{ "name": "Winoground", "release_date": "2022-01-01", "subsets": { "main": { "language": [ "en" ], "modalities": [ "multi_image_start" ], "task_type": "multiple_choice_vqa", "prompt_template": "{{ question }}", "mapping_from_source": { "media": { "from": "image", "type": "list", "min_items": 2, "max_items": 2 }, "id": { "from": "id" }, "question": { "from": "question" }, "answer": { "from": "answer", "optional": true }, "extra": { "caption_0": { "from": "caption_0" }, "caption_1": { "from": "caption_1" }, "tag": { "from": "tag" } }, "source": { "format": "json", "url": { "test": "https://huggingface.co/datasets/facebook/winoground" } } }, "prompt_template_source": { "origin": "official", "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/winoground/utils.py (winoground_doc_to_text — two-image visio-linguistic compositional reasoning)", "notes": "Tier 4: lmms-eval Winoground canonical evaluation prompt." } } } }