0b0eabbfd972c9e3f6323bff9d39ac5fc3ba9cc7,transformer/Translator.py,Translator,translate_batch,#Translator#,52

Before Change

            // -- Preparing decoded data seq -- //
            // size: batch x beam x seq
            dec_partial_seq = torch.stack([
                b.get_current_state() for b in beams if not b.done])
            // size: (batch * beam) x seq
            dec_partial_seq = dec_partial_seq.view(-1, len_dec_seq)
            // wrap into a Variable
            dec_partial_seq = Variable(dec_partial_seq, volatile=True)

            // -- Preparing decoded pos seq -- //
            // size: 1 x seq
            dec_partial_pos = torch.arange(1, len_dec_seq + 1).unsqueeze(0)
            // size: (batch * beam) x seq
            dec_partial_pos = dec_partial_pos.repeat(n_remaining_sents * beam_size, 1)
            // wrap into a Variable
            dec_partial_pos = Variable(dec_partial_pos.type(torch.LongTensor), volatile=True)

            if self.opt.cuda:
                dec_partial_seq = dec_partial_seq.cuda()
                dec_partial_pos = dec_partial_pos.cuda()

            // -- Decoding -- //

After Change

            // -- Preparing decoded data seq -- //
            // size: batch x beam x seq
            dec_partial_seq = torch.stack([
                b.get_current_state() for b in beams if not b.done]).to(self.device)
            // size: (batch * beam) x seq
            dec_partial_seq = dec_partial_seq.view(-1, len_dec_seq)

            // -- Preparing decoded pos seq -- //
            // size: 1 x seq
            dec_partial_pos = torch.arange(1, len_dec_seq + 1, dtype=torch.long, device=self.device)
            // size: (batch * beam) x seq
            dec_partial_pos = dec_partial_pos.unsqueeze(0).repeat(n_remaining_sents * sz_beam, 1)

            if self.opt.cuda:
                dec_partial_seq = dec_partial_seq.cuda()
                dec_partial_pos = dec_partial_pos.cuda()

            // -- Decoding -- //

Instances