[PD] Optimize MHA models pp util calculation logic

[ShangmingCai]

Motivation

• Refactor comments.
• Add support when dst_num_total_layers // num_kv_layers != 1

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Summary of Changes

Hello @ShangmingCai, 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 optimizes the utility calculation logic for Multi-Head Attention (MHA) models, specifically addressing how KV pointers are handled in disaggregated setups. The primary goal is to improve the robustness and correctness of KV cache management when there's a mismatch in the number of layers between the destination and KV layers, ensuring that the system can correctly map KV pointers in more diverse model configurations.

Highlights

• Comment Refactoring: Refactored an existing comment in conn.py for improved clarity, removing a potentially misleading statement about preventing empty Value Cache.
• MHA KV Pointer Calculation Logic Enhancement: Enhanced the get_mha_kv_ptrs_with_pp function to correctly handle scenarios where the destination total layers are not a direct multiple of the number of KV layers (i.e., dst_num_total_layers // num_kv_layers != 1). This ensures accurate KV pointer calculation in more complex model disaggregation setups, particularly when a decode model uses a draft model KV while prefill is deployed without speculative decoding.

🧠 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.

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. ↩

[ShangmingCai]

/rerun-stage unit-test-backend-8-gpu-h20

[github-actions]

✅ Triggered unit-test-backend-8-gpu-h20 to run independently (skipping dependencies).

[github-actions]

🔗 View workflow run

[ShangmingCai]

/tag-and-rerun-ci

[gemini-code-assist]

Code Review

This pull request optimizes the MHA model's pipeline parallelism utility calculation, specifically for handling speculative decoding scenarios where the number of layers in the destination (decode) differs from the source (prefill). The change correctly calculates the V-cache pointer offset by introducing a multiplier_ratio. The logic appears sound and addresses the intended case. I have one minor suggestion to improve the readability of the offset calculation.

[ShangmingCai]

/rerun-stage unit-test-backend-8-gpu-h20

[github-actions]

✅ Triggered unit-test-backend-8-gpu-h20 to run independently (skipping dependencies).

[github-actions]

🔗 View workflow run

[ShangmingCai]

/rerun-failed-ci

[ShangmingCai]

CIs have passed. The only failed one: gb200 test is not related to the change of this PR.