diff --git a/vllm/Dockerfile b/vllm/Dockerfile index dd9bdbf..3c34d8b 100644 --- a/vllm/Dockerfile +++ b/vllm/Dockerfile @@ -90,7 +90,11 @@ RUN mkdir -p /wheels && \ FROM build-base AS build-flashinfer ARG FLASHINFER_ENABLE_AOT=1 -ARG FLASHINFER_REF=v0.6.7 +# flashinfer version compatibility: +# - v0.6.7 works with vLLM v0.18.2rc0 (Build #43) +# - v0.6.6 works with vLLM v0.19.0 (for Gemma 4 support) +# ARG FLASHINFER_REF=v0.6.7 # For vLLM v0.18.2rc0 +ARG FLASHINFER_REF=v0.6.6 ARG FLASHINFER_BUILD_SUFFIX=cu132 ENV FLASHINFER_LOCAL_VERSION=${FLASHINFER_BUILD_SUFFIX:-} RUN git clone https://github.com/flashinfer-ai/flashinfer.git