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, 01:00 UTC
Add Dockerfile argument for VLLM_USE_PRECOMPILED environment (#20943)
Signed-off-by: dougbtv <dosmith@redhat.com>
Doug Smith committed
Jul 15, 2025 at 22:53 UTC
79764460152e8022e4bfcdcf55f80180225f5077
1 file changed
+13
docker/Dockerfile
+13
@@ -207,6 +207,19 @@ ARG SCCACHE_ENDPOINT
207
ARG SCCACHE_BUCKET_NAME=vllm-build-sccache
208
ARG SCCACHE_REGION_NAME=us-west-2
209
ARG SCCACHE_S3_NO_CREDENTIALS=0
210
+
211
+# Flag to control whether to use pre-built vLLM wheels
212
+ARG VLLM_USE_PRECOMPILED
213
+# TODO: in setup.py VLLM_USE_PRECOMPILED is sensitive to truthiness, it will take =0 as "true", this should be fixed
214
+ENV VLLM_USE_PRECOMPILED=""
215
+RUN if [ "${VLLM_USE_PRECOMPILED}" = "1" ]; then \
216
+ export VLLM_USE_PRECOMPILED=1 && \
217
+ echo "Using precompiled wheels"; \
218
+ else \
219
+ unset VLLM_USE_PRECOMPILED && \
220
+ echo "Leaving VLLM_USE_PRECOMPILED unset to build wheels from source"; \
221
+ fi
222
+
223
# if USE_SCCACHE is set, use sccache to speed up compilation
224
RUN --mount=type=cache,target=/root/.cache/uv \
225
--mount=type=bind,source=.git,target=.git \