File size: 2,531 Bytes
6e50a0a 0615392 6e50a0a 0615392 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 | {
"name": "SLAKE",
"release_date": "2026-05-25",
"subsets": {
"main": {
"language": [
"en"
],
"modalities": [
"single_image_start"
],
"task_type": "short_answer_qa",
"score_pipeline": [
"exact-match",
"rule-match",
"llm-judge"
],
"score_protocol": {
"reference": "No dedicated lmms-eval or VLMEvalKit task on disk (image_shortqa.py grep for SLAKE = empty; confirmed absent). SLAKE official metric is med-VQA accuracy: closed-ended (yes/no) exact-match accuracy + open-ended answer accuracy. rows/SLAKE.json shows short closed and open answers ('MRI','Abdomen','T2'). Classified by analogy to VLMEvalKit ShortQA grader used for PathVQA (image_shortqa.py:87-142).",
"note": "Official SLAKE = closed accuracy + open accuracy (exact/token match). Modeled as rule_llm_judge: exact match for closed yes/no, LLM judge for open medical answers. English subset only (mdwiratathya/SLAKE-vqa-english); official SLAKE is bilingual EN+ZH."
},
"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": "__index__"
},
"question": {
"from": "question"
},
"answer": {
"from": "answer",
"optional": true
},
"source": {
"format": "huggingface",
"url": {
"train": "https://huggingface.co/datasets/mdwiratathya/SLAKE-vqa-english",
"validation": "https://huggingface.co/datasets/mdwiratathya/SLAKE-vqa-english",
"test": "https://huggingface.co/datasets/mdwiratathya/SLAKE-vqa-english"
}
}
},
"prompt_template_source": {
"origin": "fallback",
"reference": "T1",
"notes": "Tier 5: no higher tier publishes a prompt. T1: SLAKE (2021) is classification-era (BAN/MEVF baselines), no generative prompt; LLaVA-Med (closest official-adjacent eval, model_vqa_med.py#L199-L223) feeds the raw question through its own conversation template — a harness convention, not a benchmark prompt spec. T2: no prompt column in mdwiratathya/SLAKE-vqa-english. T3: no VLMEvalKit slake dataset. T4: no lmms-eval tasks/slake (clone @047ec52). Template is the canonical single-image short-answer fallback."
}
}
}
} |