Downstream Datasets
Benchmarks
TaskDatasetSOTA resultTrendResults
Downstream Datasets Average
85.4Average Accuracy
68
15 Downstream Datasets
90.7Food-101
24
12 Downstream Datasets Fine-grained, Natural, Specialized (test)
72.07Accuracy (Fine-grained)
21
Multiple Downstream Datasets (LAMBADA, ARC, WinoGrande, PIQA, HellaSwag, SciQ, RACE)
45LAMBADA (OpenAI)
12
12 Downstream Datasets (ChestX, CropDisease, Deep Weeds, DTD, EuroSAT, Flowers 102, Kaokore, Omniglot, Resisc45, Sketch, SVHN, ISIC)
32.57ChestX Accuracy (FS)
8
11 downstream datasets Highly Imbalanced, τ=0.06
71.58IN. Score
6
11 downstream datasets Balanced, τ=1
73.92IN Accuracy
6