Raspberry Pi RuntimeError: Cannot insert a Tensor that requires grad as a constant

I have this script convert Pytorch to ONNX format work well on my Ubuntu laptop. But I ran on Raspberry Pi 4 Buster 32 bit and got this error. I tried this thread and this thread but without luck

pi@raspberrypi:~/thesis $ python3 test_export.py 
Traceback (most recent call last):
  File "test_export.py", line 34, in <module>
    output_names=output_names)
  File "/usr/local/lib/python3.7/dist-packages/torch/onnx/__init__.py", line 230, in export
    custom_opsets, enable_onnx_checker, use_external_data_format)
  File "/usr/local/lib/python3.7/dist-packages/torch/onnx/utils.py", line 91, in export
    use_external_data_format=use_external_data_format)
  File "/usr/local/lib/python3.7/dist-packages/torch/onnx/utils.py", line 639, in _export
    dynamic_axes=dynamic_axes)
  File "/usr/local/lib/python3.7/dist-packages/torch/onnx/utils.py", line 411, in _model_to_graph
    use_new_jit_passes)
  File "/usr/local/lib/python3.7/dist-packages/torch/onnx/utils.py", line 379, in _create_jit_graph
    graph, torch_out = _trace_and_get_graph_from_model(model, args)
  File "/usr/local/lib/python3.7/dist-packages/torch/onnx/utils.py", line 342, in _trace_and_get_graph_from_model
    torch.jit._get_trace_graph(model, args, strict=False, _force_outplace=False, _return_inputs_states=True)
  File "/usr/local/lib/python3.7/dist-packages/torch/jit/_trace.py", line 1148, in _get_trace_graph
    outs = ONNXTracedModule(f, strict, _force_outplace, return_inputs, _return_inputs_states)(*args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/torch/jit/_trace.py", line 130, in forward
    self._force_outplace,
  File "/usr/local/lib/python3.7/dist-packages/torch/jit/_trace.py", line 112, in wrapper
    tuple(x.clone(memory_format=torch.preserve_format) for x in args)
  File "/usr/local/lib/python3.7/dist-packages/torch/jit/_trace.py", line 112, in <genexpr>
    tuple(x.clone(memory_format=torch.preserve_format) for x in args)
RuntimeError: Cannot insert a Tensor that requires grad as a constant. Consider making it a parameter or input, or detaching the gradient
Tensor:
(1,1,.,.) = 
  0.0133  0.0147 -0.0154 -0.0230 -0.0409 -0.0430 -0.0708
  0.0041  0.0058  0.0149  0.0206  0.0022 -0.0209 -0.0385
  0.0223  0.0236  0.0161  0.0588  0.1028  0.0626  0.0520
  0.0232  0.0042 -0.0459 -0.0487 -0.0164  0.0402  0.0658
 -0.0009  0.0278 -0.0101 -0.0554 -0.1272 -0.0766  0.0078
  0.0036  0.0480  0.0621  0.0844  0.0243 -0.0337 -0.0157
 -0.0800 -0.0322 -0.0178  0.0342  0.0354  0.0224  0.0017
...
[ torch.FloatTensor{64,3,7,7} ]

My code:

/usr/bin/python3
import os
import torch
import torchvision.models as models
from pathlib import Path

model = models.resnet50(pretrained=True)
#model.eval()

input_names=["actual_input"]
output_names=["output"]

onnx_path="output/resnet50.onnx"
output_path="output"
model_path=Path(output_path).with_suffix(".pth")
ir_path=model_path.with_suffix(".xml")

with torch.no_grad():

        dummy_input=None
        dummy_input=torch.randn(1,3,224,224)
        dummy_input.requires_grad = False

#Export Pytorch to ONNX
        torch.onnx.export(model,
        dummy_input,
        onnx_path,
        opset_version=10,
        verbose=False,
        input_names=input_names,
        output_names=output_names)
#       export_params=True)