deepseek-spark (sha256:027c9769b28df9ebd59537076b0266199263bd74793bafcb52b6f4e56acf072a)

Published 2026-10-06 18:07:05 +00:00 by FML128

Pull the image from the command line:

docker pull git.fml128.ch/fml128/deepseek-spark@sha256:027c9769b28df9ebd59537076b0266199263bd74793bafcb52b6f4e56acf072a

Digest

sha256:027c9769b28df9ebd59537076b0266199263bd74793bafcb52b6f4e56acf072a

For more information on the Container registry, see the documentation.

ARG RELEASE

ARG LAUNCHPAD_BUILD_ARCH

LABEL org.opencontainers.image.ref.name=ubuntu

LABEL org.opencontainers.image.version=24.04

ADD file:6089c6bede9eca8ec4f424e5798a0ae0712a6fe38c9b97f9afb9d24d9675024e in /

CMD ["/bin/bash"]

ARG JETPACK_HOST_MOUNTS=

ARG ENABLE_MITMPROXY=0

ENV NVIDIA_REQUIRE_JETPACK_HOST_MOUNTS=

RUN |2 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 /bin/sh -c if [ -n "${JETPACK_HOST_MOUNTS}" ]; then echo "/usr/lib/aarch64-linux-gnu/tegra" > /etc/ld.so.conf.d/nvidia-tegra.conf && echo "/usr/lib/aarch64-linux-gnu/tegra-egl" >> /etc/ld.so.conf.d/nvidia-tegra.conf; fi # buildkit

RUN |2 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 /bin/sh -c /tmp/manage_cert.sh install # buildkit

RUN |2 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 /bin/sh -c if [ -d /usr/share/ca-certificates ]; then exit; fi && export DEBIAN_FRONTEND=noninteractive && apt-get update && apt-get install -y --no-install-recommends ca-certificates && rm -rf /var/lib/apt/lists/* # buildkit

RUN |2 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 /bin/sh -c /tmp/manage_cert.sh install # buildkit

RUN |2 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 /bin/sh -c export DEBIAN_FRONTEND=noninteractive && apt-get update && apt-get install -y --no-install-recommends adduser curl libnl-route-3-200 libnl-3-200 libnl-3-dev libnl-route-3-dev patch wget && rm -rf /var/lib/apt/lists/* && echo "hsts=0" > /root/.wgetrc # buildkit

ARG GDRCOPY_VERSION=2.5.1

ARG HPCX_VERSION=2.25.1-RC2

ARG RDMACORE_VERSION=56.0

ARG MOFED_VERSION=5.4-rdmacore56.0

ARG OPENUCX_VERSION=1.20.0

ARG OPENMPI_VERSION=4.1.7

ARG EFA_VERSION=1.43.1

ARG AWS_OFI_NCCL_VERSION=1.17.0

ENV GDRCOPY_VERSION=2.5.1 HPCX_VERSION=2.25.1-RC2 MOFED_VERSION=5.4-rdmacore56.0 OPENUCX_VERSION=1.20.0 OPENMPI_VERSION=4.1.7 RDMACORE_VERSION=56.0 EFA_VERSION=1.43.1 AWS_OFI_NCCL_VERSION=1.17.0

ARG TARGETARCH=arm64

RUN |11 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 GDRCOPY_VERSION=2.5.1 HPCX_VERSION=2.25.1-RC2 RDMACORE_VERSION=56.0 MOFED_VERSION=5.4-rdmacore56.0 OPENUCX_VERSION=1.20.0 OPENMPI_VERSION=4.1.7 EFA_VERSION=1.43.1 AWS_OFI_NCCL_VERSION=1.17.0 TARGETARCH=arm64 /bin/sh -c cd /nvidia && ( cd opt/rdma-core/ && dpkg -i libibverbs1_*.deb libibverbs-dev_*.deb librdmacm1_*.deb librdmacm-dev_*.deb libibumad3_*.deb libibumad-dev_*.deb ibverbs-utils_*.deb ibverbs-providers_*.deb && rm $(dpkg-query -L libibverbs-dev librdmacm-dev libibumad-dev | grep "$\.so\|\.a$$") ) && ( cd opt/gdrcopy/ && dpkg -i libgdrapi_*.deb ) && ( cp -r opt/hpcx /opt/ && cp etc/ld.so.conf.d/hpcx.conf /etc/ld.so.conf.d/ && ln -sf /opt/hpcx/ompi /usr/local/mpi && ln -sf /opt/hpcx/ucx /usr/local/ucx && sed -i 's/^$hwloc_base_binding_policy$ = core$/\1 = none/' /opt/hpcx/ompi/etc/openmpi-mca-params.conf && sed -i 's/^$btl = self$$/#\1/' /opt/hpcx/ompi/etc/openmpi-mca-params.conf ) && ( if [ ! -f /etc/ld.so.conf.d/nvidia-tegra.conf ]; then cd opt/amazon/efa/ && dpkg -i libfabric*.deb && rm /opt/amazon/efa/lib/libfabric.a && echo "/opt/amazon/efa/lib" > /etc/ld.so.conf.d/efa.conf; fi ) && ldconfig # buildkit

ENV OPAL_PREFIX=/opt/hpcx/ompi PATH=/usr/local/mpi/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/local/ucx/bin:/opt/amazon/efa/bin

ENV OMPI_MCA_coll_hcoll_enable=0

ARG CUDA_VERSION=13.1.1.006

ARG CUDA_DRIVER_VERSION=590.48.01

ARG NVVM_VERSION=13.1.115

ARG DOCA_VERSION=3.1.0

ENV CUDA_VERSION=13.1.1.006 CUDA_DRIVER_VERSION=590.48.01 NVVM_VERSION=13.1.115 DOCA_VERSION=3.1.0

RUN |15 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 GDRCOPY_VERSION=2.5.1 HPCX_VERSION=2.25.1-RC2 RDMACORE_VERSION=56.0 MOFED_VERSION=5.4-rdmacore56.0 OPENUCX_VERSION=1.20.0 OPENMPI_VERSION=4.1.7 EFA_VERSION=1.43.1 AWS_OFI_NCCL_VERSION=1.17.0 TARGETARCH=arm64 CUDA_VERSION=13.1.1.006 CUDA_DRIVER_VERSION=590.48.01 NVVM_VERSION=13.1.115 DOCA_VERSION=3.1.0 /bin/sh -c BASE=min /nvidia/build-scripts/installCUDA.sh # buildkit

RUN |15 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 GDRCOPY_VERSION=2.5.1 HPCX_VERSION=2.25.1-RC2 RDMACORE_VERSION=56.0 MOFED_VERSION=5.4-rdmacore56.0 OPENUCX_VERSION=1.20.0 OPENMPI_VERSION=4.1.7 EFA_VERSION=1.43.1 AWS_OFI_NCCL_VERSION=1.17.0 TARGETARCH=arm64 CUDA_VERSION=13.1.1.006 CUDA_DRIVER_VERSION=590.48.01 NVVM_VERSION=13.1.115 DOCA_VERSION=3.1.0 /bin/sh -c cp -vprd /nvidia/. / && patch -p0 < /etc/startup_scripts.patch && rm -f /etc/startup_scripts.patch # buildkit

ENV _CUDA_COMPAT_PATH=/usr/local/cuda/compat ENV=/etc/shinit_v2 BASH_ENV=/etc/bash.bashrc SHELL=/bin/bash NVIDIA_REQUIRE_CUDA=cuda>=9.0

LABEL com.nvidia.volumes.needed=nvidia_driver com.nvidia.cuda.version=9.0

ARG NCCL_VERSION=2.29.stable.20260109 CUBLAS_VERSION=13.2.1.1 CUFFT_VERSION=12.1.0.78 CURAND_VERSION=10.4.1.81 CUSPARSE_VERSION=12.7.3.1 CUSOLVER_VERSION=12.0.9.81 NPP_VERSION=13.0.3.3 NVJPEG_VERSION=13.0.3.75 CUFILE_VERSION=1.16.1.26 NVJITLINK_VERSION=13.1.115 NVFATBIN_VERSION=13.1.115 CUBLASMP_VERSION=0.7.0.125 NVSHMEM_VERSION=3.4.5 CUDLA_VERSION=13.1.1.006 NVPTXCOMPILER_VERSION=13.1.115 CUDNN_VERSION=9.19.0.56 CUDNN_FRONTEND_VERSION=1.18.0 TRT_VERSION=10.15.1.29 TRTOSS_VERSION= NSIGHT_SYSTEMS_VERSION=2026.1.1.204 NSIGHT_COMPUTE_VERSION=2025.4.1.2 CUSPARSELT_VERSION=0.8.1.1 DALI_VERSION=1.53.0 DALI_BUILD= DALI_URL_SUFFIX=130 POLYGRAPHY_VERSION=0.49.26 TRANSFORMER_ENGINE_VERSION=2.12 MODEL_OPT_VERSION=0.41.0 CUDA_ARCH_LIST=8.0 8.6 9.0 10.0 11.0 12.0 MAXSMVER=121 NVRX_VERSION=0.5.0

ENV NCCL_VERSION=2.29.stable.20260109 CUBLAS_VERSION=13.2.1.1 CUFFT_VERSION=12.1.0.78 CURAND_VERSION=10.4.1.81 CUSPARSE_VERSION=12.7.3.1 CUSPARSELT_VERSION=0.8.1.1 CUSOLVER_VERSION=12.0.9.81 NPP_VERSION=13.0.3.3 NVJPEG_VERSION=13.0.3.75 CUFILE_VERSION=1.16.1.26 NVJITLINK_VERSION=13.1.115 NVFATBIN_VERSION=13.1.115 CUBLASMP_VERSION=0.7.0.125 NVSHMEM_VERSION=3.4.5 CUDLA_VERSION=13.1.1.006 NVPTXCOMPILER_VERSION=13.1.115 CUDNN_VERSION=9.19.0.56 CUDNN_FRONTEND_VERSION=1.18.0 TRT_VERSION=10.15.1.29 TRTOSS_VERSION= NSIGHT_SYSTEMS_VERSION=2026.1.1.204 NSIGHT_COMPUTE_VERSION=2025.4.1.2 DALI_VERSION=1.53.0 DALI_BUILD= DALI_URL_SUFFIX=130 POLYGRAPHY_VERSION=0.49.26 TRANSFORMER_ENGINE_VERSION=2.12 MODEL_OPT_VERSION=0.41.0 CUDA_ARCH_LIST=8.0 8.6 9.0 10.0 11.0 12.0 MAXSMVER=121 NVRX_VERSION=0.5.0

ADD docs.tgz / # buildkit

RUN |46 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 GDRCOPY_VERSION=2.5.1 HPCX_VERSION=2.25.1-RC2 RDMACORE_VERSION=56.0 MOFED_VERSION=5.4-rdmacore56.0 OPENUCX_VERSION=1.20.0 OPENMPI_VERSION=4.1.7 EFA_VERSION=1.43.1 AWS_OFI_NCCL_VERSION=1.17.0 TARGETARCH=arm64 CUDA_VERSION=13.1.1.006 CUDA_DRIVER_VERSION=590.48.01 NVVM_VERSION=13.1.115 DOCA_VERSION=3.1.0 NCCL_VERSION=2.29.stable.20260109 CUBLAS_VERSION=13.2.1.1 CUFFT_VERSION=12.1.0.78 CURAND_VERSION=10.4.1.81 CUSPARSE_VERSION=12.7.3.1 CUSOLVER_VERSION=12.0.9.81 NPP_VERSION=13.0.3.3 NVJPEG_VERSION=13.0.3.75 CUFILE_VERSION=1.16.1.26 NVJITLINK_VERSION=13.1.115 NVFATBIN_VERSION=13.1.115 CUBLASMP_VERSION=0.7.0.125 NVSHMEM_VERSION=3.4.5 CUDLA_VERSION=13.1.1.006 NVPTXCOMPILER_VERSION=13.1.115 CUDNN_VERSION=9.19.0.56 CUDNN_FRONTEND_VERSION=1.18.0 TRT_VERSION=10.15.1.29 TRTOSS_VERSION= NSIGHT_SYSTEMS_VERSION=2026.1.1.204 NSIGHT_COMPUTE_VERSION=2025.4.1.2 CUSPARSELT_VERSION=0.8.1.1 DALI_VERSION=1.53.0 DALI_BUILD= DALI_URL_SUFFIX=130 POLYGRAPHY_VERSION=0.49.26 TRANSFORMER_ENGINE_VERSION=2.12 MODEL_OPT_VERSION=0.41.0 CUDA_ARCH_LIST=8.0 8.6 9.0 10.0 11.0 12.0 MAXSMVER=121 NVRX_VERSION=0.5.0 /bin/sh -c echo "/usr/local/nvidia/lib" >> /etc/ld.so.conf.d/nvidia.conf && echo "/usr/local/nvidia/lib64" >> /etc/ld.so.conf.d/nvidia.conf # buildkit

ARG _LIBPATH_SUFFIX=

ENV PATH=/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/mpi/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/local/ucx/bin:/opt/amazon/efa/bin LD_LIBRARY_PATH=/usr/local/cuda/compat/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64 NVIDIA_VISIBLE_DEVICES=all NVIDIA_DRIVER_CAPABILITIES=compute,utility,video

RUN |47 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 GDRCOPY_VERSION=2.5.1 HPCX_VERSION=2.25.1-RC2 RDMACORE_VERSION=56.0 MOFED_VERSION=5.4-rdmacore56.0 OPENUCX_VERSION=1.20.0 OPENMPI_VERSION=4.1.7 EFA_VERSION=1.43.1 AWS_OFI_NCCL_VERSION=1.17.0 TARGETARCH=arm64 CUDA_VERSION=13.1.1.006 CUDA_DRIVER_VERSION=590.48.01 NVVM_VERSION=13.1.115 DOCA_VERSION=3.1.0 NCCL_VERSION=2.29.stable.20260109 CUBLAS_VERSION=13.2.1.1 CUFFT_VERSION=12.1.0.78 CURAND_VERSION=10.4.1.81 CUSPARSE_VERSION=12.7.3.1 CUSOLVER_VERSION=12.0.9.81 NPP_VERSION=13.0.3.3 NVJPEG_VERSION=13.0.3.75 CUFILE_VERSION=1.16.1.26 NVJITLINK_VERSION=13.1.115 NVFATBIN_VERSION=13.1.115 CUBLASMP_VERSION=0.7.0.125 NVSHMEM_VERSION=3.4.5 CUDLA_VERSION=13.1.1.006 NVPTXCOMPILER_VERSION=13.1.115 CUDNN_VERSION=9.19.0.56 CUDNN_FRONTEND_VERSION=1.18.0 TRT_VERSION=10.15.1.29 TRTOSS_VERSION= NSIGHT_SYSTEMS_VERSION=2026.1.1.204 NSIGHT_COMPUTE_VERSION=2025.4.1.2 CUSPARSELT_VERSION=0.8.1.1 DALI_VERSION=1.53.0 DALI_BUILD= DALI_URL_SUFFIX=130 POLYGRAPHY_VERSION=0.49.26 TRANSFORMER_ENGINE_VERSION=2.12 MODEL_OPT_VERSION=0.41.0 CUDA_ARCH_LIST=8.0 8.6 9.0 10.0 11.0 12.0 MAXSMVER=121 NVRX_VERSION=0.5.0 _LIBPATH_SUFFIX= /bin/sh -c if [ -n "${DOCA_VERSION}" ] && dpkg --compare-versions "${HPCX_VERSION}" "ge" "2.24"; then /nvidia/build-scripts/installDOCA.sh; else echo "Not running installDOCA.sh"; fi # buildkit

COPY /opt/amazon/aws-ofi-nccl /opt/amazon/aws-ofi-nccl # buildkit

RUN |47 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 GDRCOPY_VERSION=2.5.1 HPCX_VERSION=2.25.1-RC2 RDMACORE_VERSION=56.0 MOFED_VERSION=5.4-rdmacore56.0 OPENUCX_VERSION=1.20.0 OPENMPI_VERSION=4.1.7 EFA_VERSION=1.43.1 AWS_OFI_NCCL_VERSION=1.17.0 TARGETARCH=arm64 CUDA_VERSION=13.1.1.006 CUDA_DRIVER_VERSION=590.48.01 NVVM_VERSION=13.1.115 DOCA_VERSION=3.1.0 NCCL_VERSION=2.29.stable.20260109 CUBLAS_VERSION=13.2.1.1 CUFFT_VERSION=12.1.0.78 CURAND_VERSION=10.4.1.81 CUSPARSE_VERSION=12.7.3.1 CUSOLVER_VERSION=12.0.9.81 NPP_VERSION=13.0.3.3 NVJPEG_VERSION=13.0.3.75 CUFILE_VERSION=1.16.1.26 NVJITLINK_VERSION=13.1.115 NVFATBIN_VERSION=13.1.115 CUBLASMP_VERSION=0.7.0.125 NVSHMEM_VERSION=3.4.5 CUDLA_VERSION=13.1.1.006 NVPTXCOMPILER_VERSION=13.1.115 CUDNN_VERSION=9.19.0.56 CUDNN_FRONTEND_VERSION=1.18.0 TRT_VERSION=10.15.1.29 TRTOSS_VERSION= NSIGHT_SYSTEMS_VERSION=2026.1.1.204 NSIGHT_COMPUTE_VERSION=2025.4.1.2 CUSPARSELT_VERSION=0.8.1.1 DALI_VERSION=1.53.0 DALI_BUILD= DALI_URL_SUFFIX=130 POLYGRAPHY_VERSION=0.49.26 TRANSFORMER_ENGINE_VERSION=2.12 MODEL_OPT_VERSION=0.41.0 CUDA_ARCH_LIST=8.0 8.6 9.0 10.0 11.0 12.0 MAXSMVER=121 NVRX_VERSION=0.5.0 _LIBPATH_SUFFIX= /bin/sh -c if [ ! -f /etc/ld.so.conf.d/nvidia-tegra.conf ]; then echo "/opt/amazon/aws-ofi-nccl/lib" > /etc/ld.so.conf.d/aws-ofi-nccl.conf && ldconfig; fi # buildkit

COPY entrypoint/ /opt/nvidia/ # buildkit

ENV NVIDIA_PRODUCT_NAME=CUDA

ENTRYPOINT ["/opt/nvidia/nvidia_entrypoint.sh"]

RUN |47 JETPACK_HOST_MOUNTS= ENABLE_MITMPROXY=0 GDRCOPY_VERSION=2.5.1 HPCX_VERSION=2.25.1-RC2 RDMACORE_VERSION=56.0 MOFED_VERSION=5.4-rdmacore56.0 OPENUCX_VERSION=1.20.0 OPENMPI_VERSION=4.1.7 EFA_VERSION=1.43.1 AWS_OFI_NCCL_VERSION=1.17.0 TARGETARCH=arm64 CUDA_VERSION=13.1.1.006 CUDA_DRIVER_VERSION=590.48.01 NVVM_VERSION=13.1.115 DOCA_VERSION=3.1.0 NCCL_VERSION=2.29.stable.20260109 CUBLAS_VERSION=13.2.1.1 CUFFT_VERSION=12.1.0.78 CURAND_VERSION=10.4.1.81 CUSPARSE_VERSION=12.7.3.1 CUSOLVER_VERSION=12.0.9.81 NPP_VERSION=13.0.3.3 NVJPEG_VERSION=13.0.3.75 CUFILE_VERSION=1.16.1.26 NVJITLINK_VERSION=13.1.115 NVFATBIN_VERSION=13.1.115 CUBLASMP_VERSION=0.7.0.125 NVSHMEM_VERSION=3.4.5 CUDLA_VERSION=13.1.1.006 NVPTXCOMPILER_VERSION=13.1.115 CUDNN_VERSION=9.19.0.56 CUDNN_FRONTEND_VERSION=1.18.0 TRT_VERSION=10.15.1.29 TRTOSS_VERSION= NSIGHT_SYSTEMS_VERSION=2026.1.1.204 NSIGHT_COMPUTE_VERSION=2025.4.1.2 CUSPARSELT_VERSION=0.8.1.1 DALI_VERSION=1.53.0 DALI_BUILD= DALI_URL_SUFFIX=130 POLYGRAPHY_VERSION=0.49.26 TRANSFORMER_ENGINE_VERSION=2.12 MODEL_OPT_VERSION=0.41.0 CUDA_ARCH_LIST=8.0 8.6 9.0 10.0 11.0 12.0 MAXSMVER=121 NVRX_VERSION=0.5.0 _LIBPATH_SUFFIX= /bin/sh -c mkdir -p /workspace && cp -f -p /opt/nvidia/entrypoint.d/30-container-license.txt /workspace/license.txt # buildkit

ARG ENABLE_MITMPROXY=0

RUN |1 ENABLE_MITMPROXY=0 /bin/sh -c /tmp/manage_cert.sh install # buildkit

RUN |1 ENABLE_MITMPROXY=0 /bin/sh -c export DEBIAN_FRONTEND=noninteractive && apt-get update && apt-get install -y --no-install-recommends apt-utils build-essential libncurses6 libncursesw6 unzip jq gnupg libtcmalloc-minimal4 git libglib2.0-0 libhwloc15 libnuma-dev libnuma1 libpmi2-0-dev numactl openssh-client vim-tiny wget && rm -rf /var/lib/apt/lists/* # buildkit

ARG CUDA_COMPONENT_LIST=cccl crt nvrtc driver-dev culibos-dev cudart cudart-dev nvcc tileiras

ENV CUDA_COMPONENT_LIST=cccl crt nvrtc driver-dev culibos-dev cudart cudart-dev nvcc tileiras

RUN |2 ENABLE_MITMPROXY=0 CUDA_COMPONENT_LIST=cccl crt nvrtc driver-dev culibos-dev cudart cudart-dev nvcc tileiras /bin/sh -c set -exo pipefail if [ ! -f /etc/ld.so.conf.d/nvidia-tegra.conf ]; then COMPONENT_LIST="${CUDA_COMPONENT_LIST}" /nvidia/build-scripts/installCUDA.sh else # Don't use COMPONENT_LIST in iGPU environment /nvidia/build-scripts/installCUDA.sh fi # buildkit

ARG VERSION_LIST=CURAND_VERSION CUBLAS_VERSION CUFILE_VERSION

RUN |3 ENABLE_MITMPROXY=0 CUDA_COMPONENT_LIST=cccl crt nvrtc driver-dev culibos-dev cudart cudart-dev nvcc tileiras VERSION_LIST=CURAND_VERSION CUBLAS_VERSION CUFILE_VERSION /bin/sh -c set -exo pipefail VERSION_LIST="${VERSION_LIST}" /nvidia/build-scripts/installLIBS.sh # Hack to grab stubs from the libs stage and put them under /usr/local/cuda/lib64 without actually installing the libs HACK_LIST="cusparse cusolver cufft nvJitLink" for lib in ${HACK_LIST}; do _LIB_VERSION_ENV_VAR="${lib^^}_VERSION" _LIB_MAJOR_VERSION=$(echo "${!_LIB_VERSION_ENV_VAR}" | cut -d. -f1) if [[ -z "${_LIB_MAJOR_VERSION}" ]]; then echo "[ERROR] ${_LIB_VERSION_ENV_VAR} is not set" exit 1 fi for _stub in $(ls /tmp/stubs/*${lib}*); do cp ${_stub} /usr/local/cuda/lib64/$(basename ${_stub}).${_LIB_MAJOR_VERSION} # Create a symlink to the file with .0 extension ln -sf $(basename ${_stub}).${_LIB_MAJOR_VERSION} /usr/local/cuda/lib64/$(basename ${_stub}).0 done done ( set +x; echo "[INFO] /usr/local/cuda/lib64 after hack:"; ls -larth /usr/local/cuda/lib64/ ) /nvidia/build-scripts/installCUDNN.sh /nvidia/build-scripts/installCUSPARSELT.sh NOBUILDER=1 /nvidia/build-scripts/installTRT.sh if [ ! -f /etc/ld.so.conf.d/nvidia-tegra.conf ]; then /nvidia/build-scripts/installNCCL.sh /nvidia/build-scripts/installNVSHMEM.sh # Manually create symlinks nvshmem so we don't need to use dpkg-divert # nvshmem default installation is under /usr/lib/x86_64-linux-gnu/nvshmem/${CUDA_VERSION_MAJOR} # We need to libs to be available under /usr/local/cuda/lib64 export CUDA_VERSION_MAJOR=$(echo "${CUDA_VERSION}" | cut -d. -f1) find /usr/lib/*-linux-gnu/nvshmem/${CUDA_VERSION_MAJOR}/ -maxdepth 1 -type f -exec ln -sf {} /usr/local/cuda/lib64/ \; # For symlinks under /usr/lib/*-linux-gnu/nvshmem/${CUDA_VERSION_MAJOR}/, we want to create it under /usr/local/cuda/lib64/ and link it to the same target find /usr/lib/*-linux-gnu/nvshmem/${CUDA_VERSION_MAJOR}/ -maxdepth 1 -type l -exec ln -sf {} /usr/local/cuda/lib64/ \; # Ensure libnvshmem_host.so* files exist under /usr/local/cuda/lib64 ls /usr/local/cuda/lib64/libnvshmem_host.so* || { echo "[ERROR] libnvshmem_host.so* not found under /usr/local/cuda/lib64"; exit 1; } fi # buildkit

ENV LIBRARY_PATH=/usr/local/cuda/lib64/stubs:

LABEL com.nvidia.nccl.version=2.29.stable.20260109 com.nvidia.cublas.version=13.2.1.1 com.nvidia.cufft.version=12.1.0.78 com.nvidia.curand.version=10.4.1.81 com.nvidia.cusparse.version=12.7.3.1 com.nvidia.cusparselt.version=0.8.1.1 com.nvidia.cusolver.version=12.0.9.81 com.nvidia.npp.version=13.0.3.3 com.nvidia.nvjpeg.version=13.0.3.75 com.nvidia.cudnn.version=9.19.0.56

ARG ENABLE_MITMPROXY=0

RUN |1 ENABLE_MITMPROXY=0 /bin/sh -c /tmp/manage_cert.sh install # buildkit

ARG VERSION_LIST=CURAND_VERSION CUBLAS_VERSION CUFILE_VERSION

RUN |2 ENABLE_MITMPROXY=0 VERSION_LIST=CURAND_VERSION CUBLAS_VERSION CUFILE_VERSION /bin/sh -c set -exo pipefail export DEVEL=1 BASE=0 if [ ! -f /etc/ld.so.conf.d/nvidia-tegra.conf ]; then COMPONENT_LIST="nvrtc-dev" /nvidia/build-scripts/installCUDA.sh else # Don't use COMPONENT_LIST in iGPU environment /nvidia/build-scripts/installCUDA.sh fi VERSION_LIST="${VERSION_LIST}" /nvidia/build-scripts/installLIBS.sh if [ ! -f /etc/ld.so.conf.d/nvidia-tegra.conf ]; then /nvidia/build-scripts/installNCCL.sh fi NOBUILDER=1 /nvidia/build-scripts/installTRT.sh /nvidia/build-scripts/installCUDNN.sh /nvidia/build-scripts/installCUSPARSELT.sh # Clean up libcudnn_engines_precompiled rm -rf /usr/lib/*-linux-gnu/libcudnn_engines_precompiled.so* # buildkit

RUN |2 ENABLE_MITMPROXY=0 VERSION_LIST=CURAND_VERSION CUBLAS_VERSION CUFILE_VERSION /bin/sh -c /tmp/manage_cert.sh uninstall # buildkit

COPY xformers/ /opt/xformers/ # buildkit

COPY vllm-flash-attn/ /opt/vllm/vllm-flash-attn/ # buildkit

COPY vllm-src/ /opt/vllm/vllm-src/ # buildkit

COPY /etc/pip/original_constraint.txt /etc/pip/constraint.txt # buildkit

COPY /usr/local/cuda/lib64/libnvshmem_device.bc /usr/local/cuda/lib64/libnvshmem_device.bc # buildkit

ARG TARGETARCH=arm64

ARG PYVER=3.12

ARG VLLM_VERSION=0.15.1+befbc472

ENV VLLM_VERSION=0.15.1+befbc472

ENV LD_LIBRARY_PATH=/usr/local/lib/python3.12/dist-packages/torch/lib:/usr/local/lib/python3.12/dist-packages/torch_tensorrt/lib:/usr/local/cuda/compat/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64

ENV PIP_CONSTRAINT=/etc/pip/constraint.txt

ENV CUDA_HOME=/usr/local/cuda

ENV CPATH=/usr/local/cuda/include:

ENV PATH=/usr/local/lib/python3.12/dist-packages/torch_tensorrt/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/mpi/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/local/ucx/bin:/opt/amazon/efa/bin

ENV TORCH_CUDA_ARCH_LIST=8.0 8.6 9.0 10.0 11.0 12.0+PTX

ENV TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas

ENV TRITON_CUOBJDUMP_PATH=/usr/local/cuda/bin/cuobjdump

ENV TRITON_NVDISASM_PATH=/usr/local/cuda/bin/nvdisasm

ENV TRITON_CUDACRT_PATH=/usr/local/cuda/include

ENV TRITON_CUDART_PATH=/usr/local/cuda/include

ENV TRITON_CUPTI_PATH=/usr/local/cuda/include

ENV MAX_JOBS=12

ENV PIP_BREAK_SYSTEM_PACKAGES=1

ENV PIP_NO_BUILD_ISOLATION=1

ENV PYTORCH_TRITON_VERSION=3.6.0+git9844da95.nv26.2

RUN |3 TARGETARCH=arm64 PYVER=3.12 VLLM_VERSION=0.15.1+befbc472 /bin/sh -c export PYSFX=`echo "${PYVER}" | cut -c1-1` && export DEBIAN_FRONTEND=noninteractive && apt-get update && apt-get install -y --no-install-recommends python$PYVER-dev python$PYSFX python$PYSFX-dev python$PYSFX-venv python-is-python$PYSFX patchelf libxcb1-dev lld && rm -rf /var/lib/apt/lists/* && curl -O https://bootstrap.pypa.io/get-pip.py && python get-pip.py && rm get-pip.py # buildkit

RUN |3 TARGETARCH=arm64 PYVER=3.12 VLLM_VERSION=0.15.1+befbc472 /bin/sh -c curl -Of "https://gitlab-master.nvidia.com/api/v4/projects/9703/packages/generic/pytorch_triton/wheel/pytorch_triton-${PYTORCH_TRITON_VERSION}-cp312-cp312-linux_$(uname -m).whl" && mkdir -p /tmp/dist && mv *.whl /tmp/dist && pip install --no-cache-dir /tmp/dist/*.whl && rm -rf /tmp/dist && cd /usr/local/lib/python${PYVER}/dist-packages/ && ls -la | grep pytorch_triton && mv pytorch_triton-${PYTORCH_TRITON_VERSION}.dist-info triton-${PYTORCH_TRITON_VERSION}.dist-info && cd triton-${PYTORCH_TRITON_VERSION}.dist-info && echo "Current directory: $(pwd)" && echo "Files in directory:" && ls -la && sed -i 's/^Name: pytorch-triton/Name: triton/' METADATA && sed -i "s|pytorch_triton-${PYTORCH_TRITON_VERSION}.dist-info/|triton-${PYTORCH_TRITON_VERSION}.dist-info/|g" RECORD && echo "METADATA after update:" && grep "^Name:" METADATA # buildkit

RUN |3 TARGETARCH=arm64 PYVER=3.12 VLLM_VERSION=0.15.1+befbc472 /bin/sh -c pip install /opt/transfer/*.whl # buildkit

RUN |3 TARGETARCH=arm64 PYVER=3.12 VLLM_VERSION=0.15.1+befbc472 /bin/sh -c pip install "apache-tvm-ffi<=0.1.7" # buildkit

RUN |3 TARGETARCH=arm64 PYVER=3.12 VLLM_VERSION=0.15.1+befbc472 /bin/sh -c mkdir -p /root/.cache/tiktoken_cache && cd /root/.cache/tiktoken_cache && wget -q https://openaipublic.blob.core.windows.net/encodings/o200k_base.tiktoken && mv o200k_base.tiktoken fb374d419588a4632f3f557e76b4b70aebbca790 # buildkit

ENV TIKTOKEN_CACHE_DIR=/root/.cache/tiktoken_cache TIKTOKEN_RS_CACHE_DIR=/root/.cache/tiktoken_cache

RUN |3 TARGETARCH=arm64 PYVER=3.12 VLLM_VERSION=0.15.1+befbc472 /bin/sh -c set -xeo pipefail pip --version python -c 'import sys; print(sys.platform)' # Remove PyTorch's numba and llvmlite constraints due to vllm version conflict sed -i -e 's/numba==.*//' /etc/pip/constraint.txt sed -i -e 's/llvmlite==.*//' /etc/pip/constraint.txt echo "[PIP_CONSTRAINT(${PIP_CONSTRAINT})]" cat ${PIP_CONSTRAINT} # L0_correctness_TinyLlama--Thor_AGX if [ $TARGETARCH = "arm64" ]; then cd /opt curl "https://gitlab-master.nvidia.com/api/v4/projects/105799/packages/generic/nvpl_slim_24.04/sbsa/nvpl_slim_24.04.tar" --output nvpl_slim_24.04.tar tar -xf nvpl_slim_24.04.tar cp -r nvpl_slim_24.04/lib/* /usr/local/lib cp -r nvpl_slim_24.04/include/* /usr/local/include rm -rf nvpl_slim_24.04.tar nvpl_slim_24.04 fi # L0_correctness_TinyLlama--H100 if [ $TARGETARCH = "amd64" ]; then pip install --no-cache-dir mkl mkl-include mkl-devel # Clean up duplicate libraries from mkl/tbb python wheel install which makes copies for symlinks. # This can be removed once upgrading to an MKL pip wheel that fixes the symlinks find /usr/local/lib -maxdepth 1 -type f -regex '.*\/lib$tbb\|mkl$.*\.so$$\|\.[0-9]*\.[0-9]*$' -exec rm -v {} + patchelf --set-rpath '/usr/local/lib' /usr/local/lib/python${PYVER}/dist-packages/torch/lib/libtorch_global_deps.so find / -name "*libmkl_intel_lp64*" fi find / -name "cuda.h" ldconfig sed -i -E 's/^([[:space:]]*)from \.modeling_utils import PreTrainedAudioTokenizerBase/\1pass/' /usr/local/lib/python${PYVER}/dist-packages/transformers/processing_utils.py # Install build dependencies to enable rebuilding inside the container. # Note: This installs a newer version of cmake than "apt install cmake" does, # which we require to support CMAKE_LINKER_TYPE=LLD. pip install --no-cache-dir uv setuptools-scm cmake change_wheel_version # Install cuda devel libs required for compilation. BASE=1 DEVEL=1 VERSION_LIST="CUSPARSE_VERSION CUSOLVER_VERSION" PURGESTUBS=1 /nvidia/build-scripts/installLIBS.sh # WAR cccl include dir change to enable re-building sglang from source inside the container. ln -s /usr/local/cuda/include/cccl/cuda /usr/local/cuda/include/cuda # buildkit

ENV VLLM_FLASH_ATTN_SRC_DIR=/opt/vllm/vllm-flash-attn/

WORKDIR /workspace

ENV NVIDIA_PRODUCT_NAME=vLLM

LABEL com.nvidia.vllm.version=0.15.1+befbc472

ARG NVIDIA_VLLM_VERSION=26.02

ENV NVIDIA_VLLM_VERSION=26.02

ARG NVIDIA_BUILD_ID=270699521

ENV NVIDIA_BUILD_ID=270699521

LABEL com.nvidia.build.id=270699521

ARG NVIDIA_BUILD_REF=db7600950a476c031b123f25536f8f1d7bcb992b

LABEL com.nvidia.build.ref=db7600950a476c031b123f25536f8f1d7bcb992b

ARG VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15

ARG SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f

ENV DEBIAN_FRONTEND=noninteractive CUDA_HOME=/usr/local/cuda CUTE_DSL_ARCH=sm_121a PYTHONPATH=/opt/vllm:/opt/sparkinfer PATH=/opt/runtime-venv/bin:/usr/local/cuda/bin:/usr/bin:/bin KV_FP8_ROPE=0 VLLM_DSV4_PADDED_NVFP4=1

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c test "$(uname -m)" = aarch64 || (echo "This image targets one DGX Spark (Linux aarch64 + GB10/SM121)." >&2; exit 2) # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c rm -rf /opt/vllm /opt/sparkinfer && git init /opt/vllm && git -C /opt/vllm remote add origin https://github.com/local-inference-lab/vllm.git && git -C /opt/vllm fetch --depth=1 origin "${VLLM_COMMIT}" && git -C /opt/vllm checkout --detach FETCH_HEAD && git init /opt/sparkinfer && git -C /opt/sparkinfer remote add origin https://github.com/local-inference-lab/sparkinfer.git && git -C /opt/sparkinfer fetch --depth=1 origin "${SPARKINFER_COMMIT}" && git -C /opt/sparkinfer checkout --detach FETCH_HEAD # buildkit

COPY patches/vllm.patch /tmp/vllm.patch # buildkit

COPY patches/vllm-padded-fp8-compat.patch /tmp/vllm-padded-fp8-compat.patch # buildkit

COPY patches/vllm-dspark-compact-draft.patch /tmp/vllm-dspark-compact-draft.patch # buildkit

COPY patches/sparkinfer.patch /tmp/sparkinfer.patch # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c git -C /opt/vllm apply --check /tmp/vllm.patch && git -C /opt/vllm apply /tmp/vllm.patch && git -C /opt/vllm apply --check /tmp/vllm-padded-fp8-compat.patch && git -C /opt/vllm apply /tmp/vllm-padded-fp8-compat.patch && git -C /opt/vllm apply --check /tmp/vllm-dspark-compact-draft.patch && git -C /opt/vllm apply /tmp/vllm-dspark-compact-draft.patch && git -C /opt/sparkinfer apply --check /tmp/sparkinfer.patch && git -C /opt/sparkinfer apply /tmp/sparkinfer.patch # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c python3 -m venv --system-site-packages /opt/runtime-venv && /opt/runtime-venv/bin/python -m pip install --upgrade pip # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c /opt/runtime-venv/bin/python -m pip install --index-url https://download.pytorch.org/whl/cu130 'torch==2.12.0+cu130' 'torchvision==0.27.0+cu130' # buildkit

COPY patches/vllm-build-fetch.patch /tmp/vllm-build-fetch.patch # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c export PATH=/usr/local/bin:${PATH} && export TORCH_CUDA_ARCH_LIST=12.0 && cmake -S /opt/vllm -B /opt/vllm-build -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_CUDA_ARCHITECTURES=120 -DVLLM_TARGET_DEVICE=cuda -DVLLM_PYTHON_EXECUTABLE=/opt/runtime-venv/bin/python -DVLLM_ENABLE_SCALED_MM_C2X=OFF -DVLLM_ENABLE_SM12X_NVFP4=OFF -DVLLM_ENABLE_SM12X_NVFP4_KV=ON -DVLLM_BUILD_RUNTIME_EXTERNALS=OFF -DVLLM_BUILD_QUTLASS=OFF -DVLLM_BUILD_VLLM_FLASH_ATTN=OFF && cmake --build /opt/vllm-build --target _C_stable_libtorch --parallel 4 && cp /opt/vllm-build/_C_stable_libtorch.abi3.so /opt/vllm/vllm/_C_stable_libtorch.abi3.so # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c git init /tmp/vllm-flash-attn && git -C /tmp/vllm-flash-attn remote add origin https://github.com/vllm-project/flash-attention.git && git -C /tmp/vllm-flash-attn fetch --depth=1 origin caaa4eb59845388a20b1f435ecaafb4bd9517ad8 && git -C /tmp/vllm-flash-attn checkout --detach FETCH_HEAD && mkdir -p /opt/vllm/vllm/vllm_flash_attn/cute && cp -a /tmp/vllm-flash-attn/flash_attn/cute/. /opt/vllm/vllm/vllm_flash_attn/cute/ && find /opt/vllm/vllm/vllm_flash_attn/cute -type f -name '*.py' -exec sed -i 's/flash_attn\.cute/vllm.vllm_flash_attn.cute/g' {} + && rm -rf /tmp/vllm-flash-attn # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c env -u PIP_CONSTRAINT /opt/runtime-venv/bin/python -m pip install 'nvidia-cutlass-dsl-libs-cu13==4.6.0' 'transformers==5.13.1' 'mistral-common==1.11.5' 'instanttensor==0.1.5' 'openai==2.44.0' 'compressed-tensors==0.17.0' 'apache-tvm-ffi==0.1.10' 'torch-c-dlpack-ext==0.1.5' 'z3-solver==4.15.4' 'tilelang==0.1.9' && env -u PIP_CONSTRAINT /opt/runtime-venv/bin/python -m pip install --no-deps 'quack-kernels==0.6.2' # buildkit

ENV PATH=/opt/runtime-venv/bin:/usr/local/cuda/bin:/usr/local/bin:/usr/bin:/bin

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c case "$(ninja --version)" in 1.13.0*) ;; *) exit 1 ;; esac # buildkit

COPY patches/instanttensor.patch /tmp/instanttensor.patch # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c patch --dry-run -p1 -d /opt/runtime-venv/lib/python3.12/site-packages < /tmp/instanttensor.patch && patch -p1 -d /opt/runtime-venv/lib/python3.12/site-packages < /tmp/instanttensor.patch # buildkit

COPY scripts /opt/recipe/scripts # buildkit

COPY config /opt/recipe/config # buildkit

RUN |2 VLLM_COMMIT=30038602b71395f481ef4a6edfe4fcf8551d9c15 SPARKINFER_COMMIT=272a84bd97ce791a1e92d1f3a0da3dd5f3c6565f /bin/sh -c chmod +x /opt/recipe/scripts/*.sh # buildkit

EXPOSE map[8000/tcp:{}]

HEALTHCHECK &{["CMD-SHELL" "/opt/recipe/scripts/healthcheck.sh"] "30s" "5s" "20m0s" "0s" '\x03'}

ENTRYPOINT ["/opt/recipe/scripts/entrypoint.sh"]

USER root

LABEL org.opencontainers.image.title=DeepSeek V4 Flash Vision Kalibrated on GB10 org.opencontainers.image.source=https://git.fml128.ch/FML128/llm-containers ch.fml128.deepseek.recipe-revision=aaff897c88d840574d3bc1a9e74766672f1d995f ch.fml128.deepseek.model-revision=178652ce2ef562c9135488def954a2418db6f5e5

ADD https://codeload.github.com/GaelicThunder/DeepSeek-V4-Flash-Vision-One-DGX-Spark/tar.gz/aaff897c88d840574d3bc1a9e74766672f1d995f /tmp/recipe.tar.gz # buildkit

COPY deepseek-v4-flash-vision/install-recipe.py deepseek-v4-flash-vision/requirements.lock /tmp/build/ # buildkit

RUN /bin/sh -c /opt/runtime-venv/bin/python /tmp/build/install-recipe.py /tmp/recipe.tar.gz && /opt/runtime-venv/bin/pip install --no-cache-dir --no-deps --require-hashes --only-binary=:all: -r /tmp/build/requirements.lock && /opt/runtime-venv/bin/python -c 'import importlib.metadata as m, pathlib, re; import xgrammar; assert hasattr(xgrammar, "normalize_tool_choice"); pins = re.findall(r"^([\w-]+)==(\S+)", pathlib.Path("/tmp/build/requirements.lock").read_text(), re.M); assert pins and all(m.version(name) == version for name, version in pins)' && /opt/runtime-venv/bin/python -c 'import json; from vllm.entrypoints.openai.chat_completion.protocol import ChatMessage; from vllm.entrypoints.openai.engine.protocol import DeltaMessage; messages = [ChatMessage(role="assistant", reasoning="compatibility"), DeltaMessage(reasoning="compatibility")]; assert all(json.loads(m.model_dump_json(exclude_unset=True))["reasoning_content"] == "compatibility" for m in messages); assert "reasoning_content" not in DeltaMessage(content="text").model_dump(exclude_unset=True)' && rm -rf /tmp/build /tmp/recipe.tar.gz # buildkit

COPY deepseek-v4-flash-vision/model-cache.py deepseek-v4-flash-vision/main.py deepseek-v4-flash-vision/models.lock.json /opt/deepseek/ # buildkit

RUN /bin/sh -c mkdir -p /models /cache /hf-cache && chown 1000:1000 /models /cache /hf-cache && chmod -R a+rX /opt/deepseek # buildkit

ENV PYTHONUNBUFFERED=1 PYTHONDONTWRITEBYTECODE=1 HOME=/tmp HF_HOME=/hf-cache HF_HUB_DISABLE_TELEMETRY=1 HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 HOST=0.0.0.0 PORT=8000 SERVED_MODEL_NAME=deepseek-v4-flash-vision-exp MAX_MODEL_LEN=245760 MAX_NUM_SEQS=1 MAX_NUM_BATCHED_TOKENS=4096 GPU_MEMORY_UTILIZATION=0.925 MODE=dspark DSPARK_TOKENS=5 DSPARK_CAPACITY=0 DSPARK_DYNAMIC_DRAFT_DEPTH=0 DSPARK_DYNAMIC_DRAFT_DEPTH_WINDOW=8 DSPARK_DRAFT_EXPERTS=64 DSPARK_STRUCTURED_EXPERTS_PER_CATEGORY=32 REJECTION_SAMPLE_METHOD=standard DRAFT_SAMPLE_METHOD=probabilistic VLLM_DSV4_PADDED_NVFP4=0 KV_FP8_ROPE=0 VLLM_USE_B12X_WO_PROJECTION=1 MAX_CUDAGRAPH_CAPTURE_SIZE=6 CUDAGRAPH_CAPTURE_SIZES=6 VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0 VLLM_MULTI_STREAM_GEMM_TOKEN_THRESHOLD=4096 DEFAULT_CHAT_TEMPLATE_KWARGS_THINKING=true DEFAULT_CHAT_TEMPLATE_KWARGS_EFFORT=max KV_OFFLOAD_GB=0 VLLM_USE_SIMPLE_KV_OFFLOAD=0 VERIFY_MODEL_CHECKSUMS=0 EXTRA_VLLM_ARGS=--long-prefill-token-threshold 4096 --mm-processor-cache-gb 0 --mm-encoder-attn-backend TORCH_SDPA

WORKDIR /opt/deepseek

USER 1000:1000

EXPOSE [8000/tcp]

STOPSIGNAL SIGTERM

ENTRYPOINT ["/opt/runtime-venv/bin/python" "/opt/deepseek/main.py"]

CMD ["serve"]

Key	Value
ch.fml128.deepseek.model-revision	178652ce2ef562c9135488def954a2418db6f5e5
ch.fml128.deepseek.recipe-revision	aaff897c88d840574d3bc1a9e74766672f1d995f
com.nvidia.build.id	270699521
com.nvidia.build.ref	db7600950a476c031b123f25536f8f1d7bcb992b
com.nvidia.cublas.version	13.2.1.1
com.nvidia.cuda.version	9.0
com.nvidia.cudnn.version	9.19.0.56
com.nvidia.cufft.version	12.1.0.78
com.nvidia.curand.version	10.4.1.81
com.nvidia.cusolver.version	12.0.9.81
com.nvidia.cusparse.version	12.7.3.1
com.nvidia.cusparselt.version	0.8.1.1
com.nvidia.nccl.version	2.29.stable.20260109
com.nvidia.npp.version	13.0.3.3
com.nvidia.nvjpeg.version	13.0.3.75
com.nvidia.vllm.version	0.15.1+befbc472
com.nvidia.volumes.needed	nvidia_driver
org.opencontainers.image.ref.name	ubuntu
org.opencontainers.image.source	https://git.fml128.ch/FML128/llm-containers
org.opencontainers.image.title	DeepSeek V4 Flash Vision Kalibrated on GB10
org.opencontainers.image.version	24.04