Sparse Attention

Language Models Can Control Their Own Attention