[Fix] Update to v0.1.5.post4 and refine HIP attention backend selection

[yichiche]

Motivation

This PR updates the aiter backend to the latest commit (v0.1.5.post4) to include the newest MLA kernel implementation supported on AMD MI355x GPUs and meanwhile update sglang interface support speculative decoding and fixes the logic for selecting the HIP attention backend. Previously, the condition for enabling aiter was more complex and could lead to unexpected fallbacks. This change makes the condition explicit: aiter will only be used when the number of KV heads is 16 or 128, otherwise, the fallback remains triton.

Modifications

• Dockerfile.rocm
  • Bumped AITER_COMMIT from v0.1.5.post3 → v0.1.5.post4.
• model_runner.py
  • Simplified conditional logic for HIP backend choice.

Accuracy Tests

• Verified against gsm8k and other standard evaluation datasets.
• Accuracy remains unchanged after applying the fix.

Benchmarking and Profiling

Model: deepseek-ai/DeepSeek-R1-0528
Server: MI355x * 8

Concurrency | Baseline | Speculative decoding
1 | 13459.11| 6260.24

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.

[gemini-code-assist]

Summary of Changes

Hello @yichiche, 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 focuses on enhancing the stability and performance for AMD GPU users by updating the aiter backend to its latest version (v0.1.5.post4) to enable speculative decoding. Concurrently, it refines the conditional logic for selecting the HIP attention backend, ensuring aiter is only engaged under specific, supported KV head configurations (16 or 128), thereby preventing unexpected fallbacks and improving overall correctness.

Highlights

• Aiter Backend Update: The aiter backend has been updated to v0.1.5.post4 to incorporate the latest features, including support for speculative decoding.
• HIP Attention Backend Selection Refinement: The logic for selecting the HIP attention backend has been refined. aiter will now be explicitly used only when the number of KV heads is 16 or 128, simplifying the condition and ensuring correct fallback to triton otherwise.

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 updates the aiter backend to v0.1.5.post4 and refines the logic for selecting the HIP attention backend, particularly to enable speculative decoding. The changes are straightforward and correct, including the version bump in the Dockerfile and the simplified conditional logic in model_runner.py. I have one minor suggestion to further improve the code's readability and maintainability.