torch.Tensor.to_sparse_bsr — PyTorch 2.7 documentation (original) (raw)

Tensor.to_sparse_bsr(blocksize, dense_dim) → Tensor¶

Convert a tensor to a block sparse row (BSR) storage format of given blocksize. If the self is strided, then the number of dense dimensions could be specified, and a hybrid BSR tensor will be created, with dense_dim dense dimensions and self.dim() - 2 - dense_dim batch dimension.

Parameters

• blocksize (list, tuple, torch.Size, optional) – Block size of the resulting BSR tensor. A block size must be a tuple of length two such that its items evenly divide the two sparse dimensions.
• dense_dim (int, optional) – Number of dense dimensions of the resulting BSR tensor. This argument should be used only ifself is a strided tensor, and must be a value between 0 and dimension of self tensor minus two.

Example:

dense = torch.randn(10, 10) sparse = dense.to_sparse_csr() sparse_bsr = sparse.to_sparse_bsr((5, 5)) sparse_bsr.col_indices() tensor([0, 1, 0, 1])

dense = torch.zeros(4, 3, 1) dense[0:2, 0] = dense[0:2, 2] = dense[2:4, 1] = 1 dense.to_sparse_bsr((2, 1), 1) tensor(crow_indices=tensor([0, 2, 3]), col_indices=tensor([0, 2, 1]), values=tensor([[[[1.]],

                   [[1.]]],


                  [[[1.]],

                   [[1.]]],


                  [[[1.]],

                   [[1.]]]]), size=(4, 3, 1), nnz=3,
   layout=torch.sparse_bsr)