{
    "model_type": "llama",
    "hidden_size": 512,
    "num_hidden_layers": 12,
    "num_attention_heads": 8,
    "intermediate_size": 2048,
    "hidden_act": "gelu",
    "initializer_range": 0.02,
    "layer_norm_eps": 1e-12,
    "max_position_embeddings": 512,
    "vocab_size": 32000,
    "dropout": 0.1,
    "attention_probs_dropout_prob": 0.1
}