[ROCm] add Qwen3 235B recipe

[gbyu-amd]

This pr adds a recipe for Qwen3-235B-A22B running on ROCm platforms.

The recipe is subject to changes as some optimizations are still on the way:

• q_norm + k_norm + rope fusion fusion: [ROCm] [Bugfix] Fix fused_qknorm_rope_kernel rocm compatibility vllm#28500 enables q_norm + k_norm + rope fusion on ROCm platforms, which was initially implemented for cuda in [Kernel][Perf] fuse QK Norm and RoPE into one cuda kernel for Qwen Model vllm#27165. PR already merged to vLLM main!
• rms_norm + fp8_block_quant fusion: [Rocm][torch.compile] Adding layernorm + fp8 block quant and silu + fp8 block quant for Aiter vllm#25693 added new fusion passes to enable rms_norm + fp8_block_quant and silu + fp8_block_quant, which depends on the triton fused kernel in https://github.com/ROCm/aiter/tree/dev/perf_fused_rms_fp8_group_quant. Need to check if this triton kernel merged into AITER main. triton kernel already merged to AITER main! vLLM pr still under review
• Sequence parallel: code ready in enable sequence parallel for rocm ROCm/vllm#790. Need pr to upstream as well. But seems poor performance due to the pynccl comm op for now.
• All-reduce + rms_norm fusion: WIP in [CommFusion] Add allreduce+rmsnorm fusion kernel ROCm/vllm#803.
• FP8 block GEMM is not efficient enough, i.e., up to 1.2p ~ 1.4p flops even after tuning. Refer to [Config] add tuned moe and gemm config for qwen3 235b ROCm/aiter#1378 for the tuning pr. Optimization WIP, just need to upgrade AITER when it's done.

The current recipe only provides TP8+EP8 deployment as an example. We will try other parallel strategies for best performance across different scenarios.

[gemini-code-assist]

Summary of Changes

Hello @gbyu-amd, 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 establishes the foundational recipe for running the Qwen3-235B-A22B model on ROCm-enabled hardware using vLLM. It provides a complete guide from environment setup to model serving and initial performance/accuracy testing, serving as a baseline while acknowledging ongoing optimization efforts for improved efficiency and broader parallelism support.

Highlights

• New Model Recipe: This pull request introduces a comprehensive usage guide for deploying the Qwen3-235B-A22B large language model on ROCm platforms, specifically targeting MI355 GPUs.
• Environment Setup: Detailed instructions are provided for setting up the necessary environment, including Docker configuration and the installation of optimized vLLM and AITER versions for ROCm.
• Model Serving Configurations: The guide includes specific commands for serving the Qwen3-235B-A22B model in both BF16 and FP8 precisions, utilizing a TP8 + EP8 parallelism strategy across 8xMI355 GPUs.
• Performance and Accuracy Testing: Benchmarking scripts and accuracy verification procedures using lm_eval are outlined to assess the model's performance and correctness on the ROCm platform.

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 adds a new recipe for running Qwen3-235B-A22B on ROCm platforms. The documentation is a good starting point, but it contains several typos and critically incorrect shell commands that will fail. I have provided specific suggestions to correct these issues to ensure the recipe is usable.