Reward Modeling on Anthropic Helpful-Harmless (HHH)
0.7108RewardBench TotalFull Set
Evaluation Results
| Method | Links | |
|---|---|---|
| Full SetSelection Size=100%2025.08 | 0.7108 | |
| Difficulty-Based Preference Data SelectionSelection Size=10%2025.08 | 0.7052 | |
| RandomSelection Size=10%2025.08 | 0.6798 |