Fetching the paper…

AtMan: Understanding Transformer Predictions Through Memory Efficient Attention Manipulation · Around