Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
This means that for training, we always need an input sequence and a corresponding target sequence.