Combined Win Rate Evaluation on PKU-SafeRLHF Sev-3 (n=100)
60.9Combined Win Rate (CVaR 0.125)IPO Entropic τ=10
Evaluation Results
| Method | Links | |
|---|---|---|
| IPO Entropic τ=10tau (τ)=10, objective=Entropic2026.05 | 60.9 | |
| IPO k = 1 Neutralk=1, objective=Neutral2026.05 | 57.2 | |
| EG Entropic τ=5tau (τ)=5, objective=Entropic2026.05 | 57.2 | |
| IPO CVaR α=0.125alpha (α)=0.125, objective=CVaR2026.05 | 56.8 | |
| IPO CVaR α=0.25alpha (α)=0.25, objective=CVaR2026.05 | 56.5 | |
| EG k = 1 Neutralk=1, objective=Neutral2026.05 | 54.9 | |
| IPO Entropic τ=5tau (τ)=5, objective=Entropic2026.05 | 54.8 | |
| NMD k = 1 Neutralk=1, objective=Neutral2026.05 | 52.5 | |
| EGPO† (HF)source=Hugging Face2026.05 | 11 | |
| SFT Basetype=Baseline2026.05 | 1.9 |