Sparse Compressed Tensors represents a class of sparse tensors thathave a common feature of compressing the indices of a certain dimensionusing an encoding that enables certain optimizations on linear algebrakernels of sparse compressed tensors. This encoding is based on theCompressed Sparse Row (CSR) format that PyTorch sparse compressedtensors extend with the support of sparse tensor batches, allowingmulti-dimensional tensor values, and storing sparse tensor values indense blocks.