config.json

{
  "architectures": [
    "Sarashina2VisionForCausalLM"
  ],
  "auto_map": {
    "AutoConfig": "configuration_sarashina2_vision.Sarashina2VisionConfig",
    "AutoModelForCausalLM": "modeling_sarashina2_vision.Sarashina2VisionForCausalLM"
  },
  "end_image_token_index": 102398,
  "image_token_index": 14,
  "model_type": "sarashina2_vision",
  "start_image_token_index": 102397,
  "text_config": {
    "_name_or_path": "sbintuitions/sarashina2-7b",
    "architectures": [
      "LlamaForCausalLM"
    ],
    "max_position_embeddings": 4096,
    "model_type": "llama",
    "rms_norm_eps": 1e-05,
    "torch_dtype": "bfloat16",
    "vocab_size": 102400
  },
  "torch_dtype": "bfloat16",
  "transformers_version": "4.47.0",
  "vision_config": {
    "hidden_size": 4096,
    "in_chans": 3,
    "model_type": "qwen2_vl",
    "spatial_patch_size": 14
  }
}