infrastructure-model-deployment-when-using-vllm-with-gpt-oss-models-and-specifying-0c581235

THEN Ensure you are using a vLLM build that includes mxfp4 support. For GPT‑OSS, use the official prebuilt Docker image or wheel from the GPT‑OSS recipe (see the vLLM user guide for GPT‑OSS). Mainstream vLLM releases (e.g., 0.10) do not have this quantization method.

# Instead of a standard vllm install, use the GPT‑OSS Docker image:
# docker pull gptoss/gpt-oss:latest
# docker run ... vllm.entrypoints.openai.api_server ...