{
  "name": "MTVQA",
  "release_date": "2024-05-16",
  "subsets": {
    "main": {
      "language": [
        "ar",
        "de",
        "fr",
        "it",
        "ja",
        "ko",
        "ru",
        "th",
        "vi"
      ],
      "modalities": [
        "single_image_start"
      ],
      "task_type": "vqa",
      "prompt_template": "<image>{{ question }}\nAnswer the question using a word or phrase in the language of the question.",
      "mapping_from_source": {
        "media": {
          "from": "image",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "id": {
          "from": "id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "answer",
          "optional": true
        },
        "extra": {
          "language": {
            "from": "language"
          },
          "source_id": {
            "from": "source_id"
          }
        },
        "source": {
          "format": "json",
          "url": {
            "test": "https://huggingface.co/datasets/ByteDance/MTVQA"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/mtvqa/utils.py (MTVQA_PROMPT_SUFFIX='\\nAnswer the question using a word or phrase in the language of the question.')",
        "notes": "Tier 4: lmms-eval MTVQA canonical evaluation prompt (multi-language language-of-question short-answer)."
      }
    }
  }
}