输出每一步的损失
# pytorch中如何处理RNN输入变长序列padding:https://zhuanlan.zhihu.com/p/34418001?edition=yidianzixun&utm_source=yidianzixun&yidian_docid=0IVwLf60for i, (imgs, caps, caplens) in enumerate(train_loader): ... scores = pack_padded_sequence(scores, de