I’m trying to run a model which consists of multiple graph convolutional networks and a recurrent neural network.
I have converted the inputs of my graph convoulutional networks (i.e., feature matrix and adjacecny matrix) to torch.sparse.FloatTensor. I only use torch.sparse.mm() in the GCN and the loss function is BCEWithLogitsLoss().
In the forward pass everything works fine but as soon as loss.backward() is executed the below error is raised:
RuntimeError: sparse_.is_sparse() INTERNAL ASSERT FAILED at /pytorch/torch/csrc/autograd/generated/Functions.cpp:541, please report a bug to PyTorch.
Can someone help me with that error? thanks.
Could you post a code snippet with random inputs to reproduce this issue, please?