Args: src_sents: Tensor [batch, src_length] tensor for source sentences tgt_sents: Tensor [batch, tgt_length] tensor for target sentences src_masks: Tensor [batch, src_length] or None tensor for source mask
(self, src_sents: torch.Tensor, tgt_sents: torch,
src_masks: torch.Tensor, tgt_masks: torch.Tensor)
source not stored for this graph (policy: none)