UnifiedQA: Crossing Format Boundaries With a Single QA System

Benchmark Model Rank Results
common-sense-reasoning-on-commonsenseqaUnifiedQA 11B (fine-tuned)#8Accuracy: 79.1
common-sense-reasoning-on-commonsenseqaT5-XXL 11B (fine-tuned)#10Accuracy: 78.1
common-sense-reasoning-on-commonsenseqaUnifiedQA 11B (zero-shot)#12Accuracy: 76.2
common-sense-reasoning-on-commonsenseqaUnifiedQA 440M (fine-tuned)#23Accuracy: 64
common-sense-reasoning-on-commonsenseqaBART-large 440M (fine-tuned)#24Accuracy: 62.5
common-sense-reasoning-on-winograndeUnifiedQA 11B (fine-tuned)#5Accuracy: 89.4
common-sense-reasoning-on-winograndeUnified QA 406M (fine-tuned)#29Accuracy: 73.3
question-answering-on-openbookqaUnifiedQA 11B#2Accuracy: 87.2
question-answering-on-piqaUnifiedQA 3B#10Accuracy: 85.3
question-answering-on-social-iqaUnifiedQA 3B#8Accuracy: 79.8