Single-turn Dialogue on Anthropic HH
69.18Win RateIPO*
Evaluation Results
| Method | Links | |
|---|---|---|
| IPO*Temperature=0.72026.02 | 69.18 | |
| TUR-DPOEvaluation Protocol=LLM judge2026.04 | 67.9 | |
| PPOEvaluation Protocol=Human eval2026.04 | 67.9 | |
| TUR-DPOEvaluation Protocol=Human eval2026.04 | 67.2 | |
| IPOEvaluation Protocol=LLM judge2026.04 | 66.1 | |
| DPOEvaluation Protocol=LLM judge2026.04 | 65.5 | |
| IPO*Temperature=0.12026.02 | 64.83 | |
| DPOEvaluation Protocol=Human eval2026.04 | 64.7 | |
| DIPO+Temperature=0.72026.02 | 61.34 | |
| DIPO+Temperature=0.12026.02 | 60.53 | |
| DIPOTemperature=0.12026.02 | 60.3 | |
| Sampled IPOTemperature=0.12026.02 | 60.13 | |
| Sampled IPOTemperature=0.72026.02 | 59.99 | |
| DIPOTemperature=0.72026.02 | 58.75 | |
| IPOTemperature=0.72026.02 | 57.38 | |
| IPOTemperature=0.12026.02 | 57.19 | |
| SFTTemperature=0.72026.02 | 52.77 | |
| SFTTemperature=0.12026.02 | 48.62 |