Safety and Utility Evaluation on MaliciousGen & LMSYS-Chat
97.31Rule ScoreShadow-Level
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Shadow-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.01 | 97.31 | 96.35 | -0.92 | 5.06 | |
| Step-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.01 | 93.08 | 92.12 | -1.05 | 5.16 | |
| Client-LevelBackbone=Qwen2.5-7B, Malicious client ratio=30%2026.01 | 83.27 | 73.27 | -1.71 | 5.01 |