Disable runtime error before final validation

Author: atalman

.github/workflows/validate-linux-binaries.yml

@@ -55,12 +55,14 @@ jobs:
         export ENV_NAME="conda-env-${{ github.run_id }}"
         export TARGET_OS="linux"
         eval "$(conda shell.bash hook)"
-        source ./.github/scripts/validate_binaries.sh
 
         # Special case PyPi installation package. And Install of PyPi package via poetry
         if [[ ${MATRIX_PACKAGE_TYPE} == "manywheel" ]] && \
            ([[ ${MATRIX_GPU_ARCH_VERSION} == "12.1" && ${MATRIX_CHANNEL} != "release" ]] || \
             [[ ${MATRIX_GPU_ARCH_VERSION} == "11.7" && ${MATRIX_CHANNEL} == "release" ]]); then
-          source ./.github/scripts/validate_pipy.sh
-          source ./.github/scripts/validate_poetry.sh
+          source ./.github/scripts/validate_pipy.sh --runtime-error-check disabled
+          source ./.github/scripts/validate_poetry.sh --runtime-error-check disabled
         fi
+
+        # Standart case: Validate binaries
+        source ./.github/scripts/validate_binaries.sh

test/smoke_test/smoke_test.py

@@ -102,7 +102,7 @@ def test_cuda_runtime_errors_captured() -> None:
     if(cuda_exception_missed):
         raise RuntimeError( f"Expected CUDA RuntimeError but have not received!")
 
-def smoke_test_cuda(package: str) -> None:
+def smoke_test_cuda(package: str, runtime_error_check: str) -> None:
     if not torch.cuda.is_available() and is_cuda_system:
         raise RuntimeError(f"Expected CUDA {gpu_arch_ver}. However CUDA is not loaded.")
 
@@ -132,7 +132,8 @@ def smoke_test_cuda(package: str) -> None:
         if (sys.platform == "linux" or sys.platform == "linux2") and sys.version_info < (3, 11, 0):
             smoke_test_compile()
 
-        test_cuda_runtime_errors_captured()
+        if(runtime_error_check == "enabled")
+            test_cuda_runtime_errors_captured()
 
 
 def smoke_test_conv2d() -> None:
@@ -231,6 +232,13 @@ def main() -> None:
         choices=["all", "torchonly"],
         default="all",
     )
+    parser.add_argument(
+        "--runtime-error-check",
+        help="No Runtime Error check",
+        type=str,
+        choices=["enabled", "disabled"],
+        default="enabled",
+    )
     options = parser.parse_args()
     print(f"torch: {torch.__version__}")
     check_version(options.package)
@@ -240,7 +248,7 @@ def main() -> None:
     if options.package == "all":
         smoke_test_modules()
 
-    smoke_test_cuda(options.package)
+    smoke_test_cuda(options.package, options.runtime_error_check)
 
 
 if __name__ == "__main__":