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 3, 17:45 UTC
[PD] Fix broken NIXL EP installation (#42542)
Signed-off-by: Ovidiu Mara <ovidium@nvidia.com>
ovidiusm committed
May 13, 2026 at 22:55 UTC
cca32d55a296879b8624b957c20d3e4ca7e7cade
1 file changed
+2
-3
docker/Dockerfile
+2
-3
@@ -845,9 +845,6 @@ RUN --mount=type=cache,target=/root/.cache/uv \
845
libcublas-dev-${CUDA_VERSION_DASH} \
846
libcusolver-dev-${CUDA_VERSION_DASH}"; \
847
if [ "$INSTALL_KV_CONNECTORS" = "true" ]; then \
848
- if [ "$CUDA_MAJOR" -ge 13 ]; then \
849
- uv pip install --system nixl-cu13; \
850
- fi; \
848
uv pip install --system -r /tmp/kv_connectors.txt --no-build || ( \
849
# if the above fails, install from source
850
apt-get update -y && \
@@ -857,6 +854,8 @@ RUN --mount=type=cache,target=/root/.cache/uv \
854
# clean up -dev packages, keep runtime libraries
855
rm -rf /var/lib/apt/lists/* \
856
); \
857
+ # Force-reinstall the matching CUDA wheel so the correct nixl_ep_cpp.so is installed.
858
+ uv pip install --system --force-reinstall --no-deps nixl-cu${CUDA_MAJOR}; \
859
fi
860
861
ENV VLLM_USAGE_SOURCE production-docker-image