LLM Safety and Informativeness Evaluation on AdvBench
97.2Safety RateSafeMoE-XL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SafeMoE-XLVariant=XL2026.05 | 97.2 | 8.2 | |
| Oyster-I 14BParameters=14B2026.05 | 80.5 | 6.9 | |
| SafeMoE-QwenBackbone=Qwen2026.05 | 73.2 | 7.8 | |
| RealSafe-R12026.05 | 60.8 | 5.8 | |
| SN-Tune2026.05 | 55.2 | 6.9 | |
| Deepseek-R1-QwenBackbone=Qwen2026.05 | 48.9 | 6.7 | |
| Zephyr2026.05 | 44.5 | 7.2 | |
| Qwen-3BParameters=3B2026.05 | 31.1 | 4.4 | |
| Mistral-SFT2026.05 | 27.7 | 6.7 | |
| SafeLoRA2026.05 | 25.1 | 5.9 | |
| Mistral-7BParameters=7B2026.05 | 11.9 | 4.6 |