Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
During pre-training, the authors shared the
relative position bias among the several self-attention layers.