/bin/sh -c #(nop) ARG RELEASE

/bin/sh -c #(nop) ARG LAUNCHPAD_BUILD_ARCH

/bin/sh -c #(nop) LABEL org.opencontainers.image.ref.name=ubuntu

/bin/sh -c #(nop) LABEL org.opencontainers.image.version=24.04

/bin/sh -c #(nop) ADD file:d77dea5c49828eb0de89439d2b631bc8ea27cb9ef774412b56a060ba1673487b in /

/bin/sh -c #(nop) CMD ["/bin/bash"]

ENV NVARCH=sbsa

ENV NVIDIA_REQUIRE_CUDA=cuda>=13.0 brand=unknown,driver>=535,driver<536 brand=grid,driver>=535,driver<536 brand=tesla,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=vapps,driver>=535,driver<536 brand=vpc,driver>=535,driver<536 brand=vcs,driver>=535,driver<536 brand=vws,driver>=535,driver<536 brand=cloudgaming,driver>=535,driver<536 brand=unknown,driver>=550,driver<551 brand=grid,driver>=550,driver<551 brand=tesla,driver>=550,driver<551 brand=nvidia,driver>=550,driver<551 brand=quadro,driver>=550,driver<551 brand=quadrortx,driver>=550,driver<551 brand=nvidiartx,driver>=550,driver<551 brand=vapps,driver>=550,driver<551 brand=vpc,driver>=550,driver<551 brand=vcs,driver>=550,driver<551 brand=vws,driver>=550,driver<551 brand=cloudgaming,driver>=550,driver<551 brand=unknown,driver>=565,driver<566 brand=grid,driver>=565,driver<566 brand=tesla,driver>=565,driver<566 brand=nvidia,driver>=565,driver<566 brand=quadro,driver>=565,driver<566 brand=quadrortx,driver>=565,driver<566 brand=nvidiartx,driver>=565,driver<566 brand=vapps,driver>=565,driver<566 brand=vpc,driver>=565,driver<566 brand=vcs,driver>=565,driver<566 brand=vws,driver>=565,driver<566 brand=cloudgaming,driver>=565,driver<566 brand=unknown,driver>=570,driver<571 brand=grid,driver>=570,driver<571 brand=tesla,driver>=570,driver<571 brand=nvidia,driver>=570,driver<571 brand=quadro,driver>=570,driver<571 brand=quadrortx,driver>=570,driver<571 brand=nvidiartx,driver>=570,driver<571 brand=vapps,driver>=570,driver<571 brand=vpc,driver>=570,driver<571 brand=vcs,driver>=570,driver<571 brand=vws,driver>=570,driver<571 brand=cloudgaming,driver>=570,driver<571 brand=unknown,driver>=575,driver<576 brand=grid,driver>=575,driver<576 brand=tesla,driver>=575,driver<576 brand=nvidia,driver>=575,driver<576 brand=quadro,driver>=575,driver<576 brand=quadrortx,driver>=575,driver<576 brand=nvidiartx,driver>=575,driver<576 brand=vapps,driver>=575,driver<576 brand=vpc,driver>=575,driver<576 brand=vcs,driver>=575,driver<576 brand=vws,driver>=575,driver<576 brand=cloudgaming,driver>=575,driver<576

ENV NV_CUDA_CUDART_VERSION=13.0.96-1

ARG TARGETARCH

LABEL maintainer=NVIDIA CORPORATION <cudatools@nvidia.com>

RUN |1 TARGETARCH=arm64 /bin/sh -c apt-get update && apt-get install -y --no-install-recommends gnupg2 curl ca-certificates && curl -fsSL https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/${NVARCH}/3bf863cc.pub | apt-key add - && echo "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/${NVARCH} /" > /etc/apt/sources.list.d/cuda.list && apt-get purge --autoremove -y curl && rm -rf /var/lib/apt/lists/* # buildkit

ENV CUDA_VERSION=13.0.2

RUN |1 TARGETARCH=arm64 /bin/sh -c apt-get update && apt-get install -y --no-install-recommends cuda-cudart-13-0=${NV_CUDA_CUDART_VERSION} cuda-compat-13-0 && rm -rf /var/lib/apt/lists/* # buildkit

RUN |1 TARGETARCH=arm64 /bin/sh -c echo "/usr/local/cuda/lib64" >> /etc/ld.so.conf.d/nvidia.conf # buildkit

ENV PATH=/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64

COPY NGC-DL-CONTAINER-LICENSE / # buildkit

ENV NVIDIA_VISIBLE_DEVICES=all

ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility

ARG CUDA_VERSION

ARG PYTHON_VERSION

ARG NCCL_VERSION

ARG DEADSNAKES_MIRROR_URL

ARG DEADSNAKES_GPGKEY_URL

ARG GET_PIP_URL

ENV DEBIAN_FRONTEND=noninteractive

WORKDIR /vllm-workspace

RUN |6 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c PYTHON_VERSION_STR=$(echo ${PYTHON_VERSION} | sed 's/\.//g') && echo "export PYTHON_VERSION_STR=${PYTHON_VERSION_STR}" >> /etc/environment # buildkit

RUN |6 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c apt-get update -y && apt-get install -y --no-install-recommends software-properties-common curl sudo ffmpeg libsm6 libxext6 libgl1 && if [ ! -z ${DEADSNAKES_MIRROR_URL} ] ; then if [ ! -z "${DEADSNAKES_GPGKEY_URL}" ] ; then mkdir -p -m 0755 /etc/apt/keyrings ; curl -L ${DEADSNAKES_GPGKEY_URL} | gpg --dearmor > /etc/apt/keyrings/deadsnakes.gpg ; sudo chmod 644 /etc/apt/keyrings/deadsnakes.gpg ; echo "deb [signed-by=/etc/apt/keyrings/deadsnakes.gpg] ${DEADSNAKES_MIRROR_URL} $(lsb_release -cs) main" > /etc/apt/sources.list.d/deadsnakes.list ; fi ; else for i in 1 2 3; do add-apt-repository -y ppa:deadsnakes/ppa && break || { echo "Attempt $i failed, retrying in 5s..."; sleep 5; }; done ; fi && apt-get update -y && apt-get install -y --no-install-recommends python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv libibverbs-dev && rm -rf /var/lib/apt/lists/* && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config && rm -f /usr/lib/python${PYTHON_VERSION}/EXTERNALLY-MANAGED && curl -sS ${GET_PIP_URL} | python${PYTHON_VERSION} && python3 --version && python3 -m pip --version # buildkit

RUN |6 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c CUDA_VERSION_DASH=$(echo $CUDA_VERSION | cut -d. -f1,2 | tr '.' '-') && apt-get update -y && apt-get install -y --no-install-recommends --allow-change-held-packages cuda-nvcc-${CUDA_VERSION_DASH} cuda-cudart-${CUDA_VERSION_DASH} cuda-nvrtc-${CUDA_VERSION_DASH} cuda-cuobjdump-${CUDA_VERSION_DASH} libcurand-dev-${CUDA_VERSION_DASH} libcublas-dev-${CUDA_VERSION_DASH} libnuma-dev numactl && rm -rf /var/lib/apt/lists/* # buildkit

RUN |6 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c python3 -m pip install uv # buildkit

ENV UV_HTTP_TIMEOUT=500

ENV UV_INDEX_STRATEGY=unsafe-best-match

ENV UV_LINK_MODE=copy

ENV UV_PYTHON_INSTALL_DIR=/opt/uv/python

ENV UV_CACHE_DIR=/opt/uv/cache

RUN |6 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c mkdir -p "${UV_PYTHON_INSTALL_DIR}" "${UV_CACHE_DIR}" && chgrp -R 0 /opt/uv && chmod -R g+rwX,a+rX /opt/uv # buildkit

RUN |6 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c CUDA_MAJOR=$(echo $CUDA_VERSION | cut -d. -f1) && if [ -n "$NCCL_VERSION" ]; then echo "nvidia-nccl-cu${CUDA_MAJOR}==${NCCL_VERSION}" > /etc/uv-overrides.txt; else touch /etc/uv-overrides.txt; fi # buildkit

ENV UV_OVERRIDE=/etc/uv-overrides.txt

RUN |6 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c useradd --uid 2000 --gid 0 --create-home --home-dir /home/vllm --shell /bin/bash vllm && mkdir -p /home/vllm/.cache /home/vllm/.config && chown -R 2000:0 /home/vllm && chmod -R g+rwX /home/vllm && chgrp 0 /etc/passwd /etc/group && chmod g=u /etc/passwd /etc/group # buildkit

COPY docker/entrypoints/vllm-nonroot-entrypoint.sh /usr/local/bin/vllm-nonroot-entrypoint.sh # buildkit

RUN |6 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c chmod 0755 /usr/local/bin/vllm-nonroot-entrypoint.sh # buildkit

ENV VLLM_ENABLE_CUDA_COMPATIBILITY=0

ARG PYTORCH_CUDA_INDEX_BASE_URL

COPY requirements/common.txt /tmp/common.txt # buildkit

COPY requirements/cuda.txt /tmp/requirements-cuda.txt # buildkit

RUN |7 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl /bin/sh -c if [ "$(echo $CUDA_VERSION | cut -d. -f1)" = "12" ]; then sed -i 's/^nvidia-cutlass-dsl\[cu13\]/nvidia-cutlass-dsl/' /tmp/requirements-cuda.txt; sed -i 's/^humming-kernels\[cu13\]/humming-kernels[cu12]/' /tmp/requirements-cuda.txt; fi && uv pip install --system -r /tmp/requirements-cuda.txt --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') && rm /tmp/requirements-cuda.txt /tmp/common.txt # buildkit

ARG FLASHINFER_VERSION=0.6.16.post3

RUN |8 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl FLASHINFER_VERSION=0.6.16.post3 /bin/sh -c uv pip install --system flashinfer-jit-cache==${FLASHINFER_VERSION} --index-url https://flashinfer.ai/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') # buildkit

ARG GDRCOPY_CUDA_VERSION=12.8

ARG GDRCOPY_OS_VERSION=Ubuntu22_04

ARG TARGETPLATFORM

COPY tools/install_gdrcopy.sh /tmp/install_gdrcopy.sh # buildkit

RUN |11 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl FLASHINFER_VERSION=0.6.16.post3 GDRCOPY_CUDA_VERSION=12.8 GDRCOPY_OS_VERSION=Ubuntu24_04 TARGETPLATFORM=linux/arm64 /bin/sh -c set -eux; case "${TARGETPLATFORM}" in linux/arm64) UUARCH="aarch64" ;; linux/amd64) UUARCH="x64" ;; *) echo "Unsupported TARGETPLATFORM: ${TARGETPLATFORM}" >&2; exit 1 ;; esac; /tmp/install_gdrcopy.sh "${GDRCOPY_OS_VERSION}" "${GDRCOPY_CUDA_VERSION}" "${UUARCH}" && rm /tmp/install_gdrcopy.sh # buildkit

ARG BITSANDBYTES_VERSION_X86=0.46.1

ARG BITSANDBYTES_VERSION_ARM64=0.42.0

ARG TIMM_VERSION=>=1.0.17

ARG RUNAI_MODEL_STREAMER_VERSION=>=0.15.7

RUN |15 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl FLASHINFER_VERSION=0.6.16.post3 GDRCOPY_CUDA_VERSION=12.8 GDRCOPY_OS_VERSION=Ubuntu24_04 TARGETPLATFORM=linux/arm64 BITSANDBYTES_VERSION_X86=0.46.1 BITSANDBYTES_VERSION_ARM64=0.42.0 TIMM_VERSION=>=1.0.17 RUNAI_MODEL_STREAMER_VERSION=>=0.15.7 /bin/sh -c if [ "$TARGETPLATFORM" = "linux/arm64" ]; then BITSANDBYTES_VERSION="${BITSANDBYTES_VERSION_ARM64}"; else BITSANDBYTES_VERSION="${BITSANDBYTES_VERSION_X86}"; fi; uv pip install --system accelerate 'modelscope<1.38' "bitsandbytes>=${BITSANDBYTES_VERSION}" "timm${TIMM_VERSION}" "runai-model-streamer[s3,gcs,azure]${RUNAI_MODEL_STREAMER_VERSION}" # buildkit

ARG PIP_INDEX_URL UV_INDEX_URL

ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL

ARG PYTORCH_CUDA_INDEX_BASE_URL

ARG PIP_KEYRING_PROVIDER UV_KEYRING_PROVIDER

ARG PYTORCH_NIGHTLY

COPY /workspace/torch_lib_versions.txt torch_lib_versions.txt # buildkit

COPY /workspace/dist/wheel.sha256 /tmp/vllm-wheel.sha256 # buildkit

RUN |23 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl FLASHINFER_VERSION=0.6.16.post3 GDRCOPY_CUDA_VERSION=12.8 GDRCOPY_OS_VERSION=Ubuntu24_04 TARGETPLATFORM=linux/arm64 BITSANDBYTES_VERSION_X86=0.46.1 BITSANDBYTES_VERSION_ARM64=0.42.0 TIMM_VERSION=>=1.0.17 RUNAI_MODEL_STREAMER_VERSION=>=0.15.7 PIP_INDEX_URL= UV_INDEX_URL= PIP_EXTRA_INDEX_URL= UV_EXTRA_INDEX_URL= PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl PIP_KEYRING_PROVIDER=disabled UV_KEYRING_PROVIDER=disabled PYTORCH_NIGHTLY= /bin/sh -c if [ "${PYTORCH_NIGHTLY}" = "1" ]; then echo "Installing torch nightly..." && uv pip install --system $(cat torch_lib_versions.txt | xargs) --pre --index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/nightly/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') && echo "Installing vLLM..." && uv pip install --system dist/*.whl --verbose --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/nightly/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); else echo "Installing vLLM..." && uv pip install --system dist/*.whl --verbose --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); fi # buildkit

RUN |23 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl FLASHINFER_VERSION=0.6.16.post3 GDRCOPY_CUDA_VERSION=12.8 GDRCOPY_OS_VERSION=Ubuntu24_04 TARGETPLATFORM=linux/arm64 BITSANDBYTES_VERSION_X86=0.46.1 BITSANDBYTES_VERSION_ARM64=0.42.0 TIMM_VERSION=>=1.0.17 RUNAI_MODEL_STREAMER_VERSION=>=0.15.7 PIP_INDEX_URL= UV_INDEX_URL= PIP_EXTRA_INDEX_URL= UV_EXTRA_INDEX_URL= PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl PIP_KEYRING_PROVIDER=disabled UV_KEYRING_PROVIDER=disabled PYTORCH_NIGHTLY= /bin/sh -c . /etc/environment && uv pip list # buildkit

ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64

COPY /tmp/ep_kernels_workspace/dist/wheels.sha256 /tmp/ep-kernels-wheels.sha256 # buildkit

RUN |23 CUDA_VERSION=13.0.2 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl FLASHINFER_VERSION=0.6.16.post3 GDRCOPY_CUDA_VERSION=12.8 GDRCOPY_OS_VERSION=Ubuntu24_04 TARGETPLATFORM=linux/arm64 BITSANDBYTES_VERSION_X86=0.46.1 BITSANDBYTES_VERSION_ARM64=0.42.0 TIMM_VERSION=>=1.0.17 RUNAI_MODEL_STREAMER_VERSION=>=0.15.7 PIP_INDEX_URL= UV_INDEX_URL= PIP_EXTRA_INDEX_URL= UV_EXTRA_INDEX_URL= PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl PIP_KEYRING_PROVIDER=disabled UV_KEYRING_PROVIDER=disabled PYTORCH_NIGHTLY= /bin/sh -c uv pip install --system ep_kernels/dist/*.whl --verbose --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') # buildkit

ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64

COPY examples examples # buildkit

COPY benchmarks benchmarks # buildkit

COPY ./vllm/collect_env.py . # buildkit

ARG TARGETPLATFORM

ARG INSTALL_KV_CONNECTORS=false

ARG CUDA_VERSION

ARG VLLM_BUILD_COMMIT

ARG VLLM_BUILD_PIPELINE

ARG VLLM_BUILD_URL

ARG VLLM_IMAGE_TAG

ARG PIP_INDEX_URL UV_INDEX_URL

ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL

ENV UV_HTTP_TIMEOUT=500

ARG torch_cuda_arch_list=7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0

ENV TORCH_CUDA_ARCH_LIST=8.0 8.7 8.9 9.0 10.0 11.0 12.0

RUN |12 TARGETPLATFORM=linux/arm64 INSTALL_KV_CONNECTORS=true CUDA_VERSION=13.0.2 VLLM_BUILD_COMMIT=6e448d0ea9bf3d88d898b65449ca6dc2aec170ac VLLM_BUILD_PIPELINE=019d130e-464e-4ff7-b84b-492992c0c06b VLLM_BUILD_URL=https://buildkite.com/vllm/release-v2/builds/5006 VLLM_IMAGE_TAG=vllm/vllm-openai:v0.27.1-ubuntu2404 PIP_INDEX_URL= UV_INDEX_URL= PIP_EXTRA_INDEX_URL= UV_EXTRA_INDEX_URL= torch_cuda_arch_list=8.0 8.7 8.9 9.0 10.0 11.0 12.0 /bin/sh -c CUDA_MAJOR="${CUDA_VERSION%%.*}"; CUDA_VERSION_DASH=$(echo $CUDA_VERSION | cut -d. -f1,2 | tr '.' '-'); CUDA_HOME=/usr/local/cuda; BUILD_PKGS="libcusparse-dev-${CUDA_VERSION_DASH} libcublas-dev-${CUDA_VERSION_DASH} libcusolver-dev-${CUDA_VERSION_DASH}"; if [ "$INSTALL_KV_CONNECTORS" = "true" ]; then uv pip install --system -r /tmp/kv_connectors.txt --no-build || ( apt-get update -y && apt-get install -y --no-install-recommends --allow-change-held-packages ${BUILD_PKGS} && uv pip install --system -r /tmp/kv_connectors.txt --no-build-isolation && apt-get purge -y ${BUILD_PKGS} && rm -rf /var/lib/apt/lists/* ); uv pip install --system --force-reinstall --no-deps nixl-cu${CUDA_MAJOR}; fi # buildkit

ARG MOONCAKE_WHEEL_AARCH64

ARG MOONCAKE_WHEEL_X86_64

RUN |14 TARGETPLATFORM=linux/arm64 INSTALL_KV_CONNECTORS=true CUDA_VERSION=13.0.2 VLLM_BUILD_COMMIT=6e448d0ea9bf3d88d898b65449ca6dc2aec170ac VLLM_BUILD_PIPELINE=019d130e-464e-4ff7-b84b-492992c0c06b VLLM_BUILD_URL=https://buildkite.com/vllm/release-v2/builds/5006 VLLM_IMAGE_TAG=vllm/vllm-openai:v0.27.1-ubuntu2404 PIP_INDEX_URL= UV_INDEX_URL= PIP_EXTRA_INDEX_URL= UV_EXTRA_INDEX_URL= torch_cuda_arch_list=8.0 8.7 8.9 9.0 10.0 11.0 12.0 MOONCAKE_WHEEL_AARCH64=https://vllm-wheels.s3.amazonaws.com/mooncake/mooncake_transfer_engine-0.3.10.post2-0da9dfea3-cp312-cp312-manylinux_2_39_aarch64.whl MOONCAKE_WHEEL_X86_64=https://vllm-wheels.s3.amazonaws.com/mooncake/mooncake_transfer_engine-0.3.10.post2-0da9dfea3-cp312-cp312-manylinux_2_35_x86_64.whl /bin/sh -c if [ "$INSTALL_KV_CONNECTORS" = "true" ]; then if [ "$TARGETPLATFORM" = "linux/arm64" ]; then WHEEL="${MOONCAKE_WHEEL_AARCH64}"; else WHEEL="${MOONCAKE_WHEEL_X86_64}"; fi && if [ -n "${WHEEL}" ]; then uv pip install --system "${WHEEL}" && CUDA_MAJOR="${CUDA_VERSION%%.*}" && if [ ! -f /usr/local/cuda/lib64/libcudart.so ] && [ -f "/usr/local/cuda/lib64/libcudart.so.${CUDA_MAJOR}" ]; then ln -s "libcudart.so.${CUDA_MAJOR}" /usr/local/cuda/lib64/libcudart.so; fi; fi; fi # buildkit

ENV VLLM_USAGE_SOURCE=production-docker-image

ENV VLLM_BUILD_COMMIT=6e448d0ea9bf3d88d898b65449ca6dc2aec170ac VLLM_BUILD_PIPELINE=019d130e-464e-4ff7-b84b-492992c0c06b VLLM_BUILD_URL=https://buildkite.com/vllm/release-v2/builds/5006 VLLM_IMAGE_TAG=vllm/vllm-openai:v0.27.1-ubuntu2404

LABEL org.opencontainers.image.source=https://github.com/vllm-project/vllm org.opencontainers.image.revision=6e448d0ea9bf3d88d898b65449ca6dc2aec170ac org.opencontainers.image.version=vllm/vllm-openai:v0.27.1-ubuntu2404 org.opencontainers.image.url=https://buildkite.com/vllm/release-v2/builds/5006 ai.vllm.build.commit=6e448d0ea9bf3d88d898b65449ca6dc2aec170ac ai.vllm.build.pipeline=019d130e-464e-4ff7-b84b-492992c0c06b ai.vllm.build.url=https://buildkite.com/vllm/release-v2/builds/5006 ai.vllm.image.tag=vllm/vllm-openai:v0.27.1-ubuntu2404

ENTRYPOINT ["vllm" "serve"]

SHELL [/bin/bash -eo pipefail -c]

ARG TARGETPLATFORM=linux/arm64

ARG TARGETOS=linux

ARG TARGETARCH=arm64

ENV DEBIAN_FRONTEND=noninteractive LANG=en_US.UTF-8 LANGUAGE=en_US:en LC_ALL=en_US.UTF-8

RUN |3 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 /bin/bash -eo pipefail -c # Tools # Refresh apt-get update -y && apt-get install -y --no-install-recommends \ software-properties-common apt-transport-https \ ca-certificates gnupg2 lsb-release gnupg-agent \ && apt-get update -y # Install apt-get install -y --no-install-recommends \ ca-certificates build-essential binutils bash openssl \ curl wget aria2 \ git git-lfs \ unzip xz-utils \ tzdata locales \ iproute2 iputils-ping ifstat net-tools dnsutils pciutils ipmitool \ rdma-core rdmacm-utils infiniband-diags \ procps sysstat htop \ tini vim jq bc tree # Update locale localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8 # Update timezone rm -f /etc/localtime \ && ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime \ && echo "Asia/Shanghai" > /etc/timezone \ && dpkg-reconfigure --frontend noninteractive tzdata # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* \ && rm -rf /var/cache/apt # buildkit

ARG PYTHON_VERSION=3.12

ENV PYTHON_VERSION=3.12

RUN |4 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 /bin/bash -eo pipefail -c # Python # Update alternatives if ! command -v python >/dev/null 2>&1 \ && command -v python${PYTHON_VERSION} >/dev/null 2>&1; then update-alternatives --install /usr/bin/python python $(which python${PYTHON_VERSION}) 1 fi # buildkit

ENV PIP_NO_CACHE_DIR=1 PIP_DISABLE_PIP_VERSION_CHECK=1 PIP_ROOT_USER_ACTION=ignore

RUN |4 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 /bin/bash -eo pipefail -c # Buildkit cat <<EOT >/tmp/requirements.txt build cmake<4 wheel pybind11<3 Cython EOT pip install -r /tmp/requirements.txt # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit

ARG CUDA_VERSION=13.0.1

ARG CUDA_ARCHS

ENV CUDA_HOME=/usr/local/cuda CUDA_VERSION=13.0.1 CUDA_ARCHS=

ENV UV_SYSTEM_PYTHON=1 UV_PRERELEASE=allow UV_SKIP_WHEEL_FILENAME_CHECK=1 UV_NO_CACHE=1 UV_HTTP_TIMEOUT=500 UV_INDEX_STRATEGY=unsafe-best-match

ARG VLLM_VERSION=0.27.1

ARG VLLM_TORCH_VERSION=2.13.0

ARG VLLM_TORCH_CUDA_VERSION=13.0.1

ENV VLLM_VERSION=0.27.1 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=13.0.1

SHELL [/bin/bash -eo pipefail -c]

ARG TARGETPLATFORM=linux/arm64

ARG TARGETOS=linux

ARG TARGETARCH=arm64

RUN |9 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=13.0.1 CUDA_ARCHS= VLLM_VERSION=0.27.1 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=13.0.1 /bin/bash -eo pipefail -c # Ray # Install Ray Client and Default. # Pinned to 2.54.0: ray 2.55 (ray-project/ray#61029) breaks distributed # ray.init when the driver has no co-located raylet (gpustack split containers). RAY_VERSION=2.54.0 cat <<EOT >/tmp/requirements.txt ray[cgraph]==${RAY_VERSION} ray[client]==${RAY_VERSION} ray[default]==${RAY_VERSION} EOT uv pip install \ -r /tmp/requirements.txt # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit

RUN |9 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=13.0.1 CUDA_ARCHS= VLLM_VERSION=0.27.1 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=13.0.1 /bin/bash -eo pipefail -c # Omni if [[ ! -d /omni/workspace ]]; then echo "Skipping vLLM Omni installation for ${TARGETARCH}..." exit 0 fi # Install uv pip install --no-build-isolation \ /omni/workspace/*.whl # Dependencies cat <<EOT >/tmp/requirements.txt onnxruntime sox EOT uv pip install \ -r /tmp/requirements.txt # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit

ARG VLLM_LMCACHE_VERSION=0.5.4

ENV VLLM_LMCACHE_VERSION=0.5.4

RUN |10 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=13.0.1 CUDA_ARCHS= VLLM_VERSION=0.27.1 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=13.0.1 VLLM_LMCACHE_VERSION=0.5.4 /bin/bash -eo pipefail -c # LMCache # Install ## Replaces the base's own lmcache —— see the note on VLLM_LMCACHE_VERSION. Must run before ## the CuPy dedup below, since lmcache declares a CUDA-major-specific cupy variant. uv pip install --no-build-isolation \ /lmcache/workspace/*.whl # Review uv pip show lmcache ## Fail here on a torch ABI mismatch: lmcache itself only WARNs and falls back to its slow ## torch baseline. RTLD_NOW resolves all symbols without module init, so this needs no GPU. cat <<EOT >/tmp/check_lmcache.py import ctypes, glob, os, sysconfig, torch platlib = sysconfig.get_paths()["platlib"] sos = glob.glob(platlib + "/lmcache/*.so") assert sos, "no lmcache extension found under " + platlib for so in sos: ctypes.CDLL(so, mode=os.RTLD_NOW) print("lmcache resolved against torch " + torch.__version__ + ": " + ", ".join(sos)) EOT python /tmp/check_lmcache.py # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit

RUN |10 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=13.0.1 CUDA_ARCHS= VLLM_VERSION=0.27.1 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=13.0.1 VLLM_LMCACHE_VERSION=0.5.4 /bin/bash -eo pipefail -c # CuPy # ray[cgraph] pulls cupy-cuda12x while the base already ships cupy-cuda13x; both share # the same cupy/ tree, so uninstalling one breaks the other. Purge both, then reinstall # the CUDA-matching variant at the version vLLM pinned in the base (its cupy-cuda13x), # so cu129 also gets that pinned version instead of ray's unconstrained cupy-cuda12x. # Our own lmcache is a third source but agrees with this reinstall: its build stage sets # LMCACHE_CUDA_MAJOR from the same CUDA major used here. IFS="." read -r CUDA_MAJOR CUDA_MINOR CUDA_PATCH <<< "${CUDA_VERSION}" PLATLIB="$(python -c 'import sysconfig; print(sysconfig.get_paths()["platlib"])')" CUPY="cupy-cuda${CUDA_MAJOR}x" CUPY_VER="$(uv pip show cupy-cuda13x 2>/dev/null | awk '/^Version:/{print $2}' || true)" uv pip uninstall cupy-cuda12x cupy-cuda13x || true rm -rf "${PLATLIB}"/cupy "${PLATLIB}"/cupyx "${PLATLIB}"/cupy_backends # cupy-cuda12x and cupy-cuda13x share the same version number, so reuse the base's # pinned version and install only the variant matching the CUDA major we build for. # Fail loudly if the base no longer ships cupy, rather than installing an unpinned latest. if [[ -z "${CUPY_VER}" ]]; then echo "cupy-cuda13x not found in base image; refusing to install unpinned cupy" >&2 exit 1 fi uv pip install "${CUPY}==${CUPY_VER}" # Review uv pip show "${CUPY}" # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit

RUN |10 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=13.0.1 CUDA_ARCHS= VLLM_VERSION=0.27.1 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=13.0.1 VLLM_LMCACHE_VERSION=0.5.4 /bin/bash -eo pipefail -c # Postprocess # Review uv pip tree \ --package vllm \ --package vllm-omni \ --package flashinfer-python \ --package torch \ --package triton \ --package deep-gemm \ --package deep-ep \ --package lmcache \ --package mooncake-transfer-engine \ --package ray # buildkit

RUN |10 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=13.0.1 CUDA_ARCHS= VLLM_VERSION=0.27.1 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=13.0.1 VLLM_LMCACHE_VERSION=0.5.4 /bin/bash -eo pipefail -c # Patch tree -hs /workspace/patches pushd $(pip show vllm | grep Location: | cut -d" " -f 2) \ && for f in /workspace/patches/vllm/*.patch; do patch -p1 < "$f" || exit 1; done if pip show vllm_omni > /dev/null 2>&1; then \ pushd $(pip show vllm_omni | grep Location: | cut -d" " -f 2) \ && patch -p1 < /workspace/patches/vllm_omni/*.patch; \ fi # buildkit

RUN |10 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=13.0.1 CUDA_ARCHS= VLLM_VERSION=0.27.1 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=13.0.1 VLLM_LMCACHE_VERSION=0.5.4 /bin/bash -eo pipefail -c # Dependencies # Install cat <<EOT >/tmp/requirements.txt # audio extras librosa soundfile mistral_common[audio] # tokenizer extras fastokens==0.2.0 EOT uv pip install \ -r /tmp/requirements.txt # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit

ENV RAY_EXPERIMENTAL_NOSET_CUDA_VISIBLE_DEVICES=1 SAFETENSORS_FAST_GPU=1 RUNAI_STREAMER_MEMORY_LIMIT=0 RUNAI_STREAMER_LOG_TO_STDERR=1 RUNAI_STREAMER_LOG_LEVEL=INFO

WORKDIR /

ENTRYPOINT ["tini" "--"]

SHELL [/bin/bash -eo pipefail -c]

ARG TARGETPLATFORM=linux/arm64

ARG TARGETOS=linux

ARG TARGETARCH=arm64

RUN |3 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 /bin/bash -eo pipefail -c # Give MooncakeConnector the Prometheus registration MultiConnector requires VLLM_DIR="$(python3 -c 'import os, vllm; print(os.path.dirname(vllm.__file__))')" MOONCAKE_DIR="${VLLM_DIR}/distributed/kv_transfer/kv_connector/v1/mooncake" # Refuse to guess at a layout this patch was not written against. test -f "${MOONCAKE_DIR}/stats.py" test -f "${MOONCAKE_DIR}/mooncake_connector.py" # Idempotent: a re-run over an already patched image must not append twice. if grep -q "class MooncakePromMetrics" "${MOONCAKE_DIR}/stats.py"; then echo "[info] MooncakePromMetrics is already present; nothing to append." exit 0 fi cat >>"${MOONCAKE_DIR}/stats.py" <<'PYEOF' from vllm.config import VllmConfig from vllm.distributed.kv_transfer.kv_connector.v1.metrics import ( KVConnectorPromMetrics, PromMetric, PromMetricT, ) from vllm.v1.metrics.utils import create_metric_per_engine class MooncakePromMetrics(KVConnectorPromMetrics): """Prometheus metrics for Mooncake point-to-point KV transfers. Mirrors NixlPromMetrics: the histograms take one observation per recorded transfer, and the counters take a list of 1s, which is the shape MooncakeKVConnectorStats.reset() already produces. The metric names are prefixed `vllm:mooncake_` rather than `vllm:mooncake_store_`, which MooncakeStorePromMetrics owns: the two connectors can be active in the same engine through a MultiConnector, and one namespace for both would merge a direct engine-to-engine transfer with a round trip to the shared store. """ def __init__( self, vllm_config: VllmConfig, metric_types: dict[type[PromMetric], type[PromMetricT]], labelnames: list[str], per_engine_labelvalues: dict[int, list[object]], ): super().__init__(vllm_config, metric_types, labelnames, per_engine_labelvalues) duration_buckets = [ 0.001, 0.005, 0.01, 0.025, 0.05, 0.075, 0.1, 0.2, 0.3, 0.5, 0.75, 1.0, 5.0, ] histogram_xfer_time = self._histogram_cls( name="vllm:mooncake_xfer_time_seconds", documentation="Histogram of transfer duration for Mooncake KV cache transfers.", buckets=duration_buckets, labelnames=labelnames, ) self.histogram_xfer_time = create_metric_per_engine( histogram_xfer_time, self.per_engine_labelvalues ) # Uniform 2KiB to 16GiB range, as NixlPromMetrics uses. byte_buckets = [2 ** (10 + i) for i in range(1, 25, 2)] histogram_bytes_transferred = self._histogram_cls( name="vllm:mooncake_bytes_transferred", documentation="Histogram of bytes transferred per Mooncake KV cache transfer.", buckets=byte_buckets, labelnames=labelnames, ) self.histogram_bytes_transferred = create_metric_per_engine( histogram_bytes_transferred, self.per_engine_labelvalues ) descriptor_buckets = [10, 20, 30, 50, 75, 100, 200, 300, 500, 750, 1000] histogram_num_descriptors = self._histogram_cls( name="vllm:mooncake_num_descriptors", documentation="Histogram of descriptors per Mooncake KV cache transfer.", buckets=descriptor_buckets, labelnames=labelnames, ) self.histogram_num_descriptors = create_metric_per_engine( histogram_num_descriptors, self.per_engine_labelvalues ) counter_num_failed_transfers = self._counter_cls( name="vllm:mooncake_num_failed_transfers", documentation="Number of failed Mooncake KV cache transfers.", labelnames=labelnames, ) self.counter_num_failed_transfers = create_metric_per_engine( counter_num_failed_transfers, self.per_engine_labelvalues ) counter_num_failed_recvs = self._counter_cls( name="vllm:mooncake_num_failed_recvs", documentation="Number of failed Mooncake KV cache receives.", labelnames=labelnames, ) self.counter_num_failed_recvs = create_metric_per_engine( counter_num_failed_recvs, self.per_engine_labelvalues ) counter_num_kv_expired_reqs = self._counter_cls( name="vllm:mooncake_num_kv_expired_reqs", documentation="Number of requests whose Mooncake KV blocks expired.", labelnames=labelnames, ) self.counter_num_kv_expired_reqs = create_metric_per_engine( counter_num_kv_expired_reqs, self.per_engine_labelvalues ) def observe(self, transfer_stats_data: dict[str, Any], engine_idx: int = 0): # A key absent from the payload is skipped rather than raising: the stats # container's key set belongs to the connector and may grow or shrink # between releases, and a metrics path is the wrong place to fail an engine. for prom_obj, key in zip( [ self.histogram_xfer_time, self.histogram_bytes_transferred, self.histogram_num_descriptors, ], ["transfer_duration", "bytes_transferred", "num_descriptors"], ): for list_item in transfer_stats_data.get(key, ()): prom_obj[engine_idx].observe(list_item) for counter_obj, key in zip( [ self.counter_num_failed_transfers, self.counter_num_failed_recvs, self.counter_num_kv_expired_reqs, ], ["num_failed_transfers", "num_failed_recvs", "num_kv_expired_reqs"], ): for list_item in transfer_stats_data.get(key, ()): counter_obj[engine_idx].inc(list_item) PYEOF cat >>"${MOONCAKE_DIR}/mooncake_connector.py" <<'PYEOF' def _build_mooncake_prom_metrics( cls, vllm_config, metric_types, labelnames, per_engine_labelvalues ): # Imported here rather than at module scope: this file already imports from # .stats, and keeping the dependency inside the call leaves the module's # import graph exactly as it was. from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.stats import ( MooncakePromMetrics, ) return MooncakePromMetrics( vllm_config, metric_types, labelnames, per_engine_labelvalues ) # Bound by assignment rather than written into the class body, so that this patch # stays an append and needs no anchor inside a file that differs between releases. MooncakeConnector.build_prom_metrics = classmethod(_build_mooncake_prom_metrics) PYEOF # Assert the patch landed. Both greps are POSITIVE: a negated grep is exempt from # set -e, so the inverted form would pass over a failure. grep -q "class MooncakePromMetrics" "${MOONCAKE_DIR}/stats.py" grep -q "MooncakeConnector.build_prom_metrics = classmethod" "${MOONCAKE_DIR}/mooncake_connector.py" # Import it for real. The greps above prove the text is there; this proves it # parses and that every name it reaches for resolves in this image. python3 -c " from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.stats import MooncakePromMetrics from vllm.distributed.kv_transfer.kv_connector.v1.metrics import KVConnectorPromMetrics assert issubclass(MooncakePromMetrics, KVConnectorPromMetrics) assert MooncakePromMetrics.observe is not KVConnectorPromMetrics.observe print('[info] MooncakePromMetrics imports and overrides observe.') " # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit

WORKDIR /

ENTRYPOINT ["tini" "--"]

SHELL [/bin/bash -eo pipefail -c]

ARG TARGETPLATFORM=linux/arm64

ARG TARGETOS=linux

ARG TARGETARCH=arm64

ARG VLLM_ROUTER_VERSION=0.1.15

RUN |4 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 VLLM_ROUTER_VERSION=0.1.15 /bin/bash -eo pipefail -c # vLLM Router uv pip install \ vllm-router==${VLLM_ROUTER_VERSION} # Fail the build rather than ship an image whose router is missing: the # alternative surfaces as a container that exits at deploy time, one # layer away from anything that explains it. command -v vllm-router >/dev/null # Review uv pip tree # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit

ARG DEPENDENCY_PACKAGES=diffusers flash-attn flashinfer-python lmcache lmcache-ascend mooncake-transfer-engine-npu mooncake-transfer-engine-rocm mooncake-transfer-engine ray runai-model-streamer sgl-kernel-npu sglang-kernel sglang-router torch torch-musa torch-npu transformers vllm-ascend vllm-metax vllm-musa vllm-omni vllm-router

RUN |5 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 VLLM_ROUTER_VERSION=0.1.15 DEPENDENCY_PACKAGES=diffusers flash-attn flashinfer-python lmcache lmcache-ascend mooncake-transfer-engine-npu mooncake-transfer-engine-rocm mooncake-transfer-engine ray runai-model-streamer sgl-kernel-npu sglang-kernel sglang-router torch torch-musa torch-npu transformers vllm-ascend vllm-metax vllm-musa vllm-omni vllm-router /bin/bash -eo pipefail -c DEPENDENCY_PACKAGES="${DEPENDENCY_PACKAGES}" bash /tmp/probe_dependencies.sh # buildkit

WORKDIR /

ENTRYPOINT ["tini" "--"]