Safety Alignment on WildChat
81.2DSRREFLECTOR
Evaluation Results
| Method | Links | |
|---|---|---|
| REFLECTORAlignment Strategy=GDPO2026.05 | 81.2 | |
| REFLECTORAlignment Strategy=SFT2026.05 | 72.4 | |
| QwQ-32B-Preview2026.05 | 49.6 | |
| Skywork-o12026.05 | 12 | |
| LLaMA-o12026.05 | 11.8 | |
| DeepSeek-r1-Dist.2026.05 | 10.4 |