config.json

{
  "architectures": [
    "BertForMaskedLM"
  ],
  "model_type" : "bert",
  "type_vocab_size": 2,
  "initializer_range": 0.02,
  "attention_probs_dropout_prob": 0.1,
  "hidden_dropout_prob": 0.1,
  "num_hidden_layers": 12,
  "pooler_fc_size": 768,
  "num_attention_heads": 12,
  "hidden_act": "gelu",
  "pooler_size_per_head": 128,
  "max_position_embeddings": 512,
  "pooler_num_attention_heads": 12,
  "vocab_size": 64000,
  "pooler_type": "first_token_transform",
  "intermediate_size": 3072,
  "pooler_num_fc_layers": 3,
  "hidden_size": 768,
  "directionality": "bidi"
}