Chat Quality Evaluation on AlpacaEval gpt-4o-mini judge v2 (test)
60LC Win Rate (%)ENPO₁/₄
Evaluation Results
| Method | Links | |
|---|---|---|
| ENPO₁/₄Iteration=3, 1/η=1.875 · 10⁻³2026.05 | 60 | |
| ENPO₁/₈Iteration=3, 1/η=0.9375 · 10⁻³2026.05 | 57.63 | |
| INPO₁/₂Iteration=3, 1/η=3.75 · 10⁻³2026.05 | 55.06 | |
| INPOIteration=3, 1/η=7.5 · 10⁻³2026.05 | 53.75 | |
| ENPO₁/₂Iteration=3, 1/η=3.75 · 10⁻³2026.05 | 53.04 | |
| INPO₁/₄Iteration=3, 1/η=1.875 · 10⁻³2026.05 | 49.58 | |
| XPOIteration=32026.05 | 48.57 | |
| Base Llama-3-8B-InstructIteration=32026.05 | 34.83 | |
| INPO₁/₈Iteration=3, 1/η=0.9375 · 10⁻³2026.05 | 32.13 |