7ffcccfa5c · tduncan/Nvidia-Optimization-Stack-VLLM

@@ -14,17 +14,22 @@ ARG PYTHON_VERSION=3.12

14 ARG TARGETPLATFORM

15 ENV DEBIAN_FRONTEND=noninteractive

16

17 -# Install minimal dependencies and uv

18 -RUN apt-get update -y \

19 - && apt-get install -y ccache git curl wget sudo \

20 - && curl -LsSf https://astral.sh/uv/install.sh | sh

21 -

22 -# Add uv to PATH

23 -ENV PATH="/root/.local/bin:$PATH"

24 -# Create venv with specified Python and activate by placing at the front of path

25 -ENV VIRTUAL_ENV="/opt/venv"

26 -RUN uv venv --python ${PYTHON_VERSION} --seed ${VIRTUAL_ENV}

27 -ENV PATH="$VIRTUAL_ENV/bin:$PATH"

17 +# Install Python and other dependencies

18 +RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \

19 + && echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections \

20 + && apt-get update -y \

21 + && apt-get install -y ccache software-properties-common git curl sudo \

22 + && add-apt-repository ppa:deadsnakes/ppa \

23 + && apt-get update -y \

24 + && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \

25 + && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \

26 + && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \

27 + && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \

28 + && curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} \

29 + && python3 --version && python3 -m pip --version

30 +# Install uv for faster pip installs

31 +RUN --mount=type=cache,target=/root/.cache/uv \

32 + python3 -m pip install uv

33

34 # This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out

35 # Reference: https://github.com/astral-sh/uv/pull/1694

@@ -46,20 +51,22 @@ RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/

51

52 WORKDIR /workspace

53

54 +# install build and runtime dependencies

55 +

56 # arm64 (GH200) build follows the practice of "use existing pytorch" build,

57 # we need to install torch and torchvision from the nightly builds first,

58 # pytorch will not appear as a vLLM dependency in all of the following steps

59 # after this step

60 RUN --mount=type=cache,target=/root/.cache/uv \

61 if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \

55 - uv pip install --index-url https://download.pytorch.org/whl/nightly/cu128 "torch==2.8.0.dev20250318+cu128" "torchvision==0.22.0.dev20250319"; \

56 - uv pip install --index-url https://download.pytorch.org/whl/nightly/cu128 --pre pytorch_triton==3.3.0+gitab727c40; \

62 + uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu128 "torch==2.8.0.dev20250318+cu128" "torchvision==0.22.0.dev20250319"; \

63 + uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu128 --pre pytorch_triton==3.3.0+gitab727c40; \

64 fi

65

66 COPY requirements/common.txt requirements/common.txt

67 COPY requirements/cuda.txt requirements/cuda.txt

68 RUN --mount=type=cache,target=/root/.cache/uv \

62 - uv pip install -r requirements/cuda.txt

69 + uv pip install --system -r requirements/cuda.txt

70

71 # cuda arch list used by torch

72 # can be useful for both `dev` and `test`

@@ -84,7 +91,7 @@ COPY requirements/build.txt requirements/build.txt

91 ENV UV_HTTP_TIMEOUT=500

92

93 RUN --mount=type=cache,target=/root/.cache/uv \

87 - uv pip install -r requirements/build.txt

94 + uv pip install --system -r requirements/build.txt

95

96 COPY . .

97 ARG GIT_REPO_CHECK=0

@@ -156,7 +163,7 @@ COPY requirements/lint.txt requirements/lint.txt

163 COPY requirements/test.txt requirements/test.txt

164 COPY requirements/dev.txt requirements/dev.txt

165 RUN --mount=type=cache,target=/root/.cache/uv \

159 - uv pip install -r requirements/dev.txt

166 + uv pip install --system -r requirements/dev.txt

167 #################### DEV IMAGE ####################

168

169 #################### vLLM installation IMAGE ####################

@@ -172,18 +179,23 @@ ARG TARGETPLATFORM

179 RUN PYTHON_VERSION_STR=$(echo ${PYTHON_VERSION} | sed 's/\.//g') && \

180 echo "export PYTHON_VERSION_STR=${PYTHON_VERSION_STR}" >> /etc/environment

181

175 -# Install minimal dependencies and uv

176 -RUN apt-get update -y \

177 - && apt-get install -y ccache git curl wget sudo vim \

178 - && apt-get install -y ffmpeg libsm6 libxext6 libgl1 libibverbs-dev \

179 - && curl -LsSf https://astral.sh/uv/install.sh | sh

180 -

181 -# Add uv to PATH

182 -ENV PATH="/root/.local/bin:$PATH"

183 -# Create venv with specified Python and activate by placing at the front of path

184 -ENV VIRTUAL_ENV="/opt/venv"

185 -RUN uv venv --python ${PYTHON_VERSION} --seed ${VIRTUAL_ENV}

186 -ENV PATH="$VIRTUAL_ENV/bin:$PATH"

182 +# Install Python and other dependencies

183 +RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \

184 + && echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections \

185 + && apt-get update -y \

186 + && apt-get install -y ccache software-properties-common git curl wget sudo vim python3-pip \

187 + && apt-get install -y ffmpeg libsm6 libxext6 libgl1 \

188 + && add-apt-repository ppa:deadsnakes/ppa \

189 + && apt-get update -y \

190 + && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv libibverbs-dev \

191 + && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \

192 + && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \

193 + && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \

194 + && curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} \

195 + && python3 --version && python3 -m pip --version

196 +# Install uv for faster pip installs

197 +RUN --mount=type=cache,target=/root/.cache/uv \

198 + python3 -m pip install uv

199

200 # This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out

201 # Reference: https://github.com/astral-sh/uv/pull/1694

@@ -201,14 +213,14 @@ RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/

213 # after this step

214 RUN --mount=type=cache,target=/root/.cache/uv \

215 if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \

204 - uv pip install --index-url https://download.pytorch.org/whl/nightly/cu128 "torch==2.8.0.dev20250318+cu128" "torchvision==0.22.0.dev20250319"; \

205 - uv pip install --index-url https://download.pytorch.org/whl/nightly/cu128 --pre pytorch_triton==3.3.0+gitab727c40; \

216 + uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu128 "torch==2.8.0.dev20250318+cu128" "torchvision==0.22.0.dev20250319"; \

217 + uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu128 --pre pytorch_triton==3.3.0+gitab727c40; \

218 fi

219

220 # Install vllm wheel first, so that torch etc will be installed.

221 RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist \

222 --mount=type=cache,target=/root/.cache/uv \

211 - uv pip install dist/*.whl --verbose

223 + uv pip install --system dist/*.whl --verbose

224

225 # If we need to build FlashInfer wheel before its release:

226 # $ export FLASHINFER_ENABLE_AOT=1

@@ -223,8 +235,9 @@ RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist

235 # $ # upload the wheel to a public location, e.g. https://wheels.vllm.ai/flashinfer/524304395bd1d8cd7d07db083859523fcaa246a4/flashinfer_python-0.2.1.post1+cu124torch2.5-cp38-abi3-linux_x86_64.whl

236

237 RUN --mount=type=cache,target=/root/.cache/uv \

238 +. /etc/environment && \

239 if [ "$TARGETPLATFORM" != "linux/arm64" ]; then \

227 - uv pip install https://github.com/flashinfer-ai/flashinfer/releases/download/v0.2.1.post2/flashinfer_python-0.2.1.post2+cu124torch2.6-cp38-abi3-linux_x86_64.whl ; \

240 + uv pip install --system https://github.com/flashinfer-ai/flashinfer/releases/download/v0.2.1.post2/flashinfer_python-0.2.1.post2+cu124torch2.6-cp38-abi3-linux_x86_64.whl ; \

241 fi

242 COPY examples examples

243

@@ -234,7 +247,7 @@ COPY examples examples

247 # TODO: Remove this once FlashInfer AOT wheel is fixed

248 COPY requirements/build.txt requirements/build.txt

249 RUN --mount=type=cache,target=/root/.cache/uv \

237 - uv pip install -r requirements/build.txt

250 + uv pip install --system -r requirements/build.txt

251

252 #################### vLLM installation IMAGE ####################

253

@@ -251,15 +264,15 @@ ENV UV_HTTP_TIMEOUT=500

264

265 # install development dependencies (for testing)

266 RUN --mount=type=cache,target=/root/.cache/uv \

254 - uv pip install -r requirements/dev.txt

267 + uv pip install --system -r requirements/dev.txt

268

269 # install development dependencies (for testing)

270 RUN --mount=type=cache,target=/root/.cache/uv \

258 - uv pip install -e tests/vllm_test_utils

271 + uv pip install --system -e tests/vllm_test_utils

272

273 # enable fast downloads from hf (for testing)

274 RUN --mount=type=cache,target=/root/.cache/uv \

262 - uv pip install hf_transfer

275 + uv pip install --system hf_transfer

276 ENV HF_HUB_ENABLE_HF_TRANSFER 1

277

278 # Copy in the v1 package for testing (it isn't distributed yet)

@@ -284,9 +297,9 @@ ENV UV_HTTP_TIMEOUT=500

297 # install additional dependencies for openai api server

298 RUN --mount=type=cache,target=/root/.cache/uv \

299 if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \

287 - uv pip install accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.42.0' 'timm==0.9.10' boto3 runai-model-streamer runai-model-streamer[s3]; \

300 + uv pip install --system accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.42.0' 'timm==0.9.10' boto3 runai-model-streamer runai-model-streamer[s3]; \

301 else \

289 - uv pip install accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.45.3' 'timm==0.9.10' boto3 runai-model-streamer runai-model-streamer[s3]; \

302 + uv pip install --system accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.45.3' 'timm==0.9.10' boto3 runai-model-streamer runai-model-streamer[s3]; \

303 fi

304

305 ENV VLLM_USAGE_SOURCE production-docker-image

Revert "[CI/Build] Use uv python for docker rather than ppa:deadsnakess/ppa (#13569)" (#15377)