A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 8, 19:00 UTC
[XPU]Replace pip in docker.xpu with uv pip (#31112)
Signed-off-by: sihao.li <sihao.li@intel.com>
sihao_li committed
Feb 6, 2026 at 14:02 UTC
6550815c3ad5fc20e6944483dd8a7a47c18b7c7d
1 file changed
+46
-28
docker/Dockerfile.xpu
+46
-28
@@ -1,5 +1,10 @@
1
FROM intel/deep-learning-essentials:2025.3.2-0-devel-ubuntu24.04 AS vllm-base
2
3
+WORKDIR /workspace/
4
+
5
+ARG PYTHON_VERSION=3.12
6
+ARG PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/xpu"
7
+
8
RUN wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null && \
9
echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" | tee /etc/apt/sources.list.d/oneAPI.list && \
10
add-apt-repository -y ppa:kobuk-team/intel-graphics
@@ -22,13 +27,16 @@ RUN apt clean && apt-get update -y && \
27
python3.12-dev \
28
python3-pip
29
25
-RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.12 1
26
-RUN update-alternatives --install /usr/bin/python python /usr/bin/python3.12 1
27
-
30
RUN apt update && apt upgrade -y && \
31
apt install -y libze1 libze-dev libze-intel-gpu1 intel-opencl-icd libze-intel-gpu-raytracing intel-ocloc && \
32
apt install -y intel-oneapi-compiler-dpcpp-cpp-2025.3
33
34
+ENV PATH="/root/.local/bin:$PATH"
35
+ENV VIRTUAL_ENV="/opt/venv"
36
+ENV UV_PYTHON_INSTALL_DIR=/opt/uv/python
37
+RUN curl -LsSf https://astral.sh/uv/install.sh | sh
38
+RUN uv venv --python ${PYTHON_VERSION} --seed ${VIRTUAL_ENV}
39
+ENV PATH="$VIRTUAL_ENV/bin:$PATH"
40
41
# This oneccl contains the BMG support which is not the case for default version of oneapi 2025.2.
42
ARG ONECCL_INSTALLER="intel-oneccl-2021.15.7.8_offline.sh"
@@ -44,20 +52,31 @@ SHELL ["bash", "-c"]
52
CMD ["bash", "-c", "source /root/.bashrc && exec bash"]
53
54
WORKDIR /workspace/vllm
47
-COPY requirements/xpu.txt /workspace/vllm/requirements/xpu.txt
48
-COPY requirements/common.txt /workspace/vllm/requirements/common.txt
55
50
-# suppress the python externally managed environment error
51
-RUN python3 -m pip config set global.break-system-packages true
52
-
53
-RUN --mount=type=cache,target=/root/.cache/pip \
54
- pip install --no-cache-dir \
55
- -r requirements/xpu.txt
56
-
57
-# arctic-inference is built from source which needs torch-xpu properly installed
58
-# used for suffix method speculative decoding
59
-RUN --mount=type=cache,target=/root/.cache/pip \
60
- pip install --no-cache-dir arctic-inference==0.1.1
56
+ENV UV_HTTP_TIMEOUT=500
57
+
58
+# Configure package index for XPU
59
+ENV PIP_EXTRA_INDEX_URL=${PIP_EXTRA_INDEX_URL}
60
+ENV UV_EXTRA_INDEX_URL=${PIP_EXTRA_INDEX_URL}
61
+ENV UV_INDEX_STRATEGY="unsafe-best-match"
62
+ENV UV_LINK_MODE="copy"
63
+
64
+RUN --mount=type=cache,target=/root/.cache/uv \
65
+ --mount=type=bind,src=requirements/common.txt,target=/workspace/vllm/requirements/common.txt \
66
+ --mount=type=bind,src=requirements/xpu.txt,target=/workspace/vllm/requirements/xpu.txt \
67
+ uv pip install --upgrade pip && \
68
+ uv pip install -r requirements/xpu.txt
69
+
70
+ # used for suffix method speculative decoding
71
+ # build deps for proto + nanobind-based extensions to set up the build environment
72
+RUN --mount=type=cache,target=/root/.cache/uv \
73
+ uv pip install grpcio-tools protobuf nanobind
74
+ # arctic-inference is built from source which needs torch-xpu properly installed first
75
+RUN --mount=type=cache,target=/root/.cache/uv \
76
+ source /opt/intel/oneapi/setvars.sh --force && \
77
+ source /opt/intel/oneapi/ccl/2021.15/env/vars.sh --force && \
78
+ export CMAKE_PREFIX_PATH="$(python -c 'import site; print(site.getsitepackages()[0])'):${CMAKE_PREFIX_PATH}" && \
79
+ uv pip install --no-build-isolation arctic-inference==0.1.1
80
81
ENV LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/lib/"
82
@@ -69,33 +88,32 @@ RUN --mount=type=bind,source=.git,target=.git \
88
ENV VLLM_TARGET_DEVICE=xpu
89
ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
90
72
-RUN --mount=type=cache,target=/root/.cache/pip \
91
+RUN --mount=type=cache,target=/root/.cache/uv \
92
--mount=type=bind,source=.git,target=.git \
74
- pip install --no-build-isolation .
93
+ uv pip install --no-build-isolation .
94
95
CMD ["/bin/bash"]
96
97
FROM vllm-base AS vllm-openai
98
99
# install additional dependencies for openai api server
81
-RUN --mount=type=cache,target=/root/.cache/pip \
82
- pip install accelerate hf_transfer pytest pytest_asyncio lm_eval[api] modelscope
100
+RUN --mount=type=cache,target=/root/.cache/uv \
101
+ uv pip install accelerate hf_transfer pytest pytest_asyncio lm_eval[api] modelscope
102
103
# install development dependencies (for testing)
85
-RUN python3 -m pip install -e tests/vllm_test_utils
104
+RUN uv pip install -e tests/vllm_test_utils
105
106
# install nixl from source code
107
ENV NIXL_VERSION=0.7.0
89
-RUN python3 /workspace/vllm/tools/install_nixl_from_source_ubuntu.py
108
+RUN python /workspace/vllm/tools/install_nixl_from_source_ubuntu.py
109
110
# FIX triton
92
-RUN --mount=type=cache,target=/root/.cache/pip pip uninstall triton triton-xpu -y && pip install triton-xpu==3.6.0 --extra-index-url=https://download.pytorch.org/whl/xpu
93
-
94
-# PyJWT-2.7.0 will influence some wheel behaviors, remove its dist-info to avoid conflicts
95
-RUN rm /usr/lib/python3/dist-packages/PyJWT-2.7.0.dist-info/ -rf
111
+RUN --mount=type=cache,target=/root/.cache/uv \
112
+ uv pip uninstall triton triton-xpu && \
113
+ uv pip install triton-xpu==3.6.0
114
115
# remove torch bundled oneccl to avoid conflicts
98
-RUN --mount=type=cache,target=/root/.cache/pip \
99
- pip uninstall oneccl oneccl-devel -y
116
+RUN --mount=type=cache,target=/root/.cache/uv \
117
+ uv pip uninstall oneccl oneccl-devel
118
119
ENTRYPOINT ["vllm", "serve"]