I recently switched to pytorch nightly from stable version and initialising an empty sparse tensor, which worked fine previously, now throws an error:
a = torch.sparse.FloatTensor(torch.empty((0), dtype=torch.long), torch.empty((0), dtype=torch.float), torch.Size([1, 5, 30, 20]))
Error:
RuntimeError: indices must be sparse_dim x nnz, but got: [0]