seq2seq_mp1.py 文件源码

def __init__(
            self, comm, n_layers, n_source_vocab, n_target_vocab, n_units):

        super(Encoder, self).__init__(
            embed_x=L.EmbedID(n_source_vocab, n_units),
            # Corresponding decoder LSTM will be invoked on process 1.
            mn_encoder=chainermn.links.create_multi_node_n_step_rnn(
                L.NStepLSTM(n_layers, n_units, n_units, 0.1),
                comm, rank_in=None, rank_out=1
            ),
        )
        self.comm = comm
        self.n_layers = n_layers
        self.n_units = n_units