ernie_args.yaml

model:
  ernie_config_path:
    val: "../LARK/ERNIE/config/ernie_config.json"
    meaning: "Path to the json file for ernie model config."
  init_checkpoint:
    val: ""
    meaning: "Path to init model"
  mode:
    val: "train"
    meaning: "Setting to train or eval or infer"
  init_pretraining_params:
    val: "pretrained/params/"
    meaning: "Init pre-training params which preforms fine-tuning from. If the arg 'init_checkpoint' has been set, this argument wouldn't be valid."

train:
  random_seed:
    val: 0
    meaning: "Random seed for training"
  batch_size:
    val: 10
    meaning: "The number of sequences contained in a mini-batch"
  epoch:
    val: 10
    meaning: "Corpus iteration num"
  use_cuda:
    val: True
    meaning: "If set, use GPU for training."
  base_learning_rate:
    val: 0.0002
    meaning: "The basic learning rate that affects the entire network."
  init_bound:
    val: 0.1
    meaning: "init bound for initialization."
  crf_learning_rate:
    val: 0.2
    meaning: "The real learning rate of the embedding layer will be (crf_learning_rate * base_learning_rate)."
  cpu_num:
    val: 10
    meaning: "The number of cpu used to train model, it works when use_cuda=False"
  print_steps:
    val: 1
    meaning: "Print the result per xxx batch of training"
  save_steps:
    val: 10
    meaning: "Save the model once per xxxx batch of training"
  validation_steps:
    val: 5
    meaning: "Do the validation once per xxxx batch of training"

data:
  vocab_path:
    val: "../LARK/ERNIE/config/vocab.txt"
    meaning: "The path of the vocabulary."
  label_map_config:
    val: "./conf/label_map.json"
    meaning: "The path of the label dictionary."
  num_labels:
    val: 57
    meaning: "label number"
  max_seq_len:
    val: 128
    meaning: "Number of words of the longest seqence."
  do_lower_case:
    val: True
    meaning: "Whether to lower case the input text. Should be True for uncased models and False for cased models."
  train_data:
    val: "./data/train.tsv"
    meaning: "The folder where the training data is located."
  test_data:
    val: "./data/test.tsv"
    meaning: "The folder where the test data is located."
  infer_data:
    val: "./data/test.tsv"
    meaning: "The folder where the infer data is located."
  model_save_dir:
    val: "./ernie_models"
    meaning: "The model will be saved in this path."