Chat Quality Evaluation on AlpacaEval gpt-4-turbo judge v2 (test)
47.89Win Rate (LC)ENPO₁/₈
Evaluation Results
| Method | Links | |
|---|---|---|
| ENPO₁/₈Iteration=3, 1/η=0.9375 · 10⁻³2026.05 | 47.89 | |
| ENPO₁/₄Iteration=3, 1/η=1.875 · 10⁻³2026.05 | 47.23 | |
| ENPO₁/₂Iteration=3, 1/η=3.75 · 10⁻³2026.05 | 46.37 | |
| INPOIteration=3, 1/η=7.5 · 10⁻³2026.05 | 46.17 | |
| INPO₁/₂Iteration=3, 1/η=3.75 · 10⁻³2026.05 | 46.17 | |
| INPO₁/₄Iteration=3, 1/η=1.875 · 10⁻³2026.05 | 41.7 | |
| XPOIteration=32026.05 | 41.43 | |
| Base Llama-3-8B-InstructIteration=32026.05 | 28.61 | |
| INPO₁/₈Iteration=3, 1/η=0.9375 · 10⁻³2026.05 | 28.42 |