Namespace
gpustack
Image / Tagrunner:cuda12.9-vllm0.31.0-linux-arm64
Content Digestsha256:64981b70c82b53b31812da60e6572c20819a6b16e22d6ea7cfc97d5b3bf502fa
Details
2026-10-16 08:05:04 UTC
12.1 GB
Labels
-
ai.vllm.build.commitdb9527a46873454610df6dbedf79a36d6bf1a7f6
-
ai.vllm.build.pipeline019d130e-464e-4ff7-b84b-492992c0c06b
-
ai.vllm.build.urlhttps://buildkite.com/vllm/release-v2/builds/7746
-
ai.vllm.image.tagvllm/vllm-openai:v0.31.0-cu129-ubuntu2404
-
maintainerNVIDIA CORPORATION <cudatools@nvidia.com>
-
org.opencontainers.image.ref.nameubuntu
-
org.opencontainers.image.revisiondb9527a46873454610df6dbedf79a36d6bf1a7f6
-
org.opencontainers.image.sourcehttps://github.com/vllm-project/vllm
-
org.opencontainers.image.urlhttps://buildkite.com/vllm/release-v2/builds/7746
-
org.opencontainers.image.versionvllm/vllm-openai:v0.31.0-cu129-ubuntu2404
Environment
CRIU_CUDA_PLUGIN_DIR/usr/local/lib/criu
CUDA_ARCHS CUDA_HOME/usr/local/cuda
CUDA_VERSION12.9.1
DEBIAN_FRONTENDnoninteractive
LANGen_US.UTF-8
LANGUAGEen_US:en
LC_ALLen_US.UTF-8
LD_LIBRARY_PATH/usr/local/nvidia/lib64:/usr/local/cuda/lib64:/usr/local/cuda/lib64:/usr/local/cuda/lib64
NVARCHsbsa
NVIDIA_DRIVER_CAPABILITIEScompute,utility
NVIDIA_REQUIRE_CUDAcuda>=12.9 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>=560,driver<561 brand=grid,driver>=560,driver<561 brand=tesla,driver>=560,driver<561 brand=nvidia,driver>=560,driver<561 brand=quadro,driver>=560,driver<561 brand=quadrortx,driver>=560,driver<561 brand=nvidiartx,driver>=560,driver<561 brand=vapps,driver>=560,driver<561 brand=vpc,driver>=560,driver<561 brand=vcs,driver>=560,driver<561 brand=vws,driver>=560,driver<561 brand=cloudgaming,driver>=560,driver<561 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
NVIDIA_VISIBLE_DEVICESall
NV_CUDA_CUDART_VERSION12.9.79-1
PATH/usr/local/cuda/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
PIP_DISABLE_PIP_VERSION_CHECK1
PIP_NO_CACHE_DIR1
PIP_ROOT_USER_ACTIONignore
PYTHON_VERSION3.12
RAY_EXPERIMENTAL_NOSET_CUDA_VISIBLE_DEVICES1
RUNAI_STREAMER_LOG_LEVELINFO
RUNAI_STREAMER_LOG_TO_STDERR1
RUNAI_STREAMER_MEMORY_LIMIT0
SAFETENSORS_FAST_GPU1
TORCH_CUDA_ARCH_LIST8.0 8.7 8.9 9.0 10.0 12.0
UV_CACHE_DIR/opt/uv/cache
UV_HTTP_TIMEOUT500
UV_INDEX_STRATEGYunsafe-best-match
UV_LINK_MODEcopy
UV_NO_CACHE1
UV_OVERRIDE/etc/uv-overrides.txt
UV_PRERELEASEallow
UV_PYTHON_INSTALL_DIR/opt/uv/python
UV_SKIP_WHEEL_FILENAME_CHECK1
UV_SYSTEM_PYTHON1
VLLM_BUILD_COMMITdb9527a46873454610df6dbedf79a36d6bf1a7f6
VLLM_BUILD_PIPELINE019d130e-464e-4ff7-b84b-492992c0c06b
VLLM_BUILD_URLhttps://buildkite.com/vllm/release-v2/builds/7746
VLLM_ENABLE_CUDA_COMPATIBILITY0
VLLM_IMAGE_TAGvllm/vllm-openai:v0.31.0-cu129-ubuntu2404
VLLM_LMCACHE_VERSION0.5.6
VLLM_TORCH_CUDA_VERSION12.9.1
VLLM_TORCH_VERSION2.13.0
VLLM_USAGE_SOURCEproduction-docker-image
VLLM_VERSION0.31.0
Layers
[#000] sha256:1fe4490d3380a3fec7f0b5cfd536dbdeed91ff50c5fbd11eba4ec838d689855e - 0.23% (28.6 MB)
[#001] sha256:ecbca12713d0ee06710d449b9cecb86e2b3be1c32e2f7e6fd6ff3dea9219cf03 - 0.03% (4.32 MB)
[#002] sha256:ed5ca6b9e891c487cf1c2321e0b3e28f63b7ffdc2ee7c75592578bfcdebac740 - 0.75% (93.1 MB)
[#003] sha256:3f0047ea99ac7fb8902bd5e3aeb361d80ffa04c40b26b7d3935b6709c3c66f0c - 0.0% (182 Bytes)
[#004] sha256:6e8af4fd0a071982e528b634ba99dec2474c21147f99748be708f36e10e3f4c2 - 0.0% (6.72 KB)
[#005] sha256:f3dec6b9e2d3d50fff968b028571bc921fe0fd1f27d2eae0e4b6d0f71bc787e9 - 0.0% (104 Bytes)
[#006] sha256:ac5250bcfb0edaabc2b9b6186788d608482ce25e840e8f434cb25db97ef43c68 - 0.0% (230 Bytes)
[#007] sha256:85365c777e733872752494efe7436c1326101bcbd8e656e8a665614e257d076b - 1.75% (216 MB)
[#008] sha256:a75624bffa7ede2fede13c281c15e5ad4784402c12846025189aa2aa0afcc582 - 12.9% (1.56 GB)
[#009] sha256:4f4fb700ef54461cfa02571ae0db9a0dc1e0cdb5577484a6d75e68dc38e8acc1 - 0.0% (32 Bytes)
[#010] sha256:0220852be227ca2a11fb22a9e8b916259d29e88f073d71599211c853b7c27bda - 0.0% (1.62 KB)
[#011] sha256:4f4fb700ef54461cfa02571ae0db9a0dc1e0cdb5577484a6d75e68dc38e8acc1 - 0.0% (32 Bytes)
[#012] sha256:72f14c846903c946ef92e14bea47f7e9f1063af83c0b9f242fa3b7c921c82a6d - 0.31% (38.6 MB)
[#013] sha256:fbc5b91e1e282c5f740c227fbd597ca1b382e3d79eaa8469323f9a49ca0fb9c8 - 0.0% (150 Bytes)
[#014] sha256:86c55cc89799201fd36b69077dd02ac6947420d8197adf5132bf5511d9b98dae - 0.0% (162 Bytes)
[#015] sha256:1f153c2a513f127e28e12efa84175f7fa86db013598bc70b31b50d1c2cb73b2c - 0.0% (3.61 KB)
[#016] sha256:0f8bf2f640c309612351b607988805ce2297de98557e95d61d89ed79ca35a6bc - 0.0% (2.03 KB)
[#017] sha256:0f8bf2f640c309612351b607988805ce2297de98557e95d61d89ed79ca35a6bc - 0.0% (2.03 KB)
[#018] sha256:cb1f9a28cc886fdab29548d7994dcc3ab7b8ea5bd316bf808e754f13a5f1fa0d - 0.0% (1.67 KB)
[#019] sha256:06f5e4c9ac6b180426a3c7534d46fb5590020acdbc2df4b4fe66c451d27659c9 - 0.0% (870 Bytes)
[#020] sha256:08901e70b50fcaf761613bac4c9522d11901ee53559afcc15109935713fd2669 - 47.21% (5.7 GB)
[#021] sha256:b7a84140fc9a454cd8398d8c8b3b8ad36c4997877f8c20f194c039cf6139796e - 9.41% (1.14 GB)
[#022] sha256:5a57eb1fd3b58bb5f4357655aec213f13f3a5b8d69ab3286ee4e354462a07e9d - 0.0% (896 Bytes)
[#023] sha256:085fa2817190048a167b344a191e684d8baf49c225d50de29a562f1cb2d4ec2c - 0.0% (517 KB)
[#024] sha256:bcb62395a84e5d271aeba370501d0f9fb20f78c0cb7c8bef95f61cca912860f8 - 2.32% (287 MB)
[#025] sha256:36f55ba0f84d83956f22faff67b836cf0e8e5b618c4bc5eb945138e2bf1cbd1a - 0.0% (202 Bytes)
[#026] sha256:fa33212f033d1f1e678d5fcb72f9cecea33e1da4598d005a7a0bf02c709f703b - 0.0% (229 Bytes)
[#027] sha256:cf4a97a791a07e8324d8f72ae3a07492269698fb37f59afe53e94751d2e1c91d - 5.87% (726 MB)
[#028] sha256:2ba84772839e1b2e67943d51fb03abc21cf5af5f40698e3a3f64710922cf12d5 - 0.0% (5.78 KB)
[#029] sha256:4f4fb700ef54461cfa02571ae0db9a0dc1e0cdb5577484a6d75e68dc38e8acc1 - 0.0% (32 Bytes)
[#030] sha256:14a737de76c586723b70297ec3d21cec42ed78d5f76f25007a5bc63d8b31dc98 - 0.0% (253 Bytes)
[#031] sha256:48efe759e6ac97174467d0a68e793aa868ab476e5600a7011996844399f2fa4d - 0.15% (18 MB)
[#032] sha256:d861c21de64eba7e7fd107a5a05c30e1ff9e0f6d5bf667753acd5dd3ff2df887 - 0.0% (298 KB)
[#033] sha256:a19e26a73373180cdd1197ed9bae6fc214f970854a44983f715b9318118297eb - 0.0% (297 KB)
[#034] sha256:e3b323d688c72a27d506010fff19efe65ca67fcfbfb17c2e4c949c599b9caeec - 0.0% (9.61 KB)
[#035] sha256:28d3c955e1f749acc15380c93913a24d89b8128f3446a100732aaaa4c7b70646 - 10.36% (1.25 GB)
[#036] sha256:3480a5342b22dab3446e3e9e1ccfef3363424c2a04aaa18b7ab9a00d77e47b0e - 0.0% (92 Bytes)
[#037] sha256:fed2d83108fb12742195b5e951933adcbb886468cf3231c0ca9b819c4148391f - 0.0% (212 Bytes)
[#038] sha256:e0f09b0aea9323c22eb95b9f0e403007569b440ba660275161f0231147bf06d4 - 0.0% (823 Bytes)
[#039] sha256:9ca7e30b4a004d3a06a6bca3d329176e30e2f5c60df7db1a936b6e41f9538fab - 0.0% (183 Bytes)
[#040] sha256:a7e0ca78228a0bc45a491572af8c23a2e914cd1a67470b304ca00a9343188225 - 0.77% (95.4 MB)
[#041] sha256:442509f7505dd33098a80748b230df09e57fe1b90953174b765f74ddecb585f9 - 0.0% (1.9 KB)
[#042] sha256:bfc81204ed752087c3efcb5d276be16b59da5c3a69e50fb7ba7d24ae1b12c4c8 - 0.25% (30.9 MB)
[#043] sha256:020e1a5ff4d3f5016162073de9cc70f00274be53492682ff93338b6e568d0b02 - 0.07% (8.1 MB)
[#044] sha256:9296c5cffa137734af0bd547dc07e42e11642753c8c6e7524e5339fa824437bc - 0.63% (77.8 MB)
[#045] sha256:9bad1dce389643a922a8ff6027df0255a6258d2c20f02cac608604ce8b5b6c88 - 3.61% (446 MB)
[#046] sha256:71c649801656413a2e7d516ed78216303fe429cfffe7615d91dc296fde1a6f17 - 2.04% (253 MB)
[#047] sha256:d3d5f194d9f8c047b0effa6d55cfcd67e37acf27ed021092dd0fd4d577354bf2 - 1.13% (140 MB)
[#048] sha256:4f4fb700ef54461cfa02571ae0db9a0dc1e0cdb5577484a6d75e68dc38e8acc1 - 0.0% (32 Bytes)
[#049] sha256:28e550521d6d9ca0dfe31b7f9b6ac047e755d1848fad8f4a5f4944769d218fe8 - 0.0% (71.6 KB)
[#050] sha256:b6c057647eff25a5a0b5d1befe78d38b24993cc31d196797209f06fff677c4d9 - 0.1% (12.1 MB)
[#051] sha256:6e54527d560122bf2561100d7f7aa180719d42b306880807d75b9745acb10115 - 0.11% (13.9 MB)
History
/bin/sh -c #(nop) ARG RELEASE
2025-07-14 16:42:11 UTC/bin/sh -c #(nop) ARG LAUNCHPAD_BUILD_ARCH
2025-07-14 16:42:11 UTC/bin/sh -c #(nop) LABEL org.opencontainers.image.ref.name=ubuntu
2025-07-14 16:42:12 UTC/bin/sh -c #(nop) LABEL org.opencontainers.image.version=24.04
2025-07-14 16:42:14 UTC/bin/sh -c #(nop) ADD file:98827745561bc42c12745b10037415621cf3b8ec8497a5953995b5fc7969bf17 in /
2025-07-14 16:42:15 UTC/bin/sh -c #(nop) CMD ["/bin/bash"]
2025-07-18 20:17:43 UTC (buildkit.dockerfile.v0)ENV NVARCH=sbsa
2025-07-18 20:17:43 UTC (buildkit.dockerfile.v0)ENV NVIDIA_REQUIRE_CUDA=cuda>=12.9 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>=560,driver<561 brand=grid,driver>=560,driver<561 brand=tesla,driver>=560,driver<561 brand=nvidia,driver>=560,driver<561 brand=quadro,driver>=560,driver<561 brand=quadrortx,driver>=560,driver<561 brand=nvidiartx,driver>=560,driver<561 brand=vapps,driver>=560,driver<561 brand=vpc,driver>=560,driver<561 brand=vcs,driver>=560,driver<561 brand=vws,driver>=560,driver<561 brand=cloudgaming,driver>=560,driver<561 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
2025-07-18 20:17:43 UTC (buildkit.dockerfile.v0)ENV NV_CUDA_CUDART_VERSION=12.9.79-1
2025-07-18 20:17:43 UTC (buildkit.dockerfile.v0)ARG TARGETARCH
2025-07-18 20:17:43 UTC (buildkit.dockerfile.v0)LABEL maintainer=NVIDIA CORPORATION <cudatools@nvidia.com>
2025-07-18 20:17:43 UTC (buildkit.dockerfile.v0)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
2025-07-18 20:17:43 UTC (buildkit.dockerfile.v0)ENV CUDA_VERSION=12.9.1
2025-07-18 20:20:54 UTC (buildkit.dockerfile.v0)RUN |1 TARGETARCH=arm64 /bin/sh -c apt-get update && apt-get install -y --no-install-recommends cuda-cudart-12-9=${NV_CUDA_CUDART_VERSION} cuda-compat-12-9 && rm -rf /var/lib/apt/lists/* # buildkit
2025-07-18 20:20:54 UTC (buildkit.dockerfile.v0)RUN |1 TARGETARCH=arm64 /bin/sh -c echo "/usr/local/cuda/lib64" >> /etc/ld.so.conf.d/nvidia.conf # buildkit
2025-07-18 20:20:54 UTC (buildkit.dockerfile.v0)ENV PATH=/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
2025-07-18 20:20:54 UTC (buildkit.dockerfile.v0)ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64
2025-07-18 20:20:54 UTC (buildkit.dockerfile.v0)COPY NGC-DL-CONTAINER-LICENSE / # buildkit
2025-07-18 20:20:54 UTC (buildkit.dockerfile.v0)ENV NVIDIA_VISIBLE_DEVICES=all
2025-07-18 20:20:54 UTC (buildkit.dockerfile.v0)ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG INSTALL_RUBIN_PRERELEASE
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG TARGETPLATFORM
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG max_jobs=2
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG CUDA_VERSION
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG PYTHON_VERSION
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG NCCL_VERSION
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG DEADSNAKES_MIRROR_URL
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG DEADSNAKES_GPGKEY_URL
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ARG GET_PIP_URL
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ENV CUDA_VERSION=12.9.1
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)ENV DEBIAN_FRONTEND=noninteractive
2026-10-03 07:45:37 UTC (buildkit.dockerfile.v0)WORKDIR /vllm-workspace
2026-10-03 07:45:38 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c PYTHON_VERSION_STR=$(echo ${PYTHON_VERSION} | sed 's/\.//g') && echo "export PYTHON_VERSION_STR=${PYTHON_VERSION_STR}" >> /etc/environment # buildkit
2026-10-03 07:46:31 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 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
2026-10-03 07:47:11 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 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-nvrtc-dev-${CUDA_VERSION_DASH} cuda-cuobjdump-${CUDA_VERSION_DASH} libcurand-dev-${CUDA_VERSION_DASH} libcublas-dev-${CUDA_VERSION_DASH} libnuma-dev numactl && rm -rf /var/lib/apt/lists/* # buildkit
2026-10-03 07:47:11 UTC (buildkit.dockerfile.v0)ENV CUDA_HOME=/usr/local/cuda
2026-10-03 07:47:11 UTC (buildkit.dockerfile.v0)ENV PATH=/usr/local/cuda/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
2026-10-03 07:47:11 UTC (buildkit.dockerfile.v0)ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:/usr/local/cuda/lib64
2026-10-03 07:47:12 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 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 if [ "${INSTALL_RUBIN_PRERELEASE}" = "true" ]; then CUDA_MAJOR_MINOR=$(echo "${CUDA_VERSION}" | cut -d. -f1,2) && if [ "${CUDA_MAJOR_MINOR}" = "13.4" ]; then CURAND_LIBRARY=$(find -H /usr/local/cuda -name libcurand.so.10 -print -quit) && test -n "${CURAND_LIBRARY}" && ln -sf "${CURAND_LIBRARY}" /usr/local/cuda/lib64/libcurand.so.11; elif [ "${CUDA_MAJOR_MINOR}" = "13.5" ]; then CURAND_LIBRARY=$(find -H /usr/local/cuda -name libcurand.so.11 -print -quit) && test -n "${CURAND_LIBRARY}" && ! readlink -f "${CURAND_LIBRARY}" | grep -q 'libcurand\.so\.10'; else echo "Unsupported Rubin CUDA version: ${CUDA_MAJOR_MINOR}" >&2; exit 1; fi && if NCCL_LIBRARY=$(find /usr/lib -name libnccl.so.2 -print -quit) && [ -n "${NCCL_LIBRARY}" ]; then if [ "${NCCL_LIBRARY}" != "/usr/lib/libnccl.so.2" ]; then ln -sf "${NCCL_LIBRARY}" /usr/lib/libnccl.so.2; fi && ln -sf libnccl.so.2 /usr/lib/libnccl.so; fi && ldconfig; fi # buildkit
2026-10-03 07:47:12 UTC (buildkit.dockerfile.v0)COPY tools/install_snapshot_runtime.sh /usr/local/bin/install_snapshot_runtime.sh # buildkit
2026-10-03 07:47:13 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c if [ "${TARGETPLATFORM}" = "linux/amd64" ] && [ "${CUDA_VERSION%%.*}" -ge 13 ]; then MAX_JOBS="${max_jobs}" /usr/local/bin/install_snapshot_runtime.sh; else echo "Initialized snapshots are not installed for ${TARGETPLATFORM} with CUDA ${CUDA_VERSION}"; fi # buildkit
2026-10-03 07:47:13 UTC (buildkit.dockerfile.v0)ENV CRIU_CUDA_PLUGIN_DIR=/usr/local/lib/criu
2026-10-03 07:47:15 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c python3 -m pip install uv # buildkit
2026-10-03 07:47:15 UTC (buildkit.dockerfile.v0)ENV UV_HTTP_TIMEOUT=500
2026-10-03 07:47:15 UTC (buildkit.dockerfile.v0)ENV UV_INDEX_STRATEGY=unsafe-best-match
2026-10-03 07:47:15 UTC (buildkit.dockerfile.v0)ENV UV_LINK_MODE=copy
2026-10-03 07:47:15 UTC (buildkit.dockerfile.v0)ENV UV_PYTHON_INSTALL_DIR=/opt/uv/python
2026-10-03 07:47:15 UTC (buildkit.dockerfile.v0)ENV UV_CACHE_DIR=/opt/uv/cache
2026-10-03 07:47:15 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c mkdir -p "${UV_PYTHON_INSTALL_DIR}" "${UV_CACHE_DIR}" && chgrp -R 0 /opt/uv && chmod -R g+rwX,a+rX /opt/uv # buildkit
2026-10-03 07:47:16 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py /bin/sh -c CUDA_MAJOR=$(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
2026-10-03 07:47:16 UTC (buildkit.dockerfile.v0)ENV UV_OVERRIDE=/etc/uv-overrides.txt
2026-10-03 07:47:17 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 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
2026-10-03 07:47:17 UTC (buildkit.dockerfile.v0)COPY docker/entrypoints/vllm-nonroot-entrypoint.sh /usr/local/bin/vllm-nonroot-entrypoint.sh # buildkit
2026-10-03 07:47:18 UTC (buildkit.dockerfile.v0)RUN |9 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 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
2026-10-03 07:47:18 UTC (buildkit.dockerfile.v0)ENV VLLM_ENABLE_CUDA_COMPATIBILITY=0
2026-10-03 07:47:18 UTC (buildkit.dockerfile.v0)ARG PYTORCH_CUDA_INDEX_BASE_URL
2026-10-03 07:47:18 UTC (buildkit.dockerfile.v0)COPY requirements/common.txt /tmp/common.txt # buildkit
2026-10-03 07:47:18 UTC (buildkit.dockerfile.v0)COPY requirements/cuda.txt /tmp/requirements-cuda.txt # buildkit
2026-10-03 07:48:20 UTC (buildkit.dockerfile.v0)RUN |10 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 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 [ "${INSTALL_RUBIN_PRERELEASE}" = "true" ] && [ ! -f /tmp/vllm-requirements/rubin-prerelease.txt ]; then echo "INSTALL_RUBIN_PRERELEASE=true requires requirements/rubin-prerelease.txt" >&2; exit 1; fi && if [ "$(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 && if [ "${INSTALL_RUBIN_PRERELEASE}" = "true" ]; then uv pip install --system --prerelease=allow --index-strategy unsafe-first-match -r /tmp/vllm-requirements/rubin-prerelease.txt && sed -Ei '/^[[:space:]]*(torch|torchvision|torchaudio)[[:space:]]*==/Id' /tmp/requirements-cuda.txt && uv pip install --system -r /tmp/requirements-cuda.txt && uv pip install --system --prerelease=allow --index-strategy unsafe-first-match -r /tmp/vllm-requirements/rubin-prerelease.txt && printf '%s\n' "/usr/local/lib/python${PYTHON_VERSION}/dist-packages/nvpl/lib" > /etc/ld.so.conf.d/vllm-rubin-python.conf && ldconfig; else uv pip install --system -r /tmp/requirements-cuda.txt --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); fi && rm /tmp/requirements-cuda.txt /tmp/common.txt # buildkit
2026-10-03 07:48:20 UTC (buildkit.dockerfile.v0)ARG FLASHINFER_VERSION=0.7.0.post1
2026-10-03 07:48:30 UTC (buildkit.dockerfile.v0)RUN |11 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl FLASHINFER_VERSION=0.7.0.post1 /bin/sh -c if [ "${INSTALL_RUBIN_PRERELEASE}" != "true" ]; then uv pip install --system flashinfer-jit-cache==${FLASHINFER_VERSION} --index-url https://flashinfer.ai/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); fi # buildkit
2026-10-03 07:48:30 UTC (buildkit.dockerfile.v0)ARG GDRCOPY_CUDA_VERSION=12.8
2026-10-03 07:48:30 UTC (buildkit.dockerfile.v0)ARG GDRCOPY_OS_VERSION=Ubuntu24_04
2026-10-03 07:48:30 UTC (buildkit.dockerfile.v0)ARG TARGETPLATFORM
2026-10-03 07:48:30 UTC (buildkit.dockerfile.v0)COPY tools/install_gdrcopy.sh /tmp/install_gdrcopy.sh # buildkit
2026-10-03 07:48:38 UTC (buildkit.dockerfile.v0)RUN |14 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 PYTHON_VERSION=3.12 NCCL_VERSION=2.30.7 DEADSNAKES_MIRROR_URL= DEADSNAKES_GPGKEY_URL= GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl FLASHINFER_VERSION=0.7.0.post1 GDRCOPY_CUDA_VERSION=12.8 GDRCOPY_OS_VERSION=Ubuntu24_04 TARGETPLATFORM=linux/arm64 /bin/sh -c set -eux; if [ "${INSTALL_RUBIN_PRERELEASE}" = "true" ] && ldconfig -p | grep -q 'libgdrapi\.so'; then true; else 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}"; fi; rm /tmp/install_gdrcopy.sh # buildkit
2026-10-03 07:48:38 UTC (buildkit.dockerfile.v0)ARG TIMM_VERSION=>=1.0.17
2026-10-03 07:48:38 UTC (buildkit.dockerfile.v0)ARG RUNAI_MODEL_STREAMER_VERSION=>=0.15.7
2026-10-03 07:48:42 UTC (buildkit.dockerfile.v0)RUN |16 INSTALL_RUBIN_PRERELEASE=false TARGETPLATFORM=linux/arm64 max_jobs=16 CUDA_VERSION=12.9.1 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=0.7.0.post1 GDRCOPY_CUDA_VERSION=12.8 GDRCOPY_OS_VERSION=Ubuntu24_04 TARGETPLATFORM=linux/arm64 TIMM_VERSION=>=1.0.17 RUNAI_MODEL_STREAMER_VERSION=>=0.15.7 /bin/sh -c uv pip install --system accelerate 'modelscope<1.38' "timm${TIMM_VERSION}" "runai-model-streamer[s3,gcs,azure]${RUNAI_MODEL_STREAMER_VERSION}" # buildkit
2026-10-03 07:48:42 UTC (buildkit.dockerfile.v0)ARG PIP_INDEX_URL UV_INDEX_URL
2026-10-03 07:48:42 UTC (buildkit.dockerfile.v0)ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL
2026-10-03 07:48:42 UTC (buildkit.dockerfile.v0)ARG PYTORCH_CUDA_INDEX_BASE_URL
2026-10-03 07:48:42 UTC (buildkit.dockerfile.v0)ARG PIP_KEYRING_PROVIDER UV_KEYRING_PROVIDER
2026-10-03 07:48:42 UTC (buildkit.dockerfile.v0)ARG PYTORCH_NIGHTLY
2026-10-03 08:02:15 UTC (buildkit.dockerfile.v0)COPY /workspace/torch_lib_versions.txt torch_lib_versions.txt # buildkit
2026-10-03 08:20:38 UTC (buildkit.dockerfile.v0)COPY /workspace/dist/wheel.sha256 /tmp/vllm-wheel.sha256 # buildkit
2026-10-03 08:20:44 UTC (buildkit.dockerfile.v0)RUN |8 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 $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); fi # buildkit
2026-10-03 08:20:44 UTC (buildkit.dockerfile.v0)RUN |8 PIP_INDEX_URL= UV_INDEX_URL= PIP_EXTRA_INDEX_URL= UV_EXTRA_INDEX_URL= PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl PIP_KEYRING_PROVIDER=disabled UV_KEYRING_PROVIDER=disabled PYTORCH_NIGHTLY= /bin/sh -c ln -s "$(python3 -c 'import sysconfig; print(sysconfig.get_path("platlib"))')/vllm/vllm-rs" /usr/local/bin/vllm-rs && vllm-rs --version # buildkit
2026-10-03 08:20:45 UTC (buildkit.dockerfile.v0)RUN |8 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
2026-10-03 08:20:45 UTC (buildkit.dockerfile.v0)ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:/usr/local/cuda/lib64:/usr/local/cuda/lib64
2026-10-03 08:20:45 UTC (buildkit.dockerfile.v0)COPY /tmp/ep_kernels_workspace/dist/wheels.sha256 /tmp/ep-kernels-wheels.sha256 # buildkit
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)RUN |8 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
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:/usr/local/cuda/lib64:/usr/local/cuda/lib64
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)COPY examples examples # buildkit
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)COPY benchmarks benchmarks # buildkit
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)COPY ./vllm/collect_env.py . # buildkit
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG TARGETPLATFORM
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG INSTALL_KV_CONNECTORS=false
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG INSTALL_RUBIN_PRERELEASE
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG TRITON_INSTALL_FROM_SOURCE_REPO
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG CUDA_VERSION
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG VLLM_BUILD_COMMIT
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG VLLM_BUILD_PIPELINE
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG VLLM_BUILD_URL
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG VLLM_IMAGE_TAG
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG PIP_INDEX_URL UV_INDEX_URL
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ENV UV_HTTP_TIMEOUT=500
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ARG torch_cuda_arch_list=7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0
2026-10-03 08:20:46 UTC (buildkit.dockerfile.v0)ENV TORCH_CUDA_ARCH_LIST=8.0 8.7 8.9 9.0 10.0 12.0
2026-10-03 08:27:46 UTC (buildkit.dockerfile.v0)RUN |14 TARGETPLATFORM=linux/arm64 INSTALL_KV_CONNECTORS=true INSTALL_RUBIN_PRERELEASE=false TRITON_INSTALL_FROM_SOURCE_REPO= CUDA_VERSION=12.9.1 VLLM_BUILD_COMMIT=db9527a46873454610df6dbedf79a36d6bf1a7f6 VLLM_BUILD_PIPELINE=019d130e-464e-4ff7-b84b-492992c0c06b VLLM_BUILD_URL=https://buildkite.com/vllm/release-v2/builds/7746 VLLM_IMAGE_TAG=vllm/vllm-openai:v0.31.0-cu129-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 12.0 /bin/sh -c set -eu; 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 [ "$INSTALL_KV_CONNECTORS" = "true" ]; then KV_REQUIREMENTS=/tmp/vllm-requirements/kv_connectors.txt; if [ "${CUDA_MAJOR}" = "12" ]; then KV_REQUIREMENTS=/tmp/vllm-requirements/kv_connectors_cu12.txt; fi; uv pip freeze --system > /tmp/kv-installed.txt; grep -E '^(torch|torchaudio|torchvision|triton|cuda-bindings|cuda-python|cuda-toolkit)==' /tmp/kv-installed.txt > /tmp/kv-constraints.txt; uv pip install --system -c /tmp/kv-constraints.txt -r "${KV_REQUIREMENTS}" --no-build || ( apt-get update -y && apt-get install -y --no-install-recommends --allow-change-held-packages ${BUILD_PKGS} && uv pip install --system -c /tmp/kv-constraints.txt -r "${KV_REQUIREMENTS}" --no-build-isolation && apt-get purge -y ${BUILD_PKGS} && rm -rf /var/lib/apt/lists/* ); MOONCAKE_VERSION=$(python3 -c "import importlib.metadata as m; print(m.version('mooncake-transfer-engine'))" 2>/dev/null || true) && if [ "${CUDA_MAJOR}" = "13" ] && [ -n "${MOONCAKE_VERSION}" ]; then uv pip uninstall --system mooncake-transfer-engine 2>/dev/null || true; uv pip install --system -c /tmp/kv-constraints.txt "mooncake-transfer-engine-cuda13==${MOONCAKE_VERSION}"; fi; rm -f /tmp/kv-installed.txt /tmp/kv-constraints.txt; fi # buildkit
2026-10-03 08:27:47 UTC (buildkit.dockerfile.v0)RUN |14 TARGETPLATFORM=linux/arm64 INSTALL_KV_CONNECTORS=true INSTALL_RUBIN_PRERELEASE=false TRITON_INSTALL_FROM_SOURCE_REPO= CUDA_VERSION=12.9.1 VLLM_BUILD_COMMIT=db9527a46873454610df6dbedf79a36d6bf1a7f6 VLLM_BUILD_PIPELINE=019d130e-464e-4ff7-b84b-492992c0c06b VLLM_BUILD_URL=https://buildkite.com/vllm/release-v2/builds/7746 VLLM_IMAGE_TAG=vllm/vllm-openai:v0.31.0-cu129-ubuntu2404 PIP_INDEX_URL= UV_INDEX_URL= PIP_EXTRA_INDEX_URL= UV_EXTRA_INDEX_URL= torch_cuda_arch_list=8.0 8.7 8.9 9.0 10.0 12.0 /bin/sh -c if [ "${INSTALL_RUBIN_PRERELEASE}" = "true" ]; then echo "Installing Rubin prerelease requirements..." && uv pip install --system --prerelease=allow --index-strategy unsafe-first-match -r /tmp/vllm-requirements/rubin-prerelease.txt; fi # buildkit
2026-10-03 08:27:47 UTC (buildkit.dockerfile.v0)COPY /tmp/triton_from_source_workspace/dist/wheel.sha256 /tmp/triton-source-wheel.sha256 # buildkit
2026-10-03 08:27:47 UTC (buildkit.dockerfile.v0)COPY tools/install_triton_from_source.sh /tmp/install_triton_from_source.sh # buildkit
2026-10-03 08:27:47 UTC (buildkit.dockerfile.v0)RUN |14 TARGETPLATFORM=linux/arm64 INSTALL_KV_CONNECTORS=true INSTALL_RUBIN_PRERELEASE=false TRITON_INSTALL_FROM_SOURCE_REPO= CUDA_VERSION=12.9.1 VLLM_BUILD_COMMIT=db9527a46873454610df6dbedf79a36d6bf1a7f6 VLLM_BUILD_PIPELINE=019d130e-464e-4ff7-b84b-492992c0c06b VLLM_BUILD_URL=https://buildkite.com/vllm/release-v2/builds/7746 VLLM_IMAGE_TAG=vllm/vllm-openai:v0.31.0-cu129-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 12.0 /bin/sh -c if [ -n "${TRITON_INSTALL_FROM_SOURCE_REPO}" ]; then CUDA_MAJOR_MINOR=$(echo "${CUDA_VERSION}" | cut -d. -f1,2); if [ "${INSTALL_RUBIN_PRERELEASE}" = "true" ]; then case "${CUDA_MAJOR_MINOR}" in 13.4|13.5) TRITON_SYSTEM_PTXAS_PATH=/usr/local/cuda/bin/ptxas TRITON_EXPECTED_PTXAS_VERSION="${CUDA_MAJOR_MINOR}" bash /tmp/install_triton_from_source.sh /tmp/triton-source-dist /tmp/triton-source-wheel.sha256 ;; *) echo "Unsupported Rubin CUDA version: ${CUDA_MAJOR_MINOR}" >&2; exit 1 ;; esac; else bash /tmp/install_triton_from_source.sh /tmp/triton-source-dist /tmp/triton-source-wheel.sha256; fi; fi && rm -f /tmp/install_triton_from_source.sh /tmp/triton-source-wheel.sha256 # buildkit
2026-10-03 08:27:47 UTC (buildkit.dockerfile.v0)ENV VLLM_USAGE_SOURCE=production-docker-image
2026-10-03 08:27:47 UTC (buildkit.dockerfile.v0)ENV VLLM_BUILD_COMMIT=db9527a46873454610df6dbedf79a36d6bf1a7f6 VLLM_BUILD_PIPELINE=019d130e-464e-4ff7-b84b-492992c0c06b VLLM_BUILD_URL=https://buildkite.com/vllm/release-v2/builds/7746 VLLM_IMAGE_TAG=vllm/vllm-openai:v0.31.0-cu129-ubuntu2404
2026-10-03 08:27:47 UTC (buildkit.dockerfile.v0)LABEL org.opencontainers.image.source=https://github.com/vllm-project/vllm org.opencontainers.image.revision=db9527a46873454610df6dbedf79a36d6bf1a7f6 org.opencontainers.image.version=vllm/vllm-openai:v0.31.0-cu129-ubuntu2404 org.opencontainers.image.url=https://buildkite.com/vllm/release-v2/builds/7746 ai.vllm.build.commit=db9527a46873454610df6dbedf79a36d6bf1a7f6 ai.vllm.build.pipeline=019d130e-464e-4ff7-b84b-492992c0c06b ai.vllm.build.url=https://buildkite.com/vllm/release-v2/builds/7746 ai.vllm.image.tag=vllm/vllm-openai:v0.31.0-cu129-ubuntu2404
2026-10-03 08:27:47 UTC (buildkit.dockerfile.v0)ENTRYPOINT ["vllm" "serve"]
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)SHELL [/bin/bash -eo pipefail -c]
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)ARG TARGETPLATFORM=linux/arm64
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)ARG TARGETOS=linux
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)ARG TARGETARCH=arm64
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)ENV DEBIAN_FRONTEND=noninteractive LANG=en_US.UTF-8 LANGUAGE=en_US:en LC_ALL=en_US.UTF-8
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)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
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)ARG PYTHON_VERSION=3.12
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)ENV PYTHON_VERSION=3.12
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)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
2026-10-11 15:55:45 UTC (buildkit.dockerfile.v0)ENV PIP_NO_CACHE_DIR=1 PIP_DISABLE_PIP_VERSION_CHECK=1 PIP_ROOT_USER_ACTION=ignore
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)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
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG CUDA_VERSION=12.9.1
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG CUDA_ARCHS
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ENV CUDA_HOME=/usr/local/cuda CUDA_VERSION=12.9.1 CUDA_ARCHS=
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)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
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG VLLM_VERSION=0.31.0
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG VLLM_TORCH_VERSION=2.13.0
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG VLLM_TORCH_CUDA_VERSION=12.9.1
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ENV VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)SHELL [/bin/bash -eo pipefail -c]
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG TARGETPLATFORM=linux/arm64
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG TARGETOS=linux
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG TARGETARCH=arm64
2026-10-11 15:55:48 UTC (buildkit.dockerfile.v0)ARG VLLM_ROUTER_VERSION=0.1.15
2026-10-11 15:55:49 UTC (buildkit.dockerfile.v0)RUN |10 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 /bin/bash -eo pipefail -c # vLLM Router # # The PD router GPUStack renders for a disaggregated group. Without it in # the image there is no router to run, and the group either falls back to # an engine example script (no metrics, no circuit breaker, no health # check) or does not start at all. uv pip install \ vllm-router==${VLLM_ROUTER_VERSION} # Fail the build rather than ship an image whose router is missing: the # alternative surfaces as a container that exits at deploy time, one # layer away from anything that explains it. command -v vllm-router >/dev/null # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit
2026-10-11 15:55:50 UTC (buildkit.dockerfile.v0)RUN |10 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 /bin/bash -eo pipefail -c # Ray # Install Ray Client and Default. # Pinned to 2.54.0: ray 2.55 (ray-project/ray#61029) breaks distributed # ray.init when the driver has no co-located raylet (gpustack split containers). RAY_VERSION=2.54.0 cat <<EOT >/tmp/requirements.txt ray[cgraph]==${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
2026-10-11 15:57:08 UTC (buildkit.dockerfile.v0)RUN |10 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 /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
2026-10-11 15:57:08 UTC (buildkit.dockerfile.v0)ARG VLLM_LMCACHE_VERSION=0.5.6
2026-10-11 15:57:08 UTC (buildkit.dockerfile.v0)ENV VLLM_LMCACHE_VERSION=0.5.6
2026-10-16 08:04:54 UTC (buildkit.dockerfile.v0)RUN |11 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 VLLM_LMCACHE_VERSION=0.5.6 /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
2026-10-16 08:04:56 UTC (buildkit.dockerfile.v0)RUN |11 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 VLLM_LMCACHE_VERSION=0.5.6 /bin/bash -eo pipefail -c # CuPy # ray[cgraph] pulls cupy-cuda12x while a cu13 base already ships cupy-cuda13x; both share # the same cupy/ tree, so uninstalling one breaks the other. Purge both, then reinstall # only the CUDA-matching variant at the version vLLM pinned in the base, so cu129 also # ends on 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" # The base ships the variant matching its own CUDA major (cupy-cuda13x on cu13, # cupy-cuda12x on cu12), so read the pinned version from that same variant. CUPY_VER="$(uv pip show "${CUPY}" 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} 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
2026-10-16 08:04:56 UTC (buildkit.dockerfile.v0)RUN |11 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 VLLM_LMCACHE_VERSION=0.5.6 /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
2026-10-16 08:04:58 UTC (buildkit.dockerfile.v0)RUN |11 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 VLLM_LMCACHE_VERSION=0.5.6 /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 $(pip show vllm_omni | grep Location: | cut -d" " -f 2) \ && patch -p1 < /workspace/patches/vllm_omni/*.patch; \ fi # buildkit
2026-10-16 08:04:58 UTC (buildkit.dockerfile.v0)RUN |11 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 VLLM_LMCACHE_VERSION=0.5.6 /bin/bash -eo pipefail -c # Dependencies # Install cat <<EOT >/tmp/requirements.txt # audio extras librosa soundfile mistral_common[audio] # tokenizer extras fastokens==0.2.0 EOT uv pip install \ -r /tmp/requirements.txt # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit
2026-10-16 08:04:58 UTC (buildkit.dockerfile.v0)ARG VLLM_TORCHCODEC_VERSION=0.17.0
2026-10-16 08:05:04 UTC (buildkit.dockerfile.v0)RUN |12 TARGETPLATFORM=linux/arm64 TARGETOS=linux TARGETARCH=arm64 PYTHON_VERSION=3.12 CUDA_VERSION=12.9.1 CUDA_ARCHS= VLLM_VERSION=0.31.0 VLLM_TORCH_VERSION=2.13.0 VLLM_TORCH_CUDA_VERSION=12.9.1 VLLM_ROUTER_VERSION=0.1.15 VLLM_LMCACHE_VERSION=0.5.6 VLLM_TORCHCODEC_VERSION=0.17.0 /bin/bash -eo pipefail -c # TorchCodec if [[ -z "${VLLM_TORCHCODEC_VERSION}" ]]; then echo "Keeping the base image's torchcodec..." exit 0 fi # Replace the base's torchcodec with the CPU build of the same version —— # see the note on VLLM_TORCHCODEC_VERSION. --no-deps keeps uv from pulling # a CPU torch over the base's CUDA build. uv pip install --no-deps \ "torchcodec==${VLLM_TORCHCODEC_VERSION}+cpu" \ --index-url https://download.pytorch.org/whl/cpu # Fail the build on the import path that crashes the stock cu129 image # (vllm-project/vllm#60406); needs no GPU. python -c "import torch, torchcodec.decoders; print('torch', torch.__version__, 'cuda', torch.version.cuda)" # Cleanup rm -rf /var/tmp/* \ && rm -rf /tmp/* # buildkit
2026-10-16 08:05:04 UTC (buildkit.dockerfile.v0)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
2026-10-16 08:05:04 UTC (buildkit.dockerfile.v0)WORKDIR /
2026-10-16 08:05:04 UTC (buildkit.dockerfile.v0)ENTRYPOINT ["tini" "--"]
Delete Tag
Please be careful as this will not just delete the reference but also the actual content!
For example when you have latest and v1.2.3 both pointing to the same image
the deletion of latest will also permanently remove v1.2.3.