{
  "name": "ScienceQA",
  "release_date": "2022-09-20",
  "subsets": {
    "main": {
      "language": [
        "en"
      ],
      "modalities": [
        "single_image_start",
        "text"
      ],
      "task_type": "multiple_choice_qa",
      "score_pipeline": [
        "exact-match",
        "rule-match"
      ],
      "score_protocol": {
        "reference": "lmms-eval@lmms_eval/tasks/scienceqa/utils.py:34-44 sqa_process_results — llava-style regex parse of the option letter, exact_match metric, no LLM. Matches official lupantech/ScienceQA get_pred_idx rule-based scoring. Confirmed in clone (pred compared to option letter, ^[A-Z]\\. pattern).",
        "note": "Mixed-modality (text-only rows have no image); all rows are MCQ, graded by exact letter match."
      },
      "prompt_template": "{% if has_image %}<image>{% endif %}{% if hint %}Context: {{ hint }}\n{% endif %}Question: {{ question }}\nOptions:\n{% for k, v in options.items() %}({{ k }}) {{ v }}\n{% endfor %}Answer with the option letter from the given choices.",
      "mapping_from_source": {
        "media": {
          "from": "image",
          "type": "list",
          "min_items": 0,
          "max_items": 1
        },
        "id": {
          "from": "id"
        },
        "question": {
          "from": "question"
        },
        "answer": {
          "from": "answer",
          "optional": true
        },
        "hint": {
          "from": "hint",
          "optional": true
        },
        "options": {
          "from": "choices",
          "optional": true,
          "note": "list source values are normalized to {A,B,...} dict"
        },
        "extra": {
          "has_image": {
            "from": "has_image"
          },
          "subject": {
            "from": "subject"
          },
          "grade": {
            "from": "grade"
          }
        },
        "source": {
          "format": "json",
          "url": {
            "train": "https://huggingface.co/datasets/derek-thomas/ScienceQA",
            "validation": "https://huggingface.co/datasets/derek-thomas/ScienceQA",
            "test": "https://huggingface.co/datasets/derek-thomas/ScienceQA"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/scienceqa/scienceqa.yaml (canonical post_prompt='\\nAnswer with the option's letter from the given choices directly.')",
        "notes": "Tier 4: lmms-eval ScienceQA canonical evaluation prompt; supports has_image/hint conditional rendering for the mixed-modality benchmark."
      }
    }
  }
}