Binary Classification on Unsafe Bench (test)
90.28Macro-F1SFT-Single
Evaluation Results
| Method | Links | |
|---|---|---|
| SFT-Singletraining=Single-task SFT2026.03 | 90.28 | |
| SFT-Unifiedtraining=Unified multi-task SFT2026.03 | 89.49 | |
| RL-Singletraining=Single-task RL2026.03 | 86.92 | |
| MT-RL-Judgetraining=Multi-task RL2026.03 | 85.22 | |
| Off-the-shelf2026.03 | 72.78 |