config.json

{
    "model_type": "llama",
    "hidden_size": 4096,
    "num_attention_heads": 32,
    "num_hidden_layers": 40,
}