Safety Evaluation on PureBad
0HRRPretrained
Evaluation results
| Method | Links | |
|---|---|---|
| PretrainedModels=LM2-Chat2026.04 | 0 | |
| Ours (I)Models=LM2-Chat2026.04 | 0 | |
| Ours (cannot)Models=LM2-Chat2026.04 | 0 | |
| Ours (I cannot)Models=LM3-Instruct2026.04 | 4 | |
| Ours (cannot)Models=LM3-Instruct2026.04 | 7.7 | |
| PretrainedModels=LM3-Instruct2026.04 | 12.7 | |
| Ours (I)Models=LM3-Instruct2026.04 | 13.5 | |
| LoRAModels=LM3-Instruct2026.04 | 47.3 | |
| LoRAModels=LM2-Chat2026.04 | 62.3 |