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 15, 22:15 UTC
[CI/Build] fix uv caching in Dockerfile (#13611)
Daniele committed
Feb 22, 2025 at 17:25 UTC
78ac0f591d8cde541ee74f2a0a8827d3769e2b86
1 file changed
+16
-16
Dockerfile
+16
-16
@@ -28,7 +28,7 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
28
&& curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} \
29
&& python3 --version && python3 -m pip --version
30
# Install uv for faster pip installs
31
-RUN --mount=type=cache,target=/root/.cache/pip \
31
+RUN --mount=type=cache,target=/root/.cache/uv \
32
python3 -m pip install uv
33
34
# Upgrade to GCC 10 to avoid https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92519
@@ -53,14 +53,14 @@ WORKDIR /workspace
53
# we need to install torch and torchvision from the nightly builds first,
54
# pytorch will not appear as a vLLM dependency in all of the following steps
55
# after this step
56
-RUN --mount=type=cache,target=/root/.cache/pip \
56
+RUN --mount=type=cache,target=/root/.cache/uv \
57
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
58
uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu126 "torch==2.7.0.dev20250121+cu126" "torchvision==0.22.0.dev20250121"; \
59
fi
60
61
COPY requirements-common.txt requirements-common.txt
62
COPY requirements-cuda.txt requirements-cuda.txt
63
-RUN --mount=type=cache,target=/root/.cache/pip \
63
+RUN --mount=type=cache,target=/root/.cache/uv \
64
uv pip install --system -r requirements-cuda.txt
65
66
# cuda arch list used by torch
@@ -81,7 +81,7 @@ ARG TARGETPLATFORM
81
# install build dependencies
82
COPY requirements-build.txt requirements-build.txt
83
84
-RUN --mount=type=cache,target=/root/.cache/pip \
84
+RUN --mount=type=cache,target=/root/.cache/uv \
85
uv pip install --system -r requirements-build.txt
86
87
COPY . .
@@ -101,7 +101,7 @@ ARG SCCACHE_BUCKET_NAME=vllm-build-sccache
101
ARG SCCACHE_REGION_NAME=us-west-2
102
ARG SCCACHE_S3_NO_CREDENTIALS=0
103
# if USE_SCCACHE is set, use sccache to speed up compilation
104
-RUN --mount=type=cache,target=/root/.cache/pip \
104
+RUN --mount=type=cache,target=/root/.cache/uv \
105
--mount=type=bind,source=.git,target=.git \
106
if [ "$USE_SCCACHE" = "1" ]; then \
107
echo "Installing sccache..." \
@@ -121,7 +121,7 @@ RUN --mount=type=cache,target=/root/.cache/pip \
121
122
ENV CCACHE_DIR=/root/.cache/ccache
123
RUN --mount=type=cache,target=/root/.cache/ccache \
124
- --mount=type=cache,target=/root/.cache/pip \
124
+ --mount=type=cache,target=/root/.cache/uv \
125
--mount=type=bind,source=.git,target=.git \
126
if [ "$USE_SCCACHE" != "1" ]; then \
127
python3 setup.py bdist_wheel --dist-dir=dist --py-limited-api=cp38; \
@@ -146,7 +146,7 @@ FROM base as dev
146
COPY requirements-lint.txt requirements-lint.txt
147
COPY requirements-test.txt requirements-test.txt
148
COPY requirements-dev.txt requirements-dev.txt
149
-RUN --mount=type=cache,target=/root/.cache/pip \
149
+RUN --mount=type=cache,target=/root/.cache/uv \
150
uv pip install --system -r requirements-dev.txt
151
#################### DEV IMAGE ####################
152
@@ -178,7 +178,7 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
178
&& curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} \
179
&& python3 --version && python3 -m pip --version
180
# Install uv for faster pip installs
181
-RUN --mount=type=cache,target=/root/.cache/pip \
181
+RUN --mount=type=cache,target=/root/.cache/uv \
182
python3 -m pip install uv
183
184
# Workaround for https://github.com/openai/triton/issues/2507 and
@@ -191,14 +191,14 @@ RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/
191
# we need to install torch and torchvision from the nightly builds first,
192
# pytorch will not appear as a vLLM dependency in all of the following steps
193
# after this step
194
-RUN --mount=type=cache,target=/root/.cache/pip \
194
+RUN --mount=type=cache,target=/root/.cache/uv \
195
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
196
uv pip install --system --index-url https://download.pytorch.org/whl/nightly/cu124 "torch==2.6.0.dev20241210+cu124" "torchvision==0.22.0.dev20241215"; \
197
fi
198
199
# Install vllm wheel first, so that torch etc will be installed.
200
RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist \
201
- --mount=type=cache,target=/root/.cache/pip \
201
+ --mount=type=cache,target=/root/.cache/uv \
202
uv pip install --system dist/*.whl --verbose
203
204
# If we need to build FlashInfer wheel before its release:
@@ -213,7 +213,7 @@ RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist
213
# $ ls dist
214
# $ # 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
215
216
-RUN --mount=type=cache,target=/root/.cache/pip \
216
+RUN --mount=type=cache,target=/root/.cache/uv \
217
. /etc/environment && \
218
if [ "$TARGETPLATFORM" != "linux/arm64" ]; then \
219
uv pip install --system https://github.com/flashinfer-ai/flashinfer/releases/download/v0.2.1.post1/flashinfer_python-0.2.1.post1+cu124torch2.5-cp38-abi3-linux_x86_64.whl ; \
@@ -225,7 +225,7 @@ COPY examples examples
225
# install build dependencies for JIT compilation.
226
# TODO: Remove this once FlashInfer AOT wheel is fixed
227
COPY requirements-build.txt requirements-build.txt
228
-RUN --mount=type=cache,target=/root/.cache/pip \
228
+RUN --mount=type=cache,target=/root/.cache/uv \
229
uv pip install --system -r requirements-build.txt
230
231
#################### vLLM installation IMAGE ####################
@@ -238,15 +238,15 @@ FROM vllm-base AS test
238
ADD . /vllm-workspace/
239
240
# install development dependencies (for testing)
241
-RUN --mount=type=cache,target=/root/.cache/pip \
241
+RUN --mount=type=cache,target=/root/.cache/uv \
242
uv pip install --system -r requirements-dev.txt
243
244
# install development dependencies (for testing)
245
-RUN --mount=type=cache,target=/root/.cache/pip \
245
+RUN --mount=type=cache,target=/root/.cache/uv \
246
uv pip install --system -e tests/vllm_test_utils
247
248
# enable fast downloads from hf (for testing)
249
-RUN --mount=type=cache,target=/root/.cache/pip \
249
+RUN --mount=type=cache,target=/root/.cache/uv \
250
uv pip install --system hf_transfer
251
ENV HF_HUB_ENABLE_HF_TRANSFER 1
252
@@ -266,7 +266,7 @@ RUN mv vllm test_docs/
266
FROM vllm-base AS vllm-openai-base
267
268
# install additional dependencies for openai api server
269
-RUN --mount=type=cache,target=/root/.cache/pip \
269
+RUN --mount=type=cache,target=/root/.cache/uv \
270
if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
271
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]; \
272
else \