software-engineering/spark-vllm-docker
2
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

Switch to uv in the main Dockerfile

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2025-12-21 13:28:40 -08:00
parent bbd3469549
commit 11db634aad
3 changed files with 44 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
Dockerfile.wheels
View File

@@ -19,6 +19,7 @@ ENV PIP_CACHE_DIR=/root/.cache/pip
ENV UV_CACHE_DIR=/root/.cache/uv
ENV UV_SYSTEM_PYTHON=1
ENV UV_LINK_MODE=copy
ENV UV_BREAK_SYSTEM_PACKAGES=1
# Install minimal runtime dependencies (NCCL, Python)
# Note: "devel" tools like cmake/gcc are NOT installed here to save space
@@ -43,7 +44,7 @@ COPY fastsafetensors.patch .
# Install fastsafetensors
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install --system --break-system-packages -U fastsafetensors
uv pip install -U fastsafetensors
# --- VLLM SOURCE CACHE BUSTER ---
# Change THIS argument to force a fresh git clone and rebuild of vLLM
@@ -57,9 +58,9 @@ ARG WHEELS_FROM_GITHUB_RELEASE=0
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
if [ "$WHEELS_FROM_GITHUB_RELEASE" = "1" ]; then \
export VLLM_VERSION=$(curl -s https://api.github.com/repos/vllm-project/vllm/releases/latest | jq -r .tag_name | sed 's/^v//') && \
uv pip install --system --break-system-packages -U https://github.com/vllm-project/vllm/releases/download/v${VLLM_VERSION}/vllm-${VLLM_VERSION}+cu130-cp38-abi3-manylinux_2_35_aarch64.whl --torch-backend=auto; \
uv pip install -U https://github.com/vllm-project/vllm/releases/download/v${VLLM_VERSION}/vllm-${VLLM_VERSION}+cu130-cp38-abi3-manylinux_2_35_aarch64.whl --torch-backend=auto; \
else \
uv pip install --system --break-system-packages -U vllm \
uv pip install -U vllm \
--torch-backend=auto \
--extra-index-url https://wheels.vllm.ai/nightly/cu130; \
fi
@@ -72,9 +73,9 @@ ARG FLASHINFER_PRE=""
# Install flashinfer helper packages
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install --system --break-system-packages ${FLASHINFER_PRE} flashinfer-python -U --no-deps --index-url https://flashinfer.ai/whl && \
uv pip install --system --break-system-packages ${FLASHINFER_PRE} flashinfer-cubin --index-url https://flashinfer.ai/whl && \
uv pip install --system --break-system-packages ${FLASHINFER_PRE} flashinfer-jit-cache --index-url https://flashinfer.ai/whl/cu130
uv pip install ${FLASHINFER_PRE} flashinfer-python -U --no-deps --index-url https://flashinfer.ai/whl && \
uv pip install ${FLASHINFER_PRE} flashinfer-cubin --index-url https://flashinfer.ai/whl && \
uv pip install ${FLASHINFER_PRE} flashinfer-jit-cache --index-url https://flashinfer.ai/whl/cu130
# Setup Env for Runtime
ENV TORCH_CUDA_ARCH_LIST=12.1a
@@ -87,5 +88,5 @@ RUN chmod +x $VLLM_BASE_DIR/run-cluster-node.sh
# Final extra deps
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install --system --break-system-packages ray[default]
uv pip install ray[default]