https://arxiv.org/abs/2205.14135
About
A conference paper on computing exact attention with less memory traffic.