| { | |
| "name": "SEEDBench", | |
| "release_date": "2026-05-15", | |
| "subsets": { | |
| "main": { | |
| "language": [ | |
| "en" | |
| ], | |
| "modalities": [ | |
| "single_image_start" | |
| ], | |
| "task_type": "multiple_choice_qa", | |
| "score_pipeline": [ | |
| "rule-match", | |
| "llm-match" | |
| ], | |
| "score_protocol": { | |
| "reference": "vlmevalkit@vlmeval/dataset/image_mcq.py:249-324 (ImageMCQDataset.evaluate_heuristic) -> vlmevalkit@vlmeval/dataset/utils/multiple_choice.py:475-508 mcq_vanilla_eval: can_infer rule prefetch, then GPT extractor (extract_answer_from_item @vlmeval/dataset/utils/multiple_choice.py:359-407) maps verbose output to a letter, rule-compare vs GT; SEEDBench_IMG registered @vlmeval/dataset/image_mcq.py:74. Cross-check: lmms-eval@lmms_eval/tasks/seedbench/utils.py:17-34 grades by first-character letter compare (pure rule)", | |
| "note": "SEED-Bench authors' original protocol is generation-free answer-likelihood ranking ('for each choice ... compute the likelihood ... select the choice with the highest likelihood', AILab-CVC/SEED-Bench EVALUATION.md) — not reproducible in a generation pipeline. The published mm-eval copy uses the generation-based ecosystem protocol (lmms-eval prompt + letter compare / VLMEvalKit MCQ extraction), which is how all current leaderboards run SEED-Bench on API/chat models." | |
| }, | |
| "prompt_template": "<image>{{ question }}\nA. {{ options.choice_a }}\nB. {{ options.choice_b }}\nC. {{ options.choice_c }}\nD. {{ options.choice_d }}\nAnswer with the option's letter from the given choices directly.", | |
| "mapping_from_source": { | |
| "media": { | |
| "from": "image", | |
| "type": "list", | |
| "min_items": 1, | |
| "max_items": 1 | |
| }, | |
| "id": { | |
| "from": "question_id" | |
| }, | |
| "question": { | |
| "from": "question" | |
| }, | |
| "answer": { | |
| "from": "answer", | |
| "optional": true | |
| }, | |
| "extra": { | |
| "choice_a": { | |
| "from": "choice_a" | |
| }, | |
| "choice_b": { | |
| "from": "choice_b" | |
| }, | |
| "choice_c": { | |
| "from": "choice_c" | |
| }, | |
| "choice_d": { | |
| "from": "choice_d" | |
| }, | |
| "question_type_id": { | |
| "from": "question_type_id" | |
| } | |
| }, | |
| "source": { | |
| "format": "json", | |
| "url": { | |
| "test": "https://huggingface.co/datasets/lmms-lab/SEED-Bench" | |
| } | |
| } | |
| }, | |
| "prompt_template_source": { | |
| "origin": "official", | |
| "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/seedbench/utils.py (seedbench_doc_to_text — 'A. choice_a\\nB. choice_b\\nC. choice_c\\nD. choice_d\\nAnswer with the option's letter from the given choices directly.')", | |
| "notes": "Tier 4: lmms-eval SEEDBench canonical evaluation prompt (direct choice_a/b/c/d fields)." | |
| } | |
| } | |
| } | |
| } |