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
[Bug]: Set LD_LIBRARY_PATH to include the 'standard' CUDA location (#25766)
Signed-off-by: Clayton Coleman <smarterclayton@gmail.com>
Clayton Coleman committed
Sep 27, 2025 at 13:36 UTC
5546acb463243ce3c166dc620c764a93351b7c69
1 file changed
+6
docker/Dockerfile
+6
@@ -474,6 +474,12 @@ ENV CUDA_HOME=/usr/local/cuda
474
RUN export TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-9.0a+PTX}" \
475
&& bash install_python_libraries.sh
476
477
+# CUDA image changed from /usr/local/nvidia to /usr/local/cuda in 12.8 but will
478
+# return to /usr/local/nvidia in 13.0 to allow container providers to mount drivers
479
+# consistently from the host (see https://github.com/vllm-project/vllm/issues/18859).
480
+# Until then, add /usr/local/nvidia/lib64 before the image cuda path to allow override.
481
+ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib64:${LD_LIBRARY_PATH}
482
+
483
#################### vLLM installation IMAGE ####################
484
485
#################### TEST IMAGE ####################