8a70c277c355fe734a01fe58baafdc2dc5164205,OpenNMT/onmt/Models.py,Decoder,forward,#Decoder#,91

Before Change

        output = Variable(emb.data.new(batch_size, self.hidden_size).zero_(),
                          requires_grad=False)
        h_0 = Variable(emb.data.new(batch_size, self.hidden_size),
                       requires_grad=False)
        c_0 = Variable(emb.data.new(batch_size, self.hidden_size),
                       requires_grad=False)
        hidden = (h_0, c_0)

        outputs = []
        for emb_t in emb.chunk(emb.size(0)):
            emb_t = emb_t.squeeze(0)
            if self.input_feed:
                emb_t = torch.cat([emb_t, output], 1)

            // FIXME: multilayer
            hidden = self.rnn(emb_t, hidden)
            output = hidden[0]
            output = self.attn(output, context.t())
            output = self.dropout(output)
            outputs += [output]

        outputs = torch.cat(outputs)
        pred = self.generator(outputs)
        return pred

After Change

        output = Variable(emb.data.new(batch_size, self.hidden_size).zero_(),
                          requires_grad=False)
        h_0 = Variable(emb.data.new(batch_size, self.hidden_size).zero_(),
                       requires_grad=False)
        c_0 = Variable(emb.data.new(batch_size, self.hidden_size).zero_(),
                       requires_grad=False)
        hidden = (h_0, c_0)

        outputs = []
        for emb_t in emb.chunk(emb.size(0)):
            emb_t = emb_t.squeeze(0)
            if self.input_feed:
                emb_t = torch.cat([emb_t, output], 1)

            // FIXME: multilayer
            hidden = self.rnn(emb_t, hidden)
            output = hidden[0]
            output = self.attn(output, context.t())
            output = self.dropout(output)
            outputs += [output]

        outputs = torch.cat(outputs)
        pred = self.generator(outputs)
        return pred

Instances