From 71babdef510ef3d95ea829ec61e5f7c3f9ae7232 Mon Sep 17 00:00:00 2001 From: ZhenshengWu <49758440+ZhenshengWu@users.noreply.github.com> Date: Sun, 1 Feb 2026 15:41:21 +0800 Subject: [PATCH] Fix CUDA 12 dependency when importing Mooncake in official CUDA 13.x image (#17540) Co-authored-by: wuzhensheng01 --- docker/Dockerfile | 23 ++++++++++++++++++++++- 1 file changed, 22 insertions(+), 1 deletion(-) diff --git a/docker/Dockerfile b/docker/Dockerfile index 366efa327..f76be5fae 100644 --- a/docker/Dockerfile +++ b/docker/Dockerfile @@ -19,6 +19,11 @@ ARG PIP_DEFAULT_INDEX ARG UBUNTU_MIRROR ARG GITHUB_ARTIFACTORY=github.com ARG INSTALL_FLASHINFER_JIT_CACHE=0 +ARG FLASHINFER_VERSION=0.6.1 +ARG MOONCAKE_VERSION=0.3.8.post1 +#if need other arg please add in MOONCAKE_COMPILE_ARG +ARG MOONCAKE_COMPILE_ARG="-DUSE_HTTP=ON -DUSE_MNNVL=ON -DUSE_CUDA=ON -DWITH_EP=ON" + ARG FLASHINFER_VERSION=0.6.2 ENV DEBIAN_FRONTEND=noninteractive \ @@ -269,11 +274,27 @@ RUN --mount=type=cache,target=/root/.cache/pip \ fi && \ TORCH_CUDA_ARCH_LIST="${CHOSEN_TORCH_CUDA_ARCH_LIST}" MAX_JOBS=${BUILD_AND_DOWNLOAD_PARALLEL} pip install --no-build-isolation . +# Install Mooncake +RUN --mount=type=cache,target=/root/.cache/pip \ + CUDA_MAJOR="${CUDA_VERSION%%.*}" && \ + if [ "$CUDA_MAJOR" -ge 13 ]; then \ + echo "CUDA >= 13, installing mooncake-transfer-engine from source code"; \ + git clone --branch v${MOONCAKE_VERSION} --depth 1 https://github.com/kvcache-ai/Mooncake.git && \ + cd Mooncake && \ + bash dependencies.sh && \ + mkdir -p build && \ + cd build && \ + cmake .. ${MOONCAKE_COMPILE_ARG} && \ + make -j$(nproc) && \ + make install; \ + else \ + echo "CUDA < 13, installing mooncake-transfer-engine from pip"; \ + python3 -m pip install mooncake-transfer-engine==${MOONCAKE_VERSION}; \ + fi # Install essential Python packages RUN --mount=type=cache,target=/root/.cache/pip \ python3 -m pip install \ datamodel_code_generator \ - mooncake-transfer-engine==0.3.8.post1 \ pre-commit \ pytest \ black \