| { | |
| "name": "GQA-Spatial", | |
| "release_date": "2019-02-25", | |
| "subsets": { | |
| "main": { | |
| "language": [ | |
| "en" | |
| ], | |
| "modalities": [ | |
| "single_image_start" | |
| ], | |
| "task_type": "short_answer_qa", | |
| "score_pipeline": [ | |
| "exact-match", | |
| "rule-match" | |
| ], | |
| "score_protocol": { | |
| "reference": "lmms-eval@lmms_eval/tasks/gqa/gqa.yaml:18-23 — exact_match (ignore_case, ignore_punctuation) against the single gold answer, matching the official GQA accuracy definition (Hudson & Manning eval: exact string match).", | |
| "note": "Subset is a non-official spatial filter of testdev_balanced (types.semantic=='rel' OR detailed startswith 'position', 5952 rows — documented in metadata prompt_template_source.notes); official GQA accuracy over full testdev is not comparable to this filtered accuracy. Rows are mixed structural types (full-scan counts: query 4213 open single-word, verify 1305 yes/no, choose 434 pick-the-alternative; extra.structural carries the type)." | |
| }, | |
| "prompt_template": "<image>{{ question }}\\nAnswer the question using a single word or phrase.", | |
| "prompt_template_source": { | |
| "origin": "official", | |
| "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/gqa/utils.py (gqa_doc_to_text — canonical short-answer)", | |
| "notes": "Tier 4: lmms-eval GQA canonical evaluation prompt, applied unchanged to the spatial subset of GQA testdev_balanced. The SAT paper (arXiv 2412.07755) reports 'GQA-Sp (Hudson & Manning, 2019b)' without specifying the spatial filter; this subset is built by filtering official testdev_balanced rows whose types.semantic=='rel' OR types.detailed.startswith('position'), yielding 5952 spatial-relation rows. Filter rule and provenance documented here for auditability." | |
| }, | |
| "mapping_from_source": { | |
| "media": { | |
| "from": "media", | |
| "type": "list", | |
| "min_items": 1, | |
| "max_items": 1 | |
| }, | |
| "id": { | |
| "from": "id" | |
| }, | |
| "question": { | |
| "from": "question" | |
| }, | |
| "answer": { | |
| "from": "answer", | |
| "optional": true | |
| }, | |
| "extra": { | |
| "imageId": { | |
| "from": "imageId" | |
| }, | |
| "fullAnswer": { | |
| "from": "fullAnswer" | |
| }, | |
| "structural": { | |
| "from": "structural" | |
| }, | |
| "semantic_type": { | |
| "from": "semantic_type" | |
| }, | |
| "detailed": { | |
| "from": "detailed" | |
| } | |
| }, | |
| "source": { | |
| "format": "huggingface", | |
| "url": { | |
| "testdev": "https://huggingface.co/datasets/lmms-lab/GQA (testdev_balanced; spatial-relation filter applied — see prompt_template_source notes)" | |
| } | |
| } | |
| } | |
| } | |
| } | |
| } |