NLVR2 / metadata.json
ZTWHHH's picture
metadata: migrate score_type -> score_pipeline (atomic stage contract; see mm-eval scorer docs/en/SCORING.md)
08e4253 verified
Raw
History Blame Contribute Delete
2.99 kB
{
"name": "NLVR2",
"release_date": "2026-05-25",
"subsets": {
"main": {
"language": [
"en"
],
"modalities": [
"multi_image_start"
],
"task_type": "yes_no_qa",
"score_pipeline": [
"exact-match",
"rule-match"
],
"score_protocol": {
"reference": "official@github.com/lil-lab/nlvr nlvr2/eval/metrics.py — deterministic comparison of predicted True/False labels against ground truth ('prediction.lower() == example[\"label\"].lower()'); reports accuracy and consistency; no LLM. (File re-verified on GitHub master 2026-07-07.)",
"note": "Ground truth is 'True'/'False' strings (statement verification), prompt appends 'Answer with True or False.'. Official also reports CONSISTENCY (all image-pairs of the same sentence correct) and per-split (balanced/unbalanced) accuracies — the org ships 6 HF splits (balanced/unbalanced x dev/test_public/test_unseen); headline is usually test_public accuracy. Per-sample scoring reproduces accuracy only. Metadata's cited lmms-eval tasks/nlvr2/utils.py does not exist in current lmms-eval main."
},
"prompt_template": "<image><image>{{ question }}\nAnswer with True or False.",
"mapping_from_source": {
"media": {
"from": "images",
"type": "list",
"min_items": 2,
"max_items": 2
},
"id": {
"from": "id"
},
"question": {
"from": "question"
},
"answer": {
"from": "answer",
"optional": true
},
"extra": {
"synset": {
"from": "synset"
},
"writer": {
"from": "writer"
}
},
"source": {
"format": "huggingface",
"url": {
"balanced_dev": "https://huggingface.co/datasets/lmms-lab/NLVR2",
"balanced_test_public": "https://huggingface.co/datasets/lmms-lab/NLVR2",
"balanced_test_unseen": "https://huggingface.co/datasets/lmms-lab/NLVR2",
"unbalanced_dev": "https://huggingface.co/datasets/lmms-lab/NLVR2",
"unbalanced_test_unseen": "https://huggingface.co/datasets/lmms-lab/NLVR2",
"unbalanced_test_public": "https://huggingface.co/datasets/lmms-lab/NLVR2"
}
}
},
"prompt_template_source": {
"origin": "fallback",
"reference": "T2-adapted",
"notes": "Tier 5: no higher tier publishes a generative prompt. T1: lil-lab/nlvr is pre-VLM (data + accuracy/consistency scripts only); lmms-lab/NLVR2 HF card states no prompt. T2: no prompt column. T3: no VLMEvalKit nlvr2. T4: lmms-eval never had a standalone nlvr2 task (code+commit search; only NLVR2_Mantis inside llava_interleave_bench with the prompt baked into the data). Template = two <image> tags + bare statement + mm-eval convention trailer 'Answer with True or False.' (T2-adapted)."
}
}
}
}