bpe-es-gl_emb.yaml

save_data: run
## Where the vocab(s) will be written
src_vocab: run/vocab/es-gl/bpe.vocab.src
tgt_vocab: run/vocab/es-gl/bpe.vocab.tgt
overwrite: True

# Corpus opts:
data:
    europarl:
        path_src: corpora/europarl/partitions/es_train.txt
        path_tgt: corpora/europarl_translit/partitions/gl_train.txt
        transforms: [bpe, filtertoolong]
        weight: 120
    opensub:
        path_src: corpora/opensub/partitions/es_train.txt
        path_tgt: corpora/opensub_translit/partitions/gl_train.txt
        transforms: [bpe, filtertoolong]
        weight: 180
    dgt:
        path_src: corpora/dgt/partitions/es_train.txt
        path_tgt: corpora/dgt_translit/partitions/gl_train.txt
        transforms: [bpe, filtertoolong]
        weight: 18
    cluvi:
        path_src: corpora/cluvi/partitions/es_train.txt
        path_tgt: corpora/cluvi/partitions/gl_train.txt
        transforms: [bpe, filtertoolong]
        weight: 40
    opensub-es-gl:
        path_src: corpora/opensub-es-gl/partitions/es_train.txt
        path_tgt: corpora/opensub-es-gl/partitions/gl_train.txt
        transforms: [bpe, filtertoolong]
        weight: 25
    ted2020:
        path_src: corpora/ted2020/partitions/es_train.txt
        path_tgt: corpora/ted2020/partitions/gl_train.txt
        transforms: [bpe, filtertoolong]
        weight: 10
    corgaback:
        path_src: corpora/corgaback/partitions/es_train.txt
        path_tgt: corpora/corgaback/partitions/gl_train.txt
        transforms: [bpe, filtertoolong]
        weight: 13
    ccmatrix:
        path_src: corpora/ccmatrix/es.txt
        path_tgt: corpora/ccmatrix/gl.txt
        transforms: [bpe, filtertoolong]
        weight: 180
    resto:
        path_src: corpora/resto/es.txt
        path_tgt: corpora/resto/gl.txt
        transforms: [bpe, filtertoolong]
        weight: 120
    opensub_2018:
        path_src: corpora/opensub_2018/es.txt
        path_tgt: corpora/opensub_2018/gl.txt
        transforms: [bpe, filtertoolong]
        weight: 25 

        
    valid:
        path_src: corpora/partitions/all-es_valid.txt
        path_tgt: corpora/partitions_translit/all-gl_valid.txt
        transforms: [bpe, filtertoolong]

### Transform related opts:
#### Subword
src_subword_model: ./bpe/es.code
tgt_subword_model: ./bpe/gl.code
src_subword_vocab: ./run/vocab/es-gl/bpe.vocab.src
tgt_subword_vocab: ./run/vocab/es-gl/bpe.vocab.tgt
src_subword_type: bpe
tgt_subord_type: bpe

src_subword_nbest: 1
src_subword_alpha: 0.0
tgt_subword_nbest: 1
tgt_subword_alpha: 0.0

##embeddings
src_embeddings: ../embeddings/es.emb.txt
tgt_embeddings: ../embeddings/gl.emb.txt

## supported types: GloVe, word2vec
embeddings_type: "word2vec"

# word_vec_size need to match with the pretrained embeddings dimensions
word_vec_size: 512


#### Filter
src_seq_length: 150
tgt_seq_length: 150

# silently ignore empty lines in the data
skip_empty_level: silent



# General opts
save_model: run/model
keep_checkpoint: 50
save_checkpoint_steps: 10000
average_decay: 0.0005
seed: 1234
report_every: 1000
train_steps: 200000
valid_steps: 10000

# Batching
queue_size: 10000
bucket_size: 32768
world_size: 1
gpu_ranks: [0]
batch_type: "tokens"
#batch_size: 4096
batch_size: 8192
valid_batch_size: 64
batch_size_multiple: 1
max_generator_batches: 2
accum_count: [4]
accum_steps: [0]

# Optimization
model_dtype: "fp16"
optim: "adam"
learning_rate: 2
#learning_rate: 0.00005
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"

# Model
encoder_type: transformer
decoder_type: transformer
position_encoding: true
enc_layers: 6
dec_layers: 6
heads: 8
rnn_size: 512
transformer_ff: 2048
dropout_steps: [0]
dropout: [0.1]
attention_dropout: [0.1]
share_decoder_embeddings: true
share_embeddings: false