First Attentions Last: Better Exploiting First Attentions for Efficient Transformer Training

Re-assign community