Harmlessness preference labeling accuracy on SafeRLHF-RMB (test)
70.6Bench Accuracy
Biased Rubric Search
Updated 4d ago
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
2026.02 | 70.6 | 54.7 | 15.9 | |
2026.02 | 70.3 | 79.8 | -9.5 | |
2026.02 | 69.8 | 76.8 | 7 | |
2026.02 | 69.7 | 80.2 | -10.5 | |
2026.02 | 69.5 | 65.3 | 4.2 | |
2026.02 | 68.9 | 54.3 | 14.6 | |
2026.02 | 68.6 | 82.6 | -14 | |
2026.02 | 68.2 | 60.5 | 7.7 | |
2026.02 | 68 | 58.2 | 9.8 | |
2026.02 | 67.8 | 73.1 | -5.3 | |
2026.02 | 66.7 | 58.7 | 8 | |
2026.02 | 65.2 | 66 | -0.8 | |
2026.02 | 63.1 | 77.2 | -14.1 | |
2026.02 | 59.7 | 82.2 | -22.5 | |
2026.02 | 57.3 | 55.4 | 1.9 |