Fast and memory-efficient exact attention - View it on GitHub