AtMan: Understanding Transformer Predictions Through Memory Efficient Attention Manipulation

Open in new window