[bugfix] Fixed the error issue when overlaying MTP and full decode on DSV3.1 C8.

[pichangping]

…DSV3.1 C8.

What this PR does / why we need it?

DeepSeek v3.1 C8 had a hanging issue when overlaying MTP and full graph modes; this pull request resolves that issue.

Does this PR introduce any user-facing change?

no

How was this patch tested?

• vLLM version: v0.18.0
• vLLM main: vllm-project/vllm@ed359c4

[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 a critical hanging issue encountered in DeepSeek v3.1 C8 when operating with overlaid MTP (Multi-Token Prediction) and full graph modes. The fix involves precise adjustments to how quantization scaling parameters are handled within the attention mechanism's decode path, specifically ensuring correct tensor dimensions and proper weak reference management for these parameters.

Highlights

• Quantization Scale Reshaping: The dequant_scale_q_nope tensor is now explicitly reshaped to (num_tokens, self.num_heads) when fa_quant_layer is active during the decode phase, ensuring correct dimensions for subsequent operations.
• Weak Reference Handling for Attention Parameters: Both dequant_scale_q_nope and self.fak_descale_float are now wrapped with weak_ref_tensors when being added to the attn_params tuple, which is crucial for managing tensor lifetimes and preventing memory leaks in certain execution environments.

🧠 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.

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.

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. ↩

[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]

Code Review

This pull request addresses a hanging issue and a potential memory leak within the MLA attention implementation for quantized models. The changes include correcting the shape of the dequant_scale_q_nope tensor during speculative decoding to prevent operator hangs, and wrapping tensors with weak_ref_tensors during graph capture to avoid memory leaks. The fixes appear correct and well-targeted to resolve the described issues.