Judge Accuracy on RMB-SafeRLHF (Bench)
85.6AccuracyBiased Rubric Search
Evaluation Results
| Method | Links | |
|---|---|---|
| Biased Rubric SearchJudge Model=Qwen3-14B2026.02 | 85.6 | |
| Seed RubricJudge Model=Qwen3-14B2026.02 | 81.7 | |
| Biased Rubric SearchJudge Model=Qwen3-14B2026.02 | 70.3 | |
| Seed RubricJudge Model=Qwen3-14B2026.02 | 69.5 |