Update gpt-oss mi300,mi325,mi355 for vllm 0.17.0

[Rohan138]

See related PR: SemiAnalysisAI/InferenceX#867, will be merged after vllm 0.17.0 is released

[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 updates the GPT-OSS documentation to align with the latest vLLM 0.17.0 release, providing revised configurations and serving commands for enhanced performance on AMD GPUs. The changes introduce new environment variables and command-line arguments, along with an example for a quantized model, ensuring users can leverage the most current optimizations.

Highlights

• vLLM Version Update: The documentation has been updated to reflect configurations validated with the vLLM 0.17.0 release, moving from the previous 0.14.1 version.
• New Environment Variable: Added export AMDGCN_USE_BUFFER_OPS=0 to the recommended environment variables for optimal performance.
• Updated vLLM Serve Commands: The vllm serve commands have been revised to include new arguments such as --attention-backend ROCM_AITER_UNIFIED_ATTN, -cc.pass_config.fuse_rope_kvcache=True, and -cc.use_inductor_graph_partition=True, while removing older flags like --compilation-config and --disable-log-request.
• Quantized Model Example: An example command now demonstrates serving the amd/gpt-oss-120b-w-mxfp4-a-fp8 model, which is a Quark-quantized version supporting fp8-quantized activations.

🧠 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

• OpenAI/GPT-OSS.md
  • Updated the referenced vLLM release version from 0.14.1 to 0.17.0.
  • Added the export AMDGCN_USE_BUFFER_OPS=0 environment variable.
  • Removed VLLM_ROCM_USE_AITER_UNIFIED_ATTENTION and VLLM_ROCM_USE_AITER_MHA environment variables.
  • Modified vllm serve commands to use new attention backend and compilation configuration flags.
  • Replaced openai/gpt-oss-120b with amd/gpt-oss-120b-w-mxfp4-a-fp8 in one example and added a description for the new quantized model.

Activity

• No human activity has been recorded on this pull request yet.

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 vLLM configurations for MI300, MI325, and MI355 to align with the vLLM 0.17.0 release. The changes include updating the validated release version, introducing new environment variables like AMDGCN_USE_BUFFER_OPS=0, and modifying vllm serve command arguments to incorporate new compilation flags and attention backend settings. Additionally, the default model for MI355x has been updated to a Quark-quantized version. One inconsistency was noted regarding the async-scheduling flag.

_{Note: Security Review has been skipped due to the limited scope of the PR.}

[gemini-code-assist]

The vllm serve command for MI355x no longer includes the --async-scheduling flag. However, the "Configs and Parameters" section (line 316) still recommends always adding this flag for best performance. Please clarify if --async-scheduling is no longer recommended or implicitly handled for MI355x with vLLM 0.17.0, or if it should be re-included in this command for consistency with the general recommendation.

[seungrokj]

@Rohan138
uv pip install vllm --extra-index-url https://wheels.vllm.ai/rocm/0.17.0/rocm700
is out