Sparse Attention Mechanisms

Block Sparse Attention

Sparse attention approach organized in blocks, where the attention matrix is divided into blocks and only certain blocks are computed to optimize hardware parallelism.

← Quay lại