Harmlessness Evaluation on HH-RLHF
94.5Harmlessness RateSFT + TTL + DPO + TPO
Evaluation Results
| Method | Links | |
|---|---|---|
| SFT + TTL + DPO + TPOInitialization=SFT + TTL, Method=DPO + TPO2026.05 | 94.5 | |
| DPO + Topo-TPO2026.05 | 94.1 | |
| DPO + TPO2026.05 | 93.5 | |
| SFT + TTL + DPOInitialization=SFT + TTL, Method=DPO2026.05 | 92.8 | |
| DPO2026.05 | 90.2 | |
| SFT (no TTL) + DPOInitialization=SFT (no TTL), Method=DPO2026.05 | 90.2 |