| { |
| "name": "CRPE", |
| "release_date": "2026-05-26", |
| "subsets": { |
| "exist": { |
| "language": [ |
| "en" |
| ], |
| "modalities": [ |
| "single_image_start" |
| ], |
| "task_type": "multiple_choice_vqa", |
| "prompt_template": "<image>Question: {{ question }}\nOptions:\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Please select the correct answer from the options above. ", |
| "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 |
| }, |
| "options": { |
| "from": "options", |
| "optional": true, |
| "note": "list source values are normalized to {A,B,...} dict" |
| }, |
| "extra": { |
| "source_id": { |
| "from": "source_id" |
| }, |
| "category": { |
| "from": "category" |
| } |
| }, |
| "source": { |
| "format": "huggingface", |
| "url": { |
| "exist": "https://huggingface.co/datasets/OpenGVLab/CRPE" |
| } |
| } |
| }, |
| "prompt_template_source": { |
| "origin": "official", |
| "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/crpe_relation/utils.py#L32-L42 (canonical eval prompt; VLMEvalKit-style 'Hint:/Question:/Options:/Please select the correct answer from the options above. ' MCQ template)", |
| "notes": "Tier 4: Authors' canonical CRPE evaluation prompt (lmms-eval crpe_relation task). The VLMEvalKit-form trailer is the deployed form; an alternative trailer 'Please answer directly with only the letter of the correct option' exists in crpe_relation.yaml for harness-internal scoring." |
| } |
| }, |
| "relation": { |
| "language": [ |
| "en" |
| ], |
| "modalities": [ |
| "single_image_start" |
| ], |
| "task_type": "multiple_choice_vqa", |
| "prompt_template": "<image>Question: {{ question }}\nOptions:\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Please select the correct answer from the options above. ", |
| "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 |
| }, |
| "options": { |
| "from": "options", |
| "optional": true, |
| "note": "list source values are normalized to {A,B,...} dict" |
| }, |
| "extra": { |
| "source_id": { |
| "from": "source_id" |
| }, |
| "category": { |
| "from": "category" |
| } |
| }, |
| "source": { |
| "format": "huggingface", |
| "url": { |
| "relation": "https://huggingface.co/datasets/OpenGVLab/CRPE" |
| } |
| } |
| }, |
| "prompt_template_source": { |
| "origin": "official", |
| "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/crpe_relation/utils.py#L32-L42 (canonical eval prompt; VLMEvalKit-style 'Hint:/Question:/Options:/Please select the correct answer from the options above. ' MCQ template)", |
| "notes": "Tier 4: Authors' canonical CRPE evaluation prompt (lmms-eval crpe_relation task). The VLMEvalKit-form trailer is the deployed form; an alternative trailer 'Please answer directly with only the letter of the correct option' exists in crpe_relation.yaml for harness-internal scoring." |
| } |
| } |
| } |
| } |
