RuntimeError: wrong! device_gemm with the specified compilation parameters does not support this GEMM problem` that there used to be for some models using problem sizes not supported by aiter.

该报错通常发生在 AMD ROCm 平台(gfx950/MI355)运行 Quark MXFP4 量化模型时,vLLM 将 `w_mxfp4_a_mxfp4` 量化路径调度到 AITER 原生 MXFP4 内核后输出损坏。优先排查 vLLM 中 `QuarkOCP_MX` 和 `QuarkOCP_M



![[Question]: About uploaded file size limit](https://www.chat-gpts.plus/wp-content/uploads/2026/08/6903-55d30477-768x403.jpg)
![[Question]:Add chat's ollama model hint 102](https://www.chat-gpts.plus/wp-content/uploads/2026/08/4074-06c47b07-768x403.jpg)



