{
  "_name_or_path": "phiva-4b-pretrained",
  "architectures": [
    "LlavaForConditionalGeneration"
  ],
  "ignore_index": -100,
  "image_token_index": 32011,
  "model_type": "llava",
  "pad_token_id": 32000,
  "projector_hidden_act": "gelu",
  "text_config": {
    "_name_or_path": "microsoft/Phi-3-mini-4k-instruct",
    "architectures": [
      "Phi3ForCausalLM"
    ],
    "embd_pdrop": 0.0,
    "eos_token_id": 32000,
    "hidden_size": 3072,
    "intermediate_size": 8192,
    "max_position_embeddings": 131072,
    "num_attention_heads": 32,
    "num_hidden_layers": 32,
    "num_key_value_heads": 32,
    "model_type": "phi3",
    "original_max_position_embeddings": 4096,
    "pad_token_id": 32000,
    "resid_pdrop": 0.0,
    "rms_norm_eps": 1e-06,
    "sliding_window": null,
    "torch_dtype": "float16",
    "vocab_size": 32013
  },
  "tie_word_embeddings": false,
  "torch_dtype": "float16",
  "transformers_version": "4.42.3",
  "vision_config": {
    "hidden_size": 768,
    "image_size": 224,
    "intermediate_size": 3072,
    "model_type": "clip_vision_model",
    "num_attention_heads": 12,
    "num_hidden_layers": 12,
    "patch_size": 16,
    "projection_dim": 512,
    "vocab_size": 32013
  },
  "vision_feature_layer": -2,
  "vision_feature_select_strategy": "default",
  "vocab_size": 32013
}