Support sugar crepe for compositionality evaluation

[mehdidc]

https://github.com/RAIVNLab/sugar-crepe/

Issue #97

[mehdidc]

Vit-BigG-14 results from the paper (https://arxiv.org/abs/2306.14610) are reproduced

{"dataset": "sugar_crepe/add_att", "model": "ViT-bigG-14", "pretrained": "laion2b_s39b_b160k", "task": "image_caption_selection", "metrics": {"acc": 0.8453757166862488}, "language": "en"}
{"dataset": "sugar_crepe/add_obj", "model": "ViT-bigG-14", "pretrained": "laion2b_s39b_b160k", "task": "image_caption_selection", "metrics": {"acc": 0.9214354753494263}, "language": "en"}
{"dataset": "sugar_crepe/replace_att", "model": "ViT-bigG-14", "pretrained": "laion2b_s39b_b160k", "task": "image_caption_selection", "metrics": {"acc": 0.8807106614112854}, "language": "en"}
{"dataset": "sugar_crepe/replace_obj", "model": "ViT-bigG-14", "pretrained": "laion2b_s39b_b160k", "task": "image_caption_selection", "metrics": {"acc": 0.9673123359680176}, "language": "en"}
{"dataset": "sugar_crepe/replace_rel", "model": "ViT-bigG-14", "pretrained": "laion2b_s39b_b160k", "task": "image_caption_selection", "metrics": {"acc": 0.7489331364631653}, "language": "en"}
{"dataset": "sugar_crepe/swap_att", "model": "ViT-bigG-14", "pretrained": "laion2b_s39b_b160k", "task": "image_caption_selection", "metrics": {"acc": 0.7492492198944092}, "language": "en"}
{"dataset": "sugar_crepe/swap_obj", "model": "ViT-bigG-14", "pretrained": "laion2b_s39b_b160k", "task": "image_caption_selection", "metrics": {"acc": 0.6260162591934204}, "language": "en"}

from the paper: