| { | |
| "name": "MathVista", | |
| "release_date": "2026-05-15", | |
| "subsets": { | |
| "main": { | |
| "language": [ | |
| "en" | |
| ], | |
| "modalities": [ | |
| "single_image_start" | |
| ], | |
| "task_type": "multiple_choice_qa", | |
| "score_pipeline": [ | |
| "rule-match", | |
| "llm-match" | |
| ], | |
| "score_protocol": { | |
| "reference": "lmms-eval@lmms_eval/tasks/mathvista/mathvista_evals.py:218-266 — rule extraction first (direct choice match, int/float parse), GPT extraction fallback via DEMO_PROMPT; utils.py:48-62 normalize_extracted_answer + safe_equal rule compare. Mirrors official lupantech/MathVista (evaluation/extract_answer.py + calculate_score.py). [Re-verified in clone 2026-07-07: extract_answer at mathvista_evals.py:221-266, process_results/normalize/safe_equal at utils.py:48-64.]", | |
| "note": "Official float normalization rounds the extracted value to a per-row `precision` column (mathvista/utils.py:60); the mm-eval copy ships answer_type but not precision (the required precision is only stated in the hint text baked into the question). Mixed MCQ/free-form subset: MAJORITY format is multi_choice (540/1000 vs 460 free_form); per-sample question_type carries the split." | |
| }, | |
| "prompt_template": "<image>{{ question }}", | |
| "mapping_from_source": { | |
| "media": { | |
| "from": "decoded_image", | |
| "type": "list", | |
| "min_items": 1, | |
| "max_items": 1 | |
| }, | |
| "id": { | |
| "from": "pid" | |
| }, | |
| "question": { | |
| "from": "query", | |
| "note": "use upstream `query` field (pre-built format-prompt with hint + question)" | |
| }, | |
| "answer": { | |
| "from": "answer", | |
| "optional": true | |
| }, | |
| "extra": { | |
| "question_type": { | |
| "from": "question_type" | |
| }, | |
| "answer_type": { | |
| "from": "answer_type" | |
| } | |
| }, | |
| "source": { | |
| "format": "json", | |
| "url": { | |
| "testmini": "https://huggingface.co/datasets/AI4Math/MathVista" | |
| } | |
| } | |
| }, | |
| "prompt_template_source": { | |
| "origin": "official", | |
| "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/mathvista/utils.py (mathvista_doc_to_text — uses 'query' column which is pre-rendered)", | |
| "notes": "Tier 4: lmms-eval MathVista canonical: pre-rendered 'query' column from source; template is bare to forward unchanged." | |
| } | |
| } | |
| } | |
| } |