pytorch

721a8ca6 - Sparse CSR: enable autograd for torch.sparse.addmm and torch.sparse.mm

Sparse CSR: enable autograd for torch.sparse.addmm and torch.sparse.mm This PR updates the derivative rule for `torch.sparse.addmm` to be working with CSR sparse matrix. Notably `torch.sparse.sampled_addmm` is used in the backward function. Pull Request resolved: https://github.com/pytorch/pytorch/pull/76591 Approved by: https://github.com/cpuhrsch