config.json

{
  "model_type": "llama",
  "architectures": [
    "MllamaForCausalLM"
  ],
  "torch_dtype": "float16",
  "text_config": {
    "model_type": "llama"
  },
  "vision_config": {
    "model_type": "clip_vision_model",
    "hidden_size": 1024
  }
}