Preference Alignment on Koala (GPT-4o-mini & RM Evaluators)
77.75Win Rate (Reward Model)Vanilla Baseline
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Vanilla BaselinePreference Dimensions=2-dim2026.05 | 77.75 | 71.06 | 72.45 | |
| p-soup & Direct Fine-tuningPreference Dimensions=2-dim2026.05 | 70.63 | 47.75 | 57.31 | |
| Direct PromptingPreference Dimensions=2-dim2026.05 | 52.75 | 50.31 | 50.47 |