test

Author: lanluo-nvidia

.github/scripts/install-torch-tensorrt.sh

@@ -1,9 +1,23 @@
-set -exou pipefail
-
+#set -exou pipefail
+set -x
 TORCH_TORCHVISION=$(grep "^torch" ${PWD}/py/requirements.txt)
 INDEX_URL=https://download.pytorch.org/whl/${CHANNEL}/${CU_VERSION}
 PLATFORM=$(python -c "import sys; print(sys.platform)")
 
+if [[ $(uname -m) == "aarch64" ]]; then
+    # install cuda 12.8 only for aarch64 for now
+    VER="12-8"
+    dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/sbsa/cuda-rhel8.repo
+    dnf -y install cuda-compiler-${VER}.aarch64 \
+                cuda-libraries-${VER}.aarch64 \
+                cuda-libraries-devel-${VER}.aarch64
+    dnf clean all
+    export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
+    ls -lart /usr/local/
+    nvcc --version
+    echo "cuda installed successfully"
+fi
+
 # Install all the dependencies required for Torch-TensorRT
 pip install --pre ${TORCH_TORCHVISION} --index-url ${INDEX_URL}
 pip install --pre -r ${PWD}/tests/py/requirements.txt
@@ -12,17 +26,6 @@ pip install --pre -r ${PWD}/tests/py/requirements.txt
 if [[ ${PLATFORM} == win32 ]]; then
     pip install ${RUNNER_ARTIFACT_DIR}/torch_tensorrt*.whl
 else
-    if [[ $(uname -m) == "aarch64" ]]; then
-        # install cuda 12.8 only for aarch64 for now
-        VER="12-8"
-        dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/sbsa/cuda-rhel8.repo
-        dnf -y install cuda-compiler-${VER}.aarch64 \
-                    cuda-libraries-${VER}.aarch64 \
-                    cuda-libraries-devel-${VER}.aarch64
-        dnf clean all
-        export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
-        echo "cuda installed successfully"
-    fi
     pip install /opt/torch-tensorrt-builds/torch_tensorrt*.whl
 fi