{
  "name": "CMMMU",
  "release_date": "2024-01-22",
  "subsets": {
    "art_and_design": {
      "language": [
        "en"
      ],
      "modalities": [
        "multi_image_interleave"
      ],
      "task_type": "multiple_choice_qa",
      "score_pipeline": [
        "exact-match",
        "rule-match"
      ],
      "score_protocol": {
        "reference": "vlmevalkit@vlmeval/dataset/cmmmu.py:227-330 CMMMU.evaluate — get_multi_choice_prediction (choice regex) for 选择, get_TF_prediction (对/错 keyword vote) for 判断, get_fill_blank_prediction+normalize_str (number/string match) for 填空; fully rule-based, no LLM. Matches official m-a-p/CMMMU eval. Confirmed in clone (dispatch on line['type']).",
        "note": "CMMMU is mixed-type: 选择 (MCQ letter), 判断 (true/false), 填空 (fill-blank number/string); all graded by deterministic parsers. Official reports 总准确率 plus per-category accuracy."
      },
      "prompt_template": "{{ question }}{% if options %}\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Answer with the option's letter from the given choices directly.{% else %}\nAnswer the question using a single word or phrase.{% endif %}",
      "mapping_from_source": {
        "media": {
          "from": "images",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "id": {
          "from": "id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "answer",
          "optional": true
        },
        "options": {
          "from": "options",
          "optional": true,
          "note": "list source values are normalized to {A,B,...} dict"
        },
        "extra": {
          "subject": {
            "from": "subject"
          },
          "type": {
            "from": "type"
          },
          "category": {
            "from": "category"
          },
          "subcategory": {
            "from": "subcategory"
          },
          "subfield": {
            "from": "subfield"
          },
          "difficulty_level": {
            "from": "difficulty_level"
          },
          "analysis": {
            "from": "analysis"
          },
          "n_images": {
            "from": "n_images"
          }
        },
        "source": {
          "format": "huggingface",
          "url": {
            "dev": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "val": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "test": "https://huggingface.co/datasets/m-a-p/CMMMU"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/cmmmu/utils.py (cmmmu construct_prompt; Chinese task_instructions preamble + 'A./B./...' options + 'Answer with the option's letter from the given choices directly.' trailer)",
        "notes": "Tier 4: lmms-eval CMMMU canonical eval prompt — used by the authors-recommended evaluation framework for CMMMU (Chinese MMMU)."
      }
    },
    "business": {
      "language": [
        "en"
      ],
      "modalities": [
        "multi_image_interleave"
      ],
      "task_type": "short_answer_qa",
      "score_pipeline": [
        "exact-match",
        "rule-match"
      ],
      "score_protocol": {
        "reference": "vlmevalkit@vlmeval/dataset/cmmmu.py:227-330 CMMMU.evaluate — rule-based get_multi_choice_prediction / get_TF_prediction / get_fill_blank_prediction+normalize_str, no LLM (official m-a-p/CMMMU protocol). Confirmed in clone.",
        "note": "Mixed-type subset dominated by 填空 fill-blank rows; 选择/判断/填空 all graded by deterministic parsers; official reports 总准确率 + per-category accuracy."
      },
      "prompt_template": "{{ question }}{% if options %}\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Answer with the option's letter from the given choices directly.{% else %}\nAnswer the question using a single word or phrase.{% endif %}",
      "mapping_from_source": {
        "media": {
          "from": "images",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "id": {
          "from": "id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "answer",
          "optional": true
        },
        "options": {
          "from": "options",
          "optional": true,
          "note": "list source values are normalized to {A,B,...} dict"
        },
        "extra": {
          "subject": {
            "from": "subject"
          },
          "type": {
            "from": "type"
          },
          "category": {
            "from": "category"
          },
          "subcategory": {
            "from": "subcategory"
          },
          "subfield": {
            "from": "subfield"
          },
          "difficulty_level": {
            "from": "difficulty_level"
          },
          "analysis": {
            "from": "analysis"
          },
          "n_images": {
            "from": "n_images"
          }
        },
        "source": {
          "format": "huggingface",
          "url": {
            "dev": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "val": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "test": "https://huggingface.co/datasets/m-a-p/CMMMU"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/cmmmu/utils.py (cmmmu construct_prompt; Chinese task_instructions preamble + 'A./B./...' options + 'Answer with the option's letter from the given choices directly.' trailer)",
        "notes": "Tier 4: lmms-eval CMMMU canonical eval prompt — used by the authors-recommended evaluation framework for CMMMU (Chinese MMMU)."
      }
    },
    "health_and_medicine": {
      "language": [
        "en"
      ],
      "modalities": [
        "multi_image_interleave"
      ],
      "task_type": "multiple_choice_qa",
      "score_pipeline": [
        "exact-match",
        "rule-match"
      ],
      "score_protocol": {
        "reference": "vlmevalkit@vlmeval/dataset/cmmmu.py:227-330 CMMMU.evaluate — rule-based multi-choice/TF/fill-blank parsers, no LLM (official m-a-p/CMMMU protocol). Confirmed in clone.",
        "note": "Mixed-type subset (选择/判断/填空) all graded by deterministic parsers; official reports 总准确率 + per-category accuracy."
      },
      "prompt_template": "{{ question }}{% if options %}\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Answer with the option's letter from the given choices directly.{% else %}\nAnswer the question using a single word or phrase.{% endif %}",
      "mapping_from_source": {
        "media": {
          "from": "images",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "id": {
          "from": "id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "answer",
          "optional": true
        },
        "options": {
          "from": "options",
          "optional": true,
          "note": "list source values are normalized to {A,B,...} dict"
        },
        "extra": {
          "subject": {
            "from": "subject"
          },
          "type": {
            "from": "type"
          },
          "category": {
            "from": "category"
          },
          "subcategory": {
            "from": "subcategory"
          },
          "subfield": {
            "from": "subfield"
          },
          "difficulty_level": {
            "from": "difficulty_level"
          },
          "analysis": {
            "from": "analysis"
          },
          "n_images": {
            "from": "n_images"
          }
        },
        "source": {
          "format": "huggingface",
          "url": {
            "dev": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "val": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "test": "https://huggingface.co/datasets/m-a-p/CMMMU"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/cmmmu/utils.py (cmmmu construct_prompt; Chinese task_instructions preamble + 'A./B./...' options + 'Answer with the option's letter from the given choices directly.' trailer)",
        "notes": "Tier 4: lmms-eval CMMMU canonical eval prompt — used by the authors-recommended evaluation framework for CMMMU (Chinese MMMU)."
      }
    },
    "humanities_and_social_sciences": {
      "language": [
        "en"
      ],
      "modalities": [
        "multi_image_interleave"
      ],
      "task_type": "multiple_choice_qa",
      "score_pipeline": [
        "exact-match",
        "rule-match"
      ],
      "score_protocol": {
        "reference": "vlmevalkit@vlmeval/dataset/cmmmu.py:227-330 CMMMU.evaluate — rule-based multi-choice/TF/fill-blank parsers, no LLM (official m-a-p/CMMMU protocol). Confirmed in clone.",
        "note": "Mixed-type subset (选择/判断/填空) all graded by deterministic parsers; official reports 总准确率 + per-category accuracy."
      },
      "prompt_template": "{{ question }}{% if options %}\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Answer with the option's letter from the given choices directly.{% else %}\nAnswer the question using a single word or phrase.{% endif %}",
      "mapping_from_source": {
        "media": {
          "from": "images",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "id": {
          "from": "id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "answer",
          "optional": true
        },
        "options": {
          "from": "options",
          "optional": true,
          "note": "list source values are normalized to {A,B,...} dict"
        },
        "extra": {
          "subject": {
            "from": "subject"
          },
          "type": {
            "from": "type"
          },
          "category": {
            "from": "category"
          },
          "subcategory": {
            "from": "subcategory"
          },
          "subfield": {
            "from": "subfield"
          },
          "difficulty_level": {
            "from": "difficulty_level"
          },
          "analysis": {
            "from": "analysis"
          },
          "n_images": {
            "from": "n_images"
          }
        },
        "source": {
          "format": "huggingface",
          "url": {
            "dev": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "val": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "test": "https://huggingface.co/datasets/m-a-p/CMMMU"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/cmmmu/utils.py (cmmmu construct_prompt; Chinese task_instructions preamble + 'A./B./...' options + 'Answer with the option's letter from the given choices directly.' trailer)",
        "notes": "Tier 4: lmms-eval CMMMU canonical eval prompt — used by the authors-recommended evaluation framework for CMMMU (Chinese MMMU)."
      }
    },
    "science": {
      "language": [
        "en"
      ],
      "modalities": [
        "multi_image_interleave"
      ],
      "task_type": "multiple_choice_qa",
      "score_pipeline": [
        "exact-match",
        "rule-match"
      ],
      "score_protocol": {
        "reference": "vlmevalkit@vlmeval/dataset/cmmmu.py:227-330 CMMMU.evaluate — rule-based multi-choice/TF/fill-blank parsers, no LLM (official m-a-p/CMMMU protocol). Confirmed in clone.",
        "note": "Mixed-type subset (选择/判断/填空) all graded by deterministic parsers; official reports 总准确率 + per-category accuracy."
      },
      "prompt_template": "{{ question }}{% if options %}\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Answer with the option's letter from the given choices directly.{% else %}\nAnswer the question using a single word or phrase.{% endif %}",
      "mapping_from_source": {
        "media": {
          "from": "images",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "id": {
          "from": "id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "answer",
          "optional": true
        },
        "options": {
          "from": "options",
          "optional": true,
          "note": "list source values are normalized to {A,B,...} dict"
        },
        "extra": {
          "subject": {
            "from": "subject"
          },
          "type": {
            "from": "type"
          },
          "category": {
            "from": "category"
          },
          "subcategory": {
            "from": "subcategory"
          },
          "subfield": {
            "from": "subfield"
          },
          "difficulty_level": {
            "from": "difficulty_level"
          },
          "analysis": {
            "from": "analysis"
          },
          "n_images": {
            "from": "n_images"
          }
        },
        "source": {
          "format": "huggingface",
          "url": {
            "dev": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "val": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "test": "https://huggingface.co/datasets/m-a-p/CMMMU"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/cmmmu/utils.py (cmmmu construct_prompt; Chinese task_instructions preamble + 'A./B./...' options + 'Answer with the option's letter from the given choices directly.' trailer)",
        "notes": "Tier 4: lmms-eval CMMMU canonical eval prompt — used by the authors-recommended evaluation framework for CMMMU (Chinese MMMU)."
      }
    },
    "technology_and_engineering": {
      "language": [
        "en"
      ],
      "modalities": [
        "multi_image_interleave"
      ],
      "task_type": "multiple_choice_qa",
      "score_pipeline": [
        "exact-match",
        "rule-match"
      ],
      "score_protocol": {
        "reference": "vlmevalkit@vlmeval/dataset/cmmmu.py:227-330 CMMMU.evaluate — rule-based multi-choice/TF/fill-blank parsers, no LLM (official m-a-p/CMMMU protocol). Confirmed in clone.",
        "note": "Mixed-type subset (选择/判断/填空) all graded by deterministic parsers; official reports 总准确率 + per-category accuracy."
      },
      "prompt_template": "{{ question }}{% if options %}\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Answer with the option's letter from the given choices directly.{% else %}\nAnswer the question using a single word or phrase.{% endif %}",
      "mapping_from_source": {
        "media": {
          "from": "images",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "id": {
          "from": "id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "answer",
          "optional": true
        },
        "options": {
          "from": "options",
          "optional": true,
          "note": "list source values are normalized to {A,B,...} dict"
        },
        "extra": {
          "subject": {
            "from": "subject"
          },
          "type": {
            "from": "type"
          },
          "category": {
            "from": "category"
          },
          "subcategory": {
            "from": "subcategory"
          },
          "subfield": {
            "from": "subfield"
          },
          "difficulty_level": {
            "from": "difficulty_level"
          },
          "analysis": {
            "from": "analysis"
          },
          "n_images": {
            "from": "n_images"
          }
        },
        "source": {
          "format": "huggingface",
          "url": {
            "dev": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "val": "https://huggingface.co/datasets/m-a-p/CMMMU",
            "test": "https://huggingface.co/datasets/m-a-p/CMMMU"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/cmmmu/utils.py (cmmmu construct_prompt; Chinese task_instructions preamble + 'A./B./...' options + 'Answer with the option's letter from the given choices directly.' trailer)",
        "notes": "Tier 4: lmms-eval CMMMU canonical eval prompt — used by the authors-recommended evaluation framework for CMMMU (Chinese MMMU)."
      }
    }
  }
}