config.json

{
    "model_type": "llama",
    "vocab_size": 32000,
    "hidden_size": 4096,
    "num_attention_heads": 32,
    "num_hidden_layers": 24,
    "intermediate_size": 11008,
    "max_position_embeddings": 2048,
    "use_cache": true,
    "layer_norm_epsilon": 1e-5,
    "activation_function": "gelu_new",
    "rotary_dim": 64,
    "quantization": {
        "format": "gguf",
        "bits": 4
    }
}