Hi, I have a question about the format when exporting model to onnx.
Pytorch version: 1.13.1
I use the example at [torch.onnx — PyTorch master documentation](torch.onnx — PyTorch master documentation)(I use resnet18 instead of alexnet). Then I visualize the resnet18.onnx file. Besides, I test this example using torch==1.10.0 and get the following result:
The left is the result of torch 1.13.1 and the right is torch 1.10.0.
This looks so strange, why the bias of Conv2d is not contained in Conv node?
It looks like that the result when using torch 1.10.0 is normal. What is the difference between torch1.10.0 and 1.13.1?