workflows/schedule_codecov_refresh.yaml

Activity

• The pull request was created to fix CI failures in schedule_test_vllm_main caused by upstream vLLM changes.
• The author plans to verify that all four previously failing jobs pass with the new vLLM commit.
• The author intends to confirm that v0.16.0 compatibility is not broken.
• The pull request was generated with Claude Code.

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 by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

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 pull request 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 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. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

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 adapts the codebase to recent upstream changes in vLLM. The modifications address API and behavior changes related to the CUDAGraph dispatcher, model compilation/warm-up, multi-modal attention, and sequence parallelism. The changes are necessary for compatibility and appear to be implemented correctly. I have no further feedback.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.