WebSRC / metadata.json
ZTWHHH's picture
metadata: migrate score_type -> score_pipeline (atomic stage contract; see mm-eval scorer docs/en/SCORING.md)
12871e7 verified
Raw
History Blame Contribute Delete
2.48 kB
{
"name": "WebSRC",
"release_date": "2021-04-21",
"subsets": {
"main": {
"language": [
"en"
],
"modalities": [
"single_image_start"
],
"task_type": "short_answer_qa",
"score_pipeline": [
"exact-match",
"rule-match"
],
"score_params": {
"string_match": "exact"
},
"score_protocol": {
"reference": "lmms-eval@lmms_eval/tasks/websrc/utils.py:133-156 — per-sample token-set F1 (lowercase, word/punct tokenization), averaged (websrc_val.yaml:14-15 metric websrc_squad_f1); official WebSRC (X-LANCE/WebSRC, arXiv:2101.09465 §5) reports Exact Match, F1 and POS. Re-opened utils.py 2026-07-07: _compute_f1 at the cited lines says exactly this.",
"note": "Official headline includes SQuAD-style F1 and POS (path overlap) which a per-sample binary exact matcher will not reproduce — only the EM component is reproduced; answers are short spans/yes-no so EM is the dominant signal. Published copy ships train+dev (test answers are withheld on the official leaderboard); domain field kept in extras but per-domain aggregation not reproduced per-sample. Dev split verified 2026-07-07: 52826/52826 rows carry non-empty answers."
},
"prompt_template": "<image>{{ question }}\nAnswer the question using a single word or phrase.",
"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": {
"domain": {
"from": "domain"
},
"page_id": {
"from": "page_id"
}
},
"source": {
"format": "huggingface",
"url": {
"train": "https://huggingface.co/datasets/rootsautomation/websrc",
"dev": "https://huggingface.co/datasets/rootsautomation/websrc"
}
}
},
"prompt_template_source": {
"origin": "official",
"reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/websrc/utils.py (websrc_doc_to_text — canonical short-answer with web-context format)",
"notes": "Tier 4: lmms-eval WebSRC canonical evaluation prompt."
}
}
}
}