{
  "builder_name": "ITT-Purpose",
  "citation": "",
  "config_name": "default",
  "dataset_name": "ITT-Purpose",
  "dataset_size": 100,
  "description": "ITT-Purpose is a premium, high-quality, hard image-to-text-to-text benchmark dataset designed to evaluate Vision-Language Models (VLM) on document parsing, data extraction, and complex visual reasoning tasks across three styles: Meaning, Text Formatting, and Table.",
  "features": {
    "id": {
      "dtype": "string",
      "_type": "Value"
    },
    "image": {
      "_type": "Image"
    },
    "prompt": {
      "dtype": "string",
      "_type": "Value"
    },
    "response": {
      "dtype": "string",
      "_type": "Value"
    },
    "style": {
      "dtype": "string",
      "_type": "Value"
    }
  },
  "homepage": "",
  "license": "apache-2.0",
  "splits": {
    "train": {
      "name": "train",
      "num_examples": 100,
      "dataset_name": "ITT-Purpose"
    }
  },
  "version": {
    "version_str": "0.1.0",
    "major": 0,
    "minor": 1,
    "patch": 0
  },
  "task_categories": [
    "image-text-to-text",
    "visual-question-answering"
  ],
  "language": [
    "en"
  ],
  "pretty_name": "ITT-Purpose",
  "size_categories": [
    "100K<n<1M"
  ],
  "tags": [
    "multimodal",
    "image-text-to-text",
    "ocr",
    "table-qa",
    "latex",
    "vlm",
    "benchmark"
  ],
  "author": "convence"
}