Does pytorch-geometric support to_onnx? I have a pytorch lightning model that contains multiple torch.nn.Modules, one of which being a GNN using pytorch geometric message passing.
I am able to get lightning’s to_onnx method to run, but it breaks apart when I try to do inference (it also adds input_names other than the ones I defined and changes input dimensions in unexpected ways)
I saw an old github post saying ONNX is not supported by pytorch-geometric, just wondering if that is still true.