Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Reward Accuracy on Cleaned-PKU-HH-SafeRLHF (test)
Loading...
91.8
Reward Accuracy
Curri-DPO
84.624
86.487
88.35
90.213
May 25, 2026
Reward Accuracy
Updated 7d ago
Evaluation Results
Method
Method
Links
Reward Accuracy
Curri-DPO
Backbone=LLaMA-3-8B
2026.05
91.8
Staged-Competence
Backbone=LLaMA-3-8B
2026.05
91.3
Curri-DPO
Backbone=Qwen3-8B
2026.05
90.4
Standard DPO
Backbone=LLaMA-3-8B
2026.05
89.8
Staged-Competence
Backbone=Qwen3-8B
2026.05
89.6
Sequential
Backbone=LLaMA-3-8B
2026.05
89.3
Sqrt-Competence
Backbone=LLaMA-3-8B
2026.05
89
Staged-Competence
Backbone=Yi-1.5-9B
2026.05
88.2
Sequential
Backbone=Qwen3-8B
2026.05
87
Standard DPO
Backbone=Qwen3-8B
2026.05
86.7
Sqrt-Competence
Backbone=Qwen3-8B
2026.05
86.7
Curri-DPO
Backbone=Yi-1.5-9B
2026.05
86.5
Sequential
Backbone=Yi-1.5-9B
2026.05
85.7
Standard DPO
Backbone=Yi-1.5-9B
2026.05
85.5
Sqrt-Competence
Backbone=Yi-1.5-9B
2026.05
84.9
Feedback
Search any
task
Search any
task