A ‘Simple Trick’ for Reducing Transformers’ (Self-)Attention Memory Requirements | Hacker News Reader