khmer-viet.yaml

# khmer-viet.yaml

## Where the samples will be written
save_data: data/run/example
## Where the vocab(s) will be written
src_vocab: data/run/example.vocab.src
tgt_vocab: data/run/example.vocab.tgt
# Prevent overwriting existing files in the folder
overwrite: True

# Corpus opts:
data:
    corpus_1:
        path_src: data/src-train-bpe.txt
        path_tgt: data/tgt-train-bpe.txt
    valid:
        path_src: data/src-val-bpe.txt
        path_tgt: data/tgt-val-bpe.txt

# Vocabulary files that were just created
src_vocab: models/run/example.vocab.src
tgt_vocab: models/run/example.vocab.tgt

# Train on a single GPU
world_size: 1
gpu_ranks: [0]

# Where to save the checkpoints
save_model: models/run/model
save_checkpoint_steps: 1000
train_steps: 30000
valid_steps: 1000

# Model
position_encoding: 'true'
enc_layers: 6
dec_layers: 6
decoder_type: transformer
encoder_type: transformer
word_vec_size: 512
rnn_size: 512
layers: 6
transformer_ff: 2048
heads: 8

# Batching
queue_size: 10000
batch_size: 4096
valid_batch_size: 4096
batch_type: tokens