# FlashAttention

An attention kernel that computes exact attention in tiles to avoid reading and writing the large attention matrix to GPU memory.

## What it is used for

Faster training and inference with long sequences.