Add libnvshmem installation (#8460)

Author: mc-nv

build.py

@@ -78,7 +78,7 @@
     "ort_openvino_version": "2025.3.0",
     "standalone_openvino_version": "2025.3.0",
     "dcgm_version": "4.4.0-1",
-    "vllm_version": "0.10.1.1",
+    "vllm_version": "0.10.2",
     "rhel_py_version": "3.12.3",
 }
 
@@ -660,6 +660,10 @@ def pytorch_cmake_args(images):
         cargs.append(
             cmake_backend_enable("pytorch", "TRITON_ENABLE_NVTX", FLAGS.enable_nvtx)
         )
+        if target_platform() == "igpu":
+            cargs.append(
+                cmake_backend_enable("pytorch", "TRITON_PYTORCH_NVSHMEM", False)
+            )
     return cargs
 
 
@@ -1514,6 +1518,22 @@ def dockerfile_prepare_container_linux(argmap, backends, enable_gpu, target_mach
 ENV PYTHONPATH=/opt/tritonserver/backends/dali/wheel/dali:$PYTHONPATH
 """
 
+    if target_platform() not in ["igpu", "windows", "rhel"]:
+        repo_arch = "sbsa" if target_machine == "aarch64" else "x86_64"
+        df += f"""
+RUN curl -o /tmp/cuda-keyring.deb \\
+        https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/{repo_arch}/cuda-keyring_1.1-1_all.deb \\
+      && apt install /tmp/cuda-keyring.deb \\
+      && rm /tmp/cuda-keyring.deb \\
+      && apt update -qq \\
+      && apt install --yes --no-install-recommends libnvshmem3-cuda-13 \\
+      && rm -rf /var/lib/apt/lists/* \\
+      && dpkg -L libnvshmem3-cuda-13 | grep libnvshmem_host.so | sed -e 's/libnvshmem_host.*//g' | sort -u > /etc/ld.so.conf.d/libnvshmem3-cuda-13.conf \\
+      && ldconfig
+""".format(
+            repo_arch=repo_arch
+        )
+
     df += """
 WORKDIR /opt/tritonserver
 RUN rm -fr /opt/tritonserver/*