PKU-SafeRLHF
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
PKU-SafeRLHF 30K
84.5Win Rate
6
PKU-SafeRLHF-30K
87.25Win Rate
6
PKU-SafeRLHF unlearning subset Du
0.93BERTScore
5
PKU-SafeRLHF (held-out)
90.6AUROC
5
PKU-SafeRLHF
83Win Rate (Llama Judge)
4
PKU-SafeRLHF 2024 (test)
0.58Win Rate
4
PKU-SafeRLHF OOD
67.8Win Rate
4
PKU-SafeRLHF-10K (test)
356.32HV Score
3
PKU-SafeRLHF
0.82Krippendorff's alpha
3
PKU-SafeRLHF (test)
28.69Win Rate
3
PKU-SafeRLHF
80Accuracy
3
PKU-SafeRLHF (test)
69.92RM Safety Accuracy
3
PKU-SafeRLHF (test)
50.17RM Length Accuracy
3
PKU-SafeRLHF w/o trigger
70.09RM Safety Acc
3
PKU-SafeRLHF w/ trigger
70.97RM Safety Acc
3
PKU-SafeRLHF w/o trigger
44.32RM Length Accuracy
3
PKU-SafeRLHF w/ trigger
64.82RM Length Acc
3
PKU-SafeRLHF 74k (train)
-1.603Safety Expected Reward (r1)
2
PKU-SafeRLHF 7k (test)
-1.604Safety Reward (r1)
1
PKU-SafeRLHF
78Accuracy
1
PKU-SafeRLHF
96Accuracy
1