Sparc.yaml

# toy_en_de.yaml

## Where the samples will be written
save_data: Sparc/
## Where the vocab(s) will be written
src_vocab: Sparc/sparc_vocab/sparc.vocab.src
tgt_vocab: Sparc/sparc_vocab/sparc.vocab.tgt
# Prevent overwriting existing files in the folder
overwrite: True
# Corpus opts:
data:
    corpus_1:
        path_src: Sparc/processed_sparc/train_src.txt
        path_tgt: Sparc/processed_sparc/train_tgt.txt
    valid:
        path_src: Sparc/processed_sparc/dev_src.txt
        path_tgt: Sparc/processed_sparc/dev_tgt.txt

## Train on a single GPU
world_size: 1
gpu_ranks: [0]
#
## Where to save the checkpoints
#save_model: TreeDST/run/model
#save_checkpoint_steps: 500
#train_steps: 1000
#valid_steps: 500

# save_model: Sparc/run/LSTM_model
# save_checkpoint_steps: 500
# train_steps: 15000
# valid_steps: 500
# copy_attn: True
# copy_attn_force: True
# encoder_type: rnn
# decoder_type: rnn
# layers: 4
# rnn_size: 500
# rnn_type: LSTM
# copy_attn: True


save_model: Sparc/run/Transformer_model
save_checkpoint_steps: 500
train_steps: 10000
valid_steps: 500
copy_attn: True
accum_count: [3]
accum_steps: [0]
average_decay: 0.0005
model_dtype: "fp32"
optim: "adam"
learning_rate: 2
warmup_steps: 8000
decay_method: "noam"
adam_beta2: 0.998
max_grad_norm: 0
label_smoothing: 0.1
param_init: 0
param_init_glorot: true
normalization: "tokens"
encoder_type: transformer
decoder_type: transformer
enc_layers: 6
dec_layers: 6
heads: 8
rnn_size: 512
word_vec_size: 512
transformer_ff: 2048
dropout_steps: [0]
dropout: [0.1]
attention_dropout: [0.1]
n_samples: 10000