config_adapt.yml

model_dir: model


data:
  train_features_file: train.src        
  train_labels_file: train.tgt         
  eval_features_file: dev.src         
  eval_labels_file: dev.tgt          
  source_words_vocabulary: vocab.src
  target_words_vocabulary: vocab.tgt


params:
  beam_width: 5
  maximum_iterations: 250
  average_loss_in_time: true
  label_smoothing: 0.1
  length_penalty: 0.6

  optimizer: LazyAdamOptimizer
  optimizer_params:
    beta1: 0.9
    beta2: 0.998
  learning_rate: 2.0            # The learning rate scale constant.
  decay_type: noam_decay
  decay_rate: 512 
  decay_step_duration: 8        # 2 - changed to 8 for using 1GPU  # 1 decay step is 2 training steps.
  decay_steps: 8000             # Warmup steps (= 16000 training steps).
  start_decay_steps: 0          # 16000

  # (optional) to freeze sub-component of the network (transformer model) 
  freeze:
    encoder: false       
    decoder: false
    # currently not necessary options 
    #shared_embs:                # used only if enc-dec is trained with shared embs
    #src_embs: true              # if encoder is trainable 
    #tgt_embs: false             # if decoder is trainable 
    #projection: false           # output_layer before softmax 
    #nlayer_encoder: 1
    #nlayer_decoder: 6


# (optional) to load sub-components of the network, use --checkpoint  
load_weights:
  encoder: true
  decoder: true
  shared_embs: false             # only if enc-dec shared embs is used
  src_embs: true
  tgt_embs: true
  projection: true
  optim: false                    
  global_step: false
  words_per_sec: false           # not in lstm models


train:
  batch_size: 4096
  batch_type: tokens
  bucket_width: 5
  save_checkpoints_steps: 1000  # update accordingly 
  keep_checkpoint_max: 10
  save_summary_steps: 50
  train_steps: 500000           # small-models 60000 steps, 200k training examples 100000 steps
  sample_buffer_size: 5000000
  maximum_features_length: 100  
  maximum_labels_length: 100


eval:
  batch_size: 32
  num_threads: 10               # default 1
  bucket_width: 5
  prefetch_buffer_size: 1
  n_best: 1
  with_scores: false
  with_alignments: null 
  external_evaluators: BLEU     # sacreBLEU, BLEU, BLEU-detok, ROUGE
  exporters: last 
  steps: 100                    # 100 (default)
  start_delay_secs: 1           # 120 (default)
  eval_delay: 1800              # every 1/2 hour  | throttle_secs: 60 # 600 (default)  


infer:
  batch_size: 32