config.json

{
  "aligner_config": {
    "cls": "MlpProjector",
    "model_type": "aligner",
    "params": {
      "depth": 2,
      "input_dim": 1024,
      "n_embed": 4096,
      "projector_type": "low_high_hybrid_split_mlp_gelu"
    }
  },
  "architectures": [
    "MultiModalityCausalLM"
  ],
  "language_config": {
    "max_position_embeddings": 16384,
    "model_type": "llama",
    "num_hidden_layers": 30,
    "torch_dtype": "float16",
    "vocab_size": 102400
  },
  "model_type": "multi_modality",
  "torch_dtype": "float16",
  "transformers_version": "4.38.2",
  "vision_config": {
    "cls": "HybridVisionTower",
    "model_type": "vision",
    "params": {
      "concat_type": "tuple",
      "freeze_high": true,
      "freeze_low": true,
      "high_res_cfg": {
        "ckpt_path": "",
        "image_size": 1024,
        "model_name": "sam_b_downsample",
        "output_dim": 1024,
        "pixel_mean": [
          0.48145466,
          0.4578275,
          0.40821073
        ],
        "pixel_std": [
          0.26862954,
          0.26130258,
          0.27577711
        ],
        "select_feature": "same",
        "select_layer": -1
      },
      "low_res_cfg": {
        "ckpt_path": "",
        "image_size": 384,
        "model_name": "siglip_large_patch16_384",
        "output_dim": 1024,
        "pixel_mean": [
          0.5,
          0.5,
          0.5
        ],
        "pixel_std": [
          0.5,
          0.5,
          0.5
        ],
        "select_feature": "same",
        "select_layer": -1
      }
    }
  }
}