Having upgraded to 1.0, all my use of einsum has stopped working, giving me instead the runtime error: sparse tensors do not have strides. What’s going on here? I can’t see from the documentation that anything has changed in the einsum. What am I missing?
I encounter the same issue…
I think this likely is a fallout from matmul not working for sparse arguments. einsum (currently) transforms the inputs and calls torch.bmm.
Best regards
Thomas