Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Reward Accuracy on Cleaned-PKU-HH-SafeRLHF (test)

91.8Reward Accuracy

Curri-DPO

84.62486.48788.3590.213May 25, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.05
91.8
2026.05
91.3
2026.05
90.4
2026.05
89.8
2026.05
89.6
2026.05
89.3
2026.05
89
2026.05
88.2
2026.05
87
2026.05
86.7
2026.05
86.7
2026.05
86.5
2026.05
85.7
2026.05
85.5
2026.05
84.9