Critique Quality Evaluation on Dialog
80Win RateRCO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RCOComparison Target=LLaMa-3-70B-Instruct2025.06 | 80 | 12.5 | 7.5 | 72.5 | |
| RCOComparison Target=LLaMa-3-8B-Instruct2025.06 | 55 | 22.5 | 22.5 | 32.5 | |
| RCOComparison Target=LLaMa-3-8B-Instruct2025.06 | 50 | 30 | 20 | 30 | |
| RCOComparison Target=LLaMa-3-70B-Instruct2025.06 | 40 | 32.5 | 27.5 | 12.5 | |
| RCOComparison Target=LLaMa-3-8B-Instruct + MSE + DPO2025.06 | 35 | 22.5 | 42.5 | -7.5 | |
| RCOComparison Target=LLaMa-3-8B-Instruct + MSE + DPO2025.06 | 25 | 32.5 | 42.5 | -17.5 |