Pangu net ONNX does not work

[Jostarndt]

Hi there,

I am trying to convert a model from this paper https://www.nature.com/articles/s41586-023-06185-3 downloaded from this GitHub https://github.com/198808xc/Pangu-Weather

First of all: thanks a lot for onnx2torch.

However, there are some issues.

First issue

First in

onnx2pytorch/onnx2pytorch/operations/slice.py

Line 83 in d4f79e4

return torch.flip(data.__getitem__(selection), flip_dims)

:

return torch.flip(data.__getitem__(selection), flip_dims)

lead to the warning lib/python3.10/site-packages/onnx2pytorch/operations/slice.py:86: UserWarning: Using a non-tuple sequence for multidimensional indexing is deprecated and will be changed in pytorch 2.9; use x[tuple(seq)] instead of x[seq]. In pytorch 2.9 this will be interpreted as tensor index, x[torch.tensor(seq)], which will result either in an error or a different result (Triggered internally at /pytorch/torch/csrc/autograd/python_variable_indexing.cpp:309.) return torch.flip(data[selection], flip_dims)

Suggested Fix?

return torch.flip(data.__getitem__(tuple(selection)), flip_dims) fixed the issue. Also, even shorter

return torch.flip(data[selection], flip_dims). I hope these operations do what I think they are supposed to do.

Second issue

onnx2pytorch/onnx2pytorch/operations/pad.py

Line 17 in d4f79e4

out = F.pad(input, list(pads), mode=self.mode, value=value)

out = F.pad(input, list(pads), mode=self.mode, value=value)

give me the error *** TypeError: pad(): argument 'value' (position 4) must be float, not Tensor

And indeed value seems to be a tensor (in my case of shape torch.Size([5, 13, 721, 1440]). Fixing this is beyond my understanding of this library unfortunately. I would appreciate any help.