Sparse Attention

Large Language Models Can Control Their Own Attention Span