Safety
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Safety
92.1Score
27
Safety
73Avg@k
15
Safety Avg.
2.691MAE
14
Safety OOD
93.14Accuracy
13
Safety ID
99.81Accuracy
13
Safety Tweet Eval, Hatecheck, Ethos (test)
83.3Accuracy
12
Safety BeaverTails, HEx-PHI (test)
95.67BeaverTails Score
10
Safety TruthfulQA CocoNot
68.97Accuracy
8
Safety Overall
32.9Reasoning Accuracy (Avg)
4