diff --git a/examples/python/onnx_inference.py b/examples/python/onnx_inference.py
new file mode 100644
index 00000000..152fe37a
--- /dev/null
+++ b/examples/python/onnx_inference.py
@@ -0,0 +1,29 @@
+import sys
+import onnx
+import torch
+import numpy as np
+from pyinfinitensor.onnx import OnnxStub, backend
+
+if __name__ == '__main__':
+    args = sys.argv
+    if len(sys.argv) != 2:
+        print("Usage: python onnx_inference.py model_name.onnx")
+        exit()
+    model_path = sys.argv[1]
+    # print(model_path)
+
+    onnx_model = onnx.load(model_path)
+    onnx_input = onnx_model.graph.input[0]
+    input_shape = [[d.dim_value for d in _input.type.tensor_type.shape.dim]
+                   for _input in onnx_model.graph.input]
+    # Assume that there is only one input tensor
+    input_shape = input_shape[0]
+    # print(input_shape)
+    input_data = np.random.random(input_shape).astype(np.float32)
+
+    model = OnnxStub(onnx_model, backend.cuda_runtime())
+    next(iter(model.inputs.values())).copyin_numpy(input_data)
+    model.run()
+    outputs = next(iter(model.outputs.values())).copyout_numpy()
+    outputs = torch.tensor(outputs)
+    print(outputs.shape)
diff --git a/examples/python/resnet_inference.py b/examples/python/resnet_inference.py
new file mode 100644
index 00000000..4c58c7a6
--- /dev/null
+++ b/examples/python/resnet_inference.py
@@ -0,0 +1,24 @@
+import sys
+import onnx
+import torch
+import numpy as np
+from pyinfinitensor.onnx import OnnxStub, backend
+import torchvision.models as models
+
+if __name__ == '__main__':
+    model_path = './resnet18.onnx'
+    tv_model = models.resnet50(weights=None)
+    input_shape = (1, 3, 224, 224)
+    param = torch.rand(input_shape)
+    torch.onnx.export(tv_model, param, model_path, verbose=False)
+
+    onnx_model = onnx.load(model_path)
+    model = OnnxStub(onnx_model, backend.cuda_runtime())
+    images = np.random.random(input_shape).astype(np.float32)
+    next(iter(model.inputs.values())).copyin_numpy(images)
+    model.run()
+    outputs = next(iter(model.outputs.values())).copyout_numpy()
+    outputs = torch.tensor(outputs)
+    outputs = torch.reshape(outputs, (1, 1000))
+    _, predicted = torch.max(outputs, 1)
+    print(predicted)