Fetching the paper…

A2SF: Accumulative Attention Scoring with Forgetting Factor for Token Pruning in Transformer Decoder · Around