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 10, 05:15 UTC
[Improvement] Persist CUDA compat libraries paths to prevent reset on `apt-get` (#30784)
Signed-off-by: emricksini-h <emrick.birivoutin@hcompany.ai>
emricksini-h committed
Jan 13, 2026 at 23:35 UTC
2a60ac91d0f5c24cdb2863b178d2f5405fae50b8
1 file changed
+4
-4
docker/Dockerfile
+4
-4
@@ -117,8 +117,8 @@ ENV UV_LINK_MODE=copy
117
# Verify GCC version
118
RUN gcc --version
119
120
-# Workaround for triton/pytorch issues
121
-RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/
120
+# Ensure CUDA compatibility library is loaded
121
+RUN echo "/usr/local/cuda-$(echo "$CUDA_VERSION" | cut -d. -f1,2)/compat/" > /etc/ld.so.conf.d/00-cuda-compat.conf && ldconfig
122
123
# ============================================================
124
# SLOW-CHANGING DEPENDENCIES BELOW
@@ -453,8 +453,8 @@ ENV UV_HTTP_TIMEOUT=500
453
ENV UV_INDEX_STRATEGY="unsafe-best-match"
454
ENV UV_LINK_MODE=copy
455
456
-# Workaround for triton/pytorch issues
457
-RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/
456
+# Ensure CUDA compatibility library is loaded
457
+RUN echo "/usr/local/cuda-$(echo "$CUDA_VERSION" | cut -d. -f1,2)/compat/" > /etc/ld.so.conf.d/00-cuda-compat.conf && ldconfig
458
459
# ============================================================
460
# SLOW-CHANGING DEPENDENCIES BELOW