Safety Guardrail on Response-side safety benchmarks
87F1 ScoreStage3-DPO
Evaluation Results
| Method | Links | |
|---|---|---|
| Stage3-DPOParams=4B2026.07 | 87 | |
| Qwen3Guard-8B-GenParams=8B2026.07 | 86.3 | |
| Stage2-SFTParams=4B2026.07 | 86.3 | |
| Stage1-SFT-v4Params=4B2026.07 | 86 | |
| Qwen3Guard-0.8B-GenParams=0.8B2026.07 | 85.8 | |
| Qwen3Guard-4B-GenParams=4B2026.07 | 85.3 | |
| Stage1-SFT-v1Params=4B2026.07 | 84.8 | |
| Stage1-SFT-v0Params=4B2026.07 | 84.7 | |
| YuFeng-XGuard-Reason-8BParams=8B2026.07 | 84.1 | |
| Stage1-SFT-v3Params=4B2026.07 | 83.9 | |
| YuFeng-XGuard-Reason-0.6BParams=0.6B2026.07 | 83.5 | |
| Stage1-SFT-v2Params=4B2026.07 | 82.4 |