78eba7b3f82b8420deac3cd28318dbfead0f9b9e,python/baseline/pytorch/seq2seq/model.py,Seq2SeqModel,encode,#Seq2SeqModel#Any#Any#,160

Before Change

        embed_in_seq = self.embed(input)
        //if self.training:
        packed = torch.nn.utils.rnn.pack_padded_sequence(embed_in_seq, src_len.data.tolist())
        output_tbh, hidden = self.encoder_rnn(packed)
        output_tbh, _ = torch.nn.utils.rnn.pad_packed_sequence(output_tbh)
        //else:
        //    output_tbh, hidden = self.encoder_rnn(embed_in_seq)
        return output_tbh, hidden

    def decoder(self, context_tbh, h_i, output_i, dst, src_mask):
        embed_out_tbh = self.tgt_embedding(dst)
        context_bth = context_tbh.transpose(0, 1)

After Change

            lengths = lengths.cuda()
        example["src_len"] = lengths
        for key in self.src_embeddings.keys():
            tensor = torch.from_numpy(batch_dict[key])
            tensor = tensor[perm_idx]
            example[key] = tensor

            if self.gpu:
                example[key] = example[key].cuda()

        if "tgt" in batch_dict:
            tgt = torch.from_numpy(batch_dict["tgt"])
            example["dst"] = tgt[:, :-1]
            example["tgt"] = tgt[:, 1:]
            example["dst"] = example["dst"][perm_idx]
            example["tgt"] = example["tgt"][perm_idx]
            if self.gpu:

Instances