{
  "architectures": [
    "InternVLForConditionalGeneration"
  ],
  "downsample_ratio": 0.5,
  "image_seq_length": 256,
  "image_token_index": 151667,
  "model_type": "internvl",
  "projector_hidden_act": "gelu",
  "text_config": {
    "_attn_implementation_autoset": true,
    "architectures": [
      "Qwen2ForCausalLM"
    ],
    "bos_token_id": 151643,
    "eos_token_id": 151645,
    "hidden_size": 896,
    "intermediate_size": 4864,
    "max_window_layers": 21,
    "model_type": "qwen2",
    "num_attention_heads": 14,
    "num_hidden_layers": 24,
    "num_key_value_heads": 2,
    "rope_theta": 1000000.0,
    "sliding_window": 32768,
    "torch_dtype": "bfloat16",
    "vocab_size": 151674
  },
  "torch_dtype": "bfloat16",
  "transformers_version": "4.50.0.dev0",
  "vision_config": {
    "_attn_implementation_autoset": true,
    "architectures": [
      "InternVisionModel"
    ],
    "attention_dropout": 0.0,
    "drop_path_rate": 0.0,
    "dropout": 0.0,
    "initializer_factor": 1.0,
    "layer_scale_init_value": 0.1,
    "model_type": "internvl_vision",
    "torch_dtype": "bfloat16"
  },
  "vision_feature_layer": -1
}