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 14, 05:30 UTC
[Build/CI] Revert back to Ubuntu 20.04, install python 3.12 with uv (#26103)
Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com> Co-authored-by: Simon Mo <simon.mo@hey.com>
Tyler Michael Smith committed
Oct 3, 2025 at 01:21 UTC
27edd2aeb410e3126655ca76ef3fd32913ca52fa
2 files changed
+21
-31
docker/Dockerfile
+21
-31
@@ -13,8 +13,13 @@ ARG PYTHON_VERSION=3.12
13
# private registries that use a different repository naming conventions.
14
#
15
# Example:
16
-# docker build --build-arg BUILD_BASE_IMAGE=registry.acme.org/mirror/nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04
17
-ARG BUILD_BASE_IMAGE=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04
16
+# docker build --build-arg BUILD_BASE_IMAGE=registry.acme.org/mirror/nvidia/cuda:${CUDA_VERSION}-devel-ubuntu20.04
17
+
18
+# Important: We build with an old version of Ubuntu to maintain broad
19
+# compatibility with other Linux OSes. The main reason for this is that the
20
+# glibc version is baked into the distro, and binaries built with one glibc
21
+# version are not backwards compatible with OSes that use an earlier version.
22
+ARG BUILD_BASE_IMAGE=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu20.04
23
# TODO: Restore to base image after FlashInfer AOT wheel fixed
24
ARG FINAL_BASE_IMAGE=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04
25
@@ -75,34 +80,19 @@ ARG TARGETPLATFORM
80
ARG INSTALL_KV_CONNECTORS=false
81
ENV DEBIAN_FRONTEND=noninteractive
82
78
-ARG DEADSNAKES_MIRROR_URL
79
-ARG DEADSNAKES_GPGKEY_URL
83
ARG GET_PIP_URL
84
82
-# Install Python and other dependencies
85
+# Install system dependencies and uv, then create Python virtual environment
86
RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
87
&& echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections \
88
&& apt-get update -y \
86
- && apt-get install -y ccache software-properties-common git curl sudo \
87
- && if [ ! -z ${DEADSNAKES_MIRROR_URL} ] ; then \
88
- if [ ! -z "${DEADSNAKES_GPGKEY_URL}" ] ; then \
89
- mkdir -p -m 0755 /etc/apt/keyrings ; \
90
- curl -L ${DEADSNAKES_GPGKEY_URL} | gpg --dearmor > /etc/apt/keyrings/deadsnakes.gpg ; \
91
- sudo chmod 644 /etc/apt/keyrings/deadsnakes.gpg ; \
92
- echo "deb [signed-by=/etc/apt/keyrings/deadsnakes.gpg] ${DEADSNAKES_MIRROR_URL} $(lsb_release -cs) main" > /etc/apt/sources.list.d/deadsnakes.list ; \
93
- fi ; \
94
- else \
95
- for i in 1 2 3; do \
96
- add-apt-repository -y ppa:deadsnakes/ppa && break || \
97
- { echo "Attempt $i failed, retrying in 5s..."; sleep 5; }; \
98
- done ; \
99
- fi \
100
- && apt-get update -y \
101
- && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \
102
- && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
103
- && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \
104
- && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \
105
- && curl -sS ${GET_PIP_URL} | python${PYTHON_VERSION} \
89
+ && apt-get install -y ccache software-properties-common git curl sudo python3-pip \
90
+ && curl -LsSf https://astral.sh/uv/install.sh | sh \
91
+ && $HOME/.local/bin/uv venv /opt/venv --python ${PYTHON_VERSION} \
92
+ && rm -f /usr/bin/python3 /usr/bin/python3-config /usr/bin/pip \
93
+ && ln -s /opt/venv/bin/python3 /usr/bin/python3 \
94
+ && ln -s /opt/venv/bin/python3-config /usr/bin/python3-config \
95
+ && ln -s /opt/venv/bin/pip /usr/bin/pip \
96
&& python3 --version && python3 -m pip --version
97
98
ARG PIP_INDEX_URL UV_INDEX_URL
@@ -111,9 +101,9 @@ ARG PYTORCH_CUDA_INDEX_BASE_URL
101
ARG PYTORCH_CUDA_NIGHTLY_INDEX_BASE_URL
102
ARG PIP_KEYRING_PROVIDER UV_KEYRING_PROVIDER
103
114
-# Install uv for faster pip installs
115
-RUN --mount=type=cache,target=/root/.cache/uv \
116
- python3 -m pip install uv
104
+# Activate virtual environment and add uv to PATH
105
+ENV PATH="/opt/venv/bin:/root/.local/bin:$PATH"
106
+ENV VIRTUAL_ENV="/opt/venv"
107
108
# This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out
109
# Reference: https://github.com/astral-sh/uv/pull/1694
@@ -142,7 +132,7 @@ WORKDIR /workspace
132
COPY requirements/common.txt requirements/common.txt
133
COPY requirements/cuda.txt requirements/cuda.txt
134
RUN --mount=type=cache,target=/root/.cache/uv \
145
- uv pip install --system -r requirements/cuda.txt \
135
+ uv pip install --python /opt/venv/bin/python3 -r requirements/cuda.txt \
136
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
137
138
# cuda arch list used by torch
@@ -172,7 +162,7 @@ ENV UV_INDEX_STRATEGY="unsafe-best-match"
162
ENV UV_LINK_MODE=copy
163
164
RUN --mount=type=cache,target=/root/.cache/uv \
175
- uv pip install --system -r requirements/build.txt \
165
+ uv pip install --python /opt/venv/bin/python3 -r requirements/build.txt \
166
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
167
168
COPY . .
@@ -269,7 +259,7 @@ COPY requirements/lint.txt requirements/lint.txt
259
COPY requirements/test.txt requirements/test.txt
260
COPY requirements/dev.txt requirements/dev.txt
261
RUN --mount=type=cache,target=/root/.cache/uv \
272
- uv pip install --system -r requirements/dev.txt \
262
+ uv pip install --python /opt/venv/bin/python3 -r requirements/dev.txt \
263
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
264
#################### DEV IMAGE ####################
265
docs/assets/contributing/dockerfile-stages-dependency.png
Binary files a/docs/assets/contributing/dockerfile-stages-dependency.png and b/docs/assets/contributing/dockerfile-stages-dependency.png differ