llm/config.json

{
    "model_type": "llava_qwen2",
    "hidden_size": 896,
    "num_hidden_layers": 24,
    "intermediate_size": 4864,
    "num_attention_heads": 14,
    "num_key_value_heads": 2,
    "rms_norm_eps": 1e-06,
    "vocab_size": 151649,
    "max_position_embeddings": 32768,
    "rope_theta": 1000000.0,
    "tie_word_embeddings": false,
    "mm_hidden_size": 3072,
    "mm_vision_tower": "mobileclip_l_1024",
    "image_token_index": 151648,
    "quantization": {
        "group_size": 64,
        "bits": 4
    }
}