{
  "architectures": [
    "MLRS"
  ],
  "auto_map": {
    "AutoConfig": "configuration_mlrs.MLRSConfig",
    "AutoModel": "modeling_mlrs.MLRS",
    "AutoProcessor": "processing_mlrs.MLRSProcessor"
  },
  "bbox_filtering": true,
  "image_first": true,
  "instruction_prompt": "You are solving visual reasoning tasks.\n\nYou may respond in one of two final formats:\n\n1. If the task requires segmenting/localizing objects in the image, call the segmentation tool by outputting:\n<segmentation>{\"noun phrase\": \"...\", \"objects\": [{\"bbox\": [x1, y1, x2, y2]}]}</segmentation>\n\nThe segmentation JSON must contain:\n- \"noun phrase\": exactly one referential noun phrase identifying the target.\n- \"objects\": a list of objects, each with one \"bbox\" field.\n- Coordinates should use the 0-999 image coordinate grid.\n\nIf the task includes several images, segmentation tool uses the last rgb image.\n\n2. If the task asks for a textual answer and does not require segmentation, output:\n<answer>your textual answer</answer>\n\nUse <think>...</think> before the final tag when reasoning is useful.\nDo not use <answer> for segmentation. Do not use <segmentation> for non-segmentation answers.\n",
  "language_adapter_subdir": "language_adapter",
  "language_base_model": "OpenGVLab/InternVL3_5-8B-HF",
  "language_base_revision": "741a7d03020411e666c6109218ab71e08151ef86",
  "mask_threshold": 0.5,
  "max_image_longest_side": 4030,
  "max_new_tokens": 1024,
  "model_type": "mlrs",
  "multi_image_downsample_min_images": 3,
  "multi_image_max_longest_side": 448,
  "pad_to_multiple_of": 8,
  "padding_side": "left",
  "sam_base_model": "facebook/sam3",
  "sam_base_revision": "3c879f39826c281e95690f02c7821c4de09afae7",
  "sam_delta_file": "sam3_delta.safetensors",
  "transformers_version": "5.18.0"
}