[Bug]: ROCm MiniMax M3 MXFP8 Disagg not working

[functionstackx]

Your current environment

vllm/vllm-openai-rocm:minimax-m3

🐛 Describe the bug

hi @andyluo7

+viz @chunfangamd @powderluv @hongxiayang

can u take a look? is there any validated DI recipe (whether MoRI or RIXL or Mooncake, the KVCache transfer engine doesn't matter to us for week 0 enablement) for ROCm M3 yet?

evals fail on TP8 Prefill + TP8 Decode SemiAnalysisAI/InferenceX#1762

https://github.com/SemiAnalysisAI/InferenceX/actions/runs/27526046669/job/81353671922?pr=1762

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[github-actions]

CC @tjtanaa @vllmellm for ROCm-related issue

[github-actions]

Hi @functionstackx, thanks for reporting this ROCm issue!

To help us investigate, please make sure the following information is included:

• A minimal reproducer (code snippet or script that triggers the issue)
• The full error message or traceback
• How you installed vLLM (Docker image name, pip install, or build from source steps)
• The command you used to launch vLLM (e.g., vllm serve ... or the Python script)
• Your GPU model and GFX architecture (e.g., MI300X / gfx942) — run rocminfo | grep gfx

Please provide any unchecked items above. This will help us reproduce and resolve the issue faster. Thank you!