How to convert a pytorch pth file to an onxx model that supports any batch size,

How to convert a pytorch pth file to an onxx model that supports any batch size,how to set torch.randn function batch_size variables

I find method that use torch.onnx.export dynamic_axes variable

I search for a way that could realizations from the file endwith"pth" to the file endwith"onxx".but emmmm
:sob:

how did you using dynamic_axes ,i use the parammeter it raises " ‘please determine the input size manually by --input-shape xxx’.format(key))
RuntimeError: The shape of input “input” has dynamic size, please determine the input size manually by --input-shape xxx