fix: add missing nvlink binary to vLLM and TRT-LLM runtime/dev/local-dev Dockerfiles (#6503)

Signed-off-by: Keiven Chang <keivenchang@users.noreply.github.com>
Co-authored-by: Keiven Chang <keivenchang@users.noreply.github.com>
Co-authored-by: Cursor <cursoragent@cursor.com>

container/templates/sglang_runtime.Dockerfile

@@ -9,6 +9,9 @@
 
 FROM ${RUNTIME_IMAGE}:${RUNTIME_IMAGE_TAG} AS runtime
 
+# NOTE: Unlike vLLM/TRTLLM, the SGLang upstream runtime image already ships with the full CUDA
+# toolkit (nvcc, nvlink, ptxas, etc.), so no selective COPY of CUDA binaries is needed here.
+
 # cleanup unnecessary libs (python3-blinker conflicts with pip-installed blinker from Flask/dash)
 RUN apt remove -y python3-apt python3-blinker && \
     pip uninstall -y termplotlib

container/templates/trtllm_runtime.Dockerfile

@@ -34,6 +34,7 @@ ENV OMPI_MCA_coll_ucc_enable=0
 
 # Copy CUDA development tools (nvcc, headers, dependencies, etc.) from PyTorch base image
 COPY --from=pytorch_base /usr/local/cuda/bin/nvcc /usr/local/cuda/bin/nvcc
+COPY --from=pytorch_base /usr/local/cuda/bin/nvlink /usr/local/cuda/bin/nvlink
 COPY --from=pytorch_base /usr/local/cuda/bin/cudafe++ /usr/local/cuda/bin/cudafe++
 COPY --from=pytorch_base /usr/local/cuda/bin/ptxas /usr/local/cuda/bin/ptxas
 COPY --from=pytorch_base /usr/local/cuda/bin/fatbinary /usr/local/cuda/bin/fatbinary

container/templates/vllm_runtime.Dockerfile

@@ -34,6 +34,7 @@ ENV CUDA_DEVICE_ORDER=PCI_BUS_ID
 
 # Copy CUDA development tools (nvcc, headers, dependencies, etc.) from base devel image
 COPY --from=dynamo_base /usr/local/cuda/bin/nvcc /usr/local/cuda/bin/nvcc
+COPY --from=dynamo_base /usr/local/cuda/bin/nvlink /usr/local/cuda/bin/nvlink
 COPY --from=dynamo_base /usr/local/cuda/bin/cudafe++ /usr/local/cuda/bin/cudafe++
 COPY --from=dynamo_base /usr/local/cuda/bin/ptxas /usr/local/cuda/bin/ptxas
 COPY --from=dynamo_base /usr/local/cuda/bin/fatbinary /usr/local/cuda/bin/fatbinary