Safety and Utility Evaluation on Safety and Utility Evaluation Suite
0.67Unsafe RateQwen1.5-7B-Chat (Beam Search Pruning)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen1.5-7B-Chat (Beam Search Pruning)Method=Beam Search Pruning2026.04 | 0.67 | 43 | 6.97 | 25 | 14.7 | 7.5 | 6.67 | |
| gemma-7b (Beam Search Pruning)Method=Beam Search Pruning2026.04 | 1 | 65 | 6.48 | 9.5 | 2.33 | 9 | 3.17 | |
| Llama-3-Instruct-8B-SimPO (Beam Search Pruning)Method=Beam Search Pruning2026.04 | 3.67 | 24.3 | 7.13 | 23.3 | 10.2 | 43 | 29.8 | |
| gemma-7bMethod=Base2026.04 | 5.67 | 42.5 | 7.35 | 25.8 | 11.8 | 16.5 | 9 | |
| Qwen1.5-7B-ChatMethod=Base2026.04 | 8 | 40.2 | 7.4 | 51 | 39.2 | 26 | 29.5 | |
| Llama-3-Instruct-8B-SimPOMethod=Base2026.04 | 20.2 | 43 | 8.61 | 58.2 | 65.5 | 68.2 | 45 |