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
Always use cache mounts when installing vllm to avoid populating pip cache in the image. Also remove apt cache. (#23270)
Signed-off-by: Valentyn Tymofieiev <valentyn@google.com>
tvalentyn committed
Aug 22, 2025 at 00:01 UTC
8ef6b8a38cc3b69e55b0bb60d5442c4efb1988de
1 file changed
+8
-3
docker/Dockerfile.tpu
+8
-3
@@ -7,7 +7,8 @@ WORKDIR /workspace/vllm
7
# Install some basic utilities
8
RUN apt-get update && apt-get install -y \
9
git \
10
- ffmpeg libsm6 libxext6 libgl1
10
+ ffmpeg libsm6 libxext6 libgl1 && \
11
+ rm -rf /var/lib/apt/lists/*
12
13
# Build vLLM.
14
COPY . .
@@ -16,6 +17,9 @@ RUN --mount=type=bind,source=.git,target=.git \
17
if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh; fi
18
19
# Remove existing versions of dependencies
20
+# TODO: These packages will remain as dead weight in the Docker image layers.
21
+# We should find a way to build the image without uninstalling these.
22
+# Consider using a different base image.
23
RUN pip uninstall -y torch torch_xla torchvision
24
25
ENV VLLM_TARGET_DEVICE="tpu"
@@ -23,9 +27,10 @@ RUN --mount=type=cache,target=/root/.cache/pip \
27
--mount=type=bind,source=.git,target=.git \
28
python3 -m pip install \
29
-r requirements/tpu.txt
26
-RUN python3 -m pip install -e .
30
+
31
+RUN --mount=type=cache,target=/root/.cache/pip python3 -m pip install -e .
32
33
# install development dependencies (for testing)
29
-RUN python3 -m pip install -e tests/vllm_test_utils
34
+RUN --mount=type=cache,target=/root/.cache/pip python3 -m pip install -e tests/vllm_test_utils
35
36
CMD ["/bin/bash"]