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

Removed unnessesary dependencies

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-01-29 09:58:44 -08:00
parent e4b57633fe
commit 36e3b7af27
1 changed files with 16 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

59
Dockerfile.mxfp4
View File

@@ -6,7 +6,7 @@ ARG BUILD_JOBS=16
# ========================================================= # =========================================================
# STAGE 1: Base Image (Installs Dependencies) # STAGE 1: Base Image (Installs Dependencies)
# ========================================================= # =========================================================
FROM nvidia/cuda:13.1.0-devel-ubuntu24.04 AS base FROM nvcr.io/nvidia/pytorch:25.12-py3 AS base
# Build parallemism # Build parallemism
ARG BUILD_JOBS ARG BUILD_JOBS
@@ -36,8 +36,8 @@ ENV PIP_BREAK_SYSTEM_PACKAGES=1
# Set pip cache directory # Set pip cache directory
ENV PIP_CACHE_DIR=/root/.cache/pip ENV PIP_CACHE_DIR=/root/.cache/pip
ENV UV_CACHE_DIR=/root/.cache/uv ENV UV_CACHE_DIR=/root/.cache/uv
#ENV UV_SYSTEM_PYTHON=1 ENV UV_SYSTEM_PYTHON=1
#ENV UV_BREAK_SYSTEM_PACKAGES=1 ENV UV_BREAK_SYSTEM_PACKAGES=1
ENV UV_LINK_MODE=copy ENV UV_LINK_MODE=copy
# Set the base directory environment variable # Set the base directory environment variable
@@ -45,15 +45,12 @@ ENV VLLM_BASE_DIR=/workspace/vllm
# 1. Install Build Dependencies & Ccache # 1. Install Build Dependencies & Ccache
# Added ccache to enable incremental compilation caching # Added ccache to enable incremental compilation caching
RUN apt update && apt upgrade -y \ RUN apt update && \
&& apt install -y --allow-change-held-packages --no-install-recommends \ apt install -y --no-install-recommends \
curl vim cmake build-essential ninja-build \ curl vim ninja-build git \
libcudnn9-cuda-13 libcudnn9-dev-cuda-13 \
python3-dev python3-pip git wget \
libnccl-dev libnccl2 libibverbs1 libibverbs-dev rdma-core \
ccache \ ccache \
&& rm -rf /var/lib/apt/lists/* \ && rm -rf /var/lib/apt/lists/* \
&& pip install uv && pip install uv && pip uninstall -y flash-attn
# Configure Ccache for CUDA/C++ # Configure Ccache for CUDA/C++
ENV PATH=/usr/lib/ccache:$PATH ENV PATH=/usr/lib/ccache:$PATH
@@ -77,18 +74,6 @@ ENV TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
# Change this argument to force a re-download of PyTorch/FlashInfer # Change this argument to force a re-download of PyTorch/FlashInfer
ARG CACHEBUST_DEPS=1 ARG CACHEBUST_DEPS=1
# Initialize virtual environment
ENV VIRTUAL_ENV=/workspace/python-venv
RUN uv venv /workspace/python-venv
ENV PATH="$VIRTUAL_ENV/bin:$PATH"
# 3. Install Python Dependencies with Cache Mounts
# Using --mount=type=cache ensures that even if this layer invalidates,
# pip reuses previously downloaded wheels.
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130
# Install additional dependencies # Install additional dependencies
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install fastsafetensors uv pip install fastsafetensors
@@ -238,7 +223,7 @@ RUN --mount=type=cache,id=ccache,target=/root/.ccache \
# ========================================================= # =========================================================
# STAGE 4: Runner (Transfers only necessary artifacts) # STAGE 4: Runner (Transfers only necessary artifacts)
# ========================================================= # =========================================================
FROM nvidia/cuda:13.1.0-devel-ubuntu24.04 AS runner FROM nvcr.io/nvidia/pytorch:25.12-py3 AS runner
ENV DEBIAN_FRONTEND=noninteractive ENV DEBIAN_FRONTEND=noninteractive
ENV PIP_BREAK_SYSTEM_PACKAGES=1 ENV PIP_BREAK_SYSTEM_PACKAGES=1
@@ -247,20 +232,18 @@ ENV VLLM_BASE_DIR=/workspace/vllm
# Set pip cache directory # Set pip cache directory
ENV PIP_CACHE_DIR=/root/.cache/pip ENV PIP_CACHE_DIR=/root/.cache/pip
ENV UV_CACHE_DIR=/root/.cache/uv ENV UV_CACHE_DIR=/root/.cache/uv
#ENV UV_SYSTEM_PYTHON=1 ENV UV_SYSTEM_PYTHON=1
#ENV UV_BREAK_SYSTEM_PACKAGES=1 ENV UV_BREAK_SYSTEM_PACKAGES=1
ENV UV_LINK_MODE=copy ENV UV_LINK_MODE=copy
# Install minimal runtime dependencies (NCCL, Python) # Install minimal runtime dependencies (NCCL, Python)
# Note: "devel" tools like cmake/gcc are NOT installed here to save space # Note: "devel" tools like cmake/gcc are NOT installed here to save space
RUN apt update && apt upgrade -y \ RUN apt update && \
&& apt install -y --allow-change-held-packages --no-install-recommends \ apt install -y --no-install-recommends \
python3 python3-pip python3-dev vim curl git wget \ curl vim git \
libcudnn9-cuda-13 \
libnccl-dev libnccl2 libibverbs1 libibverbs-dev rdma-core \
libxcb1 \ libxcb1 \
&& rm -rf /var/lib/apt/lists/* \ && rm -rf /var/lib/apt/lists/* \
&& pip install uv && pip install uv && pip uninstall -y flash-attn
# Set final working directory # Set final working directory
WORKDIR $VLLM_BASE_DIR WORKDIR $VLLM_BASE_DIR
@@ -273,11 +256,8 @@ RUN mkdir -p tiktoken_encodings && \
# Copy artifacts from Builder Stage # Copy artifacts from Builder Stage
# We copy the python packages and executables # We copy the python packages and executables
# No need to copy source code, as it's already in the site-packages # No need to copy source code, as it's already in the site-packages
COPY --from=builder /workspace/python-venv /workspace/python-venv COPY --from=builder /usr/local/lib/python3.12/dist-packages /usr/local/lib/python3.12/dist-packages
COPY --from=builder /usr/local/bin /usr/local/bin
# Activate virtual environment
ENV VIRTUAL_ENV=/workspace/python-venv
ENV PATH="$VIRTUAL_ENV/bin:$PATH"
# Setup Env for Runtime # Setup Env for Runtime
ENV TORCH_CUDA_ARCH_LIST="12.0;12.1" ENV TORCH_CUDA_ARCH_LIST="12.0;12.1"
@@ -293,10 +273,3 @@ RUN chmod +x $VLLM_BASE_DIR/run-cluster-node.sh
# Final extra deps # Final extra deps
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install ray[default] uv pip install ray[default]
# Create entrypoint script to activate venv
# RUN echo '#!/bin/bash\nsource /workspace/python-venv/bin/activate\nexec "$@"' > /entrypoint.sh && \
# chmod +x /entrypoint.sh
# ENTRYPOINT ["/entrypoint.sh"]
# CMD ["bash"]