Fetching the paper…

How Sparse Attention Approximates Exact Attention? Your Attention is Naturally $n^C$-Sparse · Around