task:
  model:
    encoder:
      type: bert
      bert:
        attention_dropout_rate: 0.1
        dropout_rate: 0.1
        hidden_activation: gelu
        hidden_size: 288
        initializer_range: 0.02
        intermediate_size: 256
        max_position_embeddings: 512
        num_attention_heads: 6
        num_layers: 2
        type_vocab_size: 4
        vocab_size: 114507
  train_data:
    input_path: 'YourData/COLA_train.tf_record'
    is_training: true
    global_batch_size: 32
  validation_data:
    input_path: 'YourData/COLA_eval.tf_record'
    is_training: false
    global_batch_size: 32
trainer:
  checkpoint_interval: 500
  max_to_keep: 5
  steps_per_loop: 100
  summary_interval: 100
  train_steps: 500
  validation_interval: 100
  validation_steps: -1
  optimizer_config:
    learning_rate:
      polynomial:
        initial_learning_rate: 0.001
        decay_steps: 740000
    warmup:
      polynomial:
        warmup_steps: 100