70K sentence completion problems testing commonsense natural language inference.
Accuracy is the reported evaluation metric for HellaSwag. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.
Higher is better