A2SF: Accumulative Attention Scoring with Forgetting Factor for Token Pruning in Transformer Decoder | Read Paper on Bytez