{
    "architectures": [
        "MegatronBertForPreTraining"
    ],
    "attention_probs_dropout_prob": 0.1,
    "hidden_act": "gelu_new",
    "hidden_dropout_prob": 0.1,
    "hidden_size": 768,
    "gradient_checkpointing": false,
    "initializer_range": 0.2,
    "intermediate_size": 3072,
    "layer_norm_eps": 1e-12,
    "max_position_embeddings": 512,
    "model_type": "megatron-bert",
    "num_attention_heads": 12,
    "num_hidden_layers": 12,
    "pad_token_id": 0,
    "position_embedding_type": "absolute",
    "type_vocab_size": 2,
    "use_cache": false,
    "vocab_size": 32000
}