rusty1s
commented
1 year ago How does your model look like? Note that you need to add type hints to the forward function, e.g. def forward(x: Tensor, adj: SparseTensor) as otherwise every argument is considered to be a PyTorch tensor.
🚀 The feature, motivation and pitch
I am trying to deploy a pyg model with libtorch and use torch script to trace the model. I mainly use GINConv. When the index is in coo format, I can manage to do that. But I want to use spmm to speedup the computation, so I use SparseTensor, but it is not supported yet.
Alternatives
No response
Additional context
No response