Safety and Helpfulness Evaluation on WildTeaming
0.3Harm RatePPO SafeSearch w/o qr. & hf.
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PPO SafeSearch w/o qr. & hf.Backbone=Qwen-2.5-7B-Instruct2025.10 | 0.3 | 1.38 | |
| GRPO SafeSearchBackbone=Qwen-2.5-7B-Instruct2025.10 | 0.6 | 2.23 | |
| PPO SafeSearchBackbone=Qwen-2.5-7B-Instruct2025.10 | 1.9 | 2.16 | |
| PPO SafeSearch w/o hf.Backbone=Qwen-2.5-7B-Instruct2025.10 | 1.9 | 1.45 | |
| PPO SafeSearch w/o qr.Backbone=Qwen-2.5-7B-Instruct2025.10 | 3.6 | 2.25 | |
| PPO SafeSearchBackbone=Qwen-2.5-3B-Instruct2025.10 | 48.9 | 2.51 | |
| GRPO Utility-Only Ft.Backbone=Qwen-2.5-7B-Instruct2025.10 | 78.8 | 2.53 | |
| BaseBackbone=Qwen-2.5-7B-Instruct2025.10 | 83.9 | 2.79 | |
| Search BlockBackbone=Qwen-2.5-7B-Instruct2025.10 | 84.7 | 2.66 | |
| Document FilterBackbone=Qwen-2.5-7B-Instruct2025.10 | 85.7 | 2.82 | |
| PPO Utility-Only Ft.Backbone=Qwen-2.5-7B-Instruct2025.10 | 87.3 | 2.88 | |
| BaseBackbone=Qwen-2.5-3B-Instruct2025.10 | 87.5 | 2.9 | |
| Document FilterBackbone=Qwen-2.5-3B-Instruct2025.10 | 88 | 2.89 | |
| PPO Utility-Only Ft.Backbone=Qwen-2.5-3B-Instruct2025.10 | 88.6 | 2.87 | |
| Search BlockBackbone=Qwen-2.5-3B-Instruct2025.10 | 88.8 | 2.97 |