Seq2Seq

Sequence to Sequence Learning with Neural Networks

https://proceedings.neurips.cc/paper/2014/file/a14ac55a4f27472c5d894ec1c3c743d2-Paper.pdf

구조

구현

• torch

결과

Seq2Seq(
  (encoder): Encoder(
    (embedding): Embedding(24745, 512)
    (lstm): LSTM(512, 256, num_layers=4, dropout=0.5)
    (dropout): Dropout(p=0.5, inplace=False)
  )
  (decoder): Decoder(
    (embedding): Embedding(8854, 256)
    (lstm): LSTM(256, 256, num_layers=4, dropout=0.5)
    (fc): Linear(in_features=256, out_features=8854, bias=True)
    (dropout): Dropout(p=0.5, inplace=False)
  )
)

The model has 21,684,374 trainable parameters.