att512.yml

batch_size: 128
buckets: 10,20,30,40
dev_source: newstest2013.tok.bpe.32000.en
dev_target: newstest2013.tok.bpe.32000.de
hparams:
  attention.dim: 512
  attention.score_type: bahdanau
  decoder.rnn_cell.cell_spec:
    class: GRUCell
    num_units: 512
  decoder.rnn_cell.dropout_input_keep_prob: 0.8
  decoder.rnn_cell.dropout_output_keep_prob: 1.0
  decoder.rnn_cell.num_layers: 2
  decoder.rnn_cell.residual_connections: false
  embedding.dim: 512
  encoder.rnn_cell.cell_spec:
    class: GRUCell
    num_units: 512
  encoder.rnn_cell.dropout_input_keep_prob: 0.8
  encoder.rnn_cell.dropout_output_keep_prob: 1.0
  encoder.rnn_cell.num_layers: 1
  encoder.rnn_cell.residual_connections: false
  encoder.type: BidirectionalRNNEncoder
  inference.max_decode_length: 100
  optimizer.clip_gradients: 5.0
  optimizer.learning_rate: 0.0001
  optimizer.name: Adam
  source.max_seq_len: 50
  source.reverse: false
  target.max_seq_len: 50
keep_checkpoint_max: 10000
model: AttentionSeq2Seq
sample_every_n_steps: 5000
save_checkpoints_secs: 1800
source_delimiter: ' '
target_delimiter: ' '
train_source: train.tok.clean.bpe.32000.en
train_steps: 2500000
train_target: train.tok.clean.bpe.32000.de
vocab_source: vocab.bpe.32000
vocab_target: vocab.bpe.32000