diff --git a/models/pytorch2onnx/ort_run_frozen_each_layer.py b/models/pytorch2onnx/ort_run_frozen_each_layer.py
index 8bcd8ada7..efdf2838e 100644
--- a/models/pytorch2onnx/ort_run_frozen_each_layer.py
+++ b/models/pytorch2onnx/ort_run_frozen_each_layer.py
@@ -98,11 +98,10 @@ def check_shape(shape):
 providers = args.provider.split(",")
 if "CPUExecutionProvider" not in providers:
     providers.append("CPUExecutionProvider")
+if 'CUDAExecutionProvider' in ort.get_available_providers() and 'CUDAExecutionProvider' not in providers:
+    providers = ['CUDAExecutionProvider'] + providers
 
-ort_session = ort.InferenceSession(args.file, sess_options, providers=providers)
-
-if args.provider != '':
-    ort_session.set_providers([args.provider])
+ort_session = ort.InferenceSession(model.SerializeToString(), sess_options, providers=providers)
 
 print("Execution Providers:", ort_session.get_providers())