Instruction Following and Safety Alignment on AlpacaEval Borderline
98WinRateBest-of-N
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Best-of-NGenerator=GPT-OSS-20B2025.10 | 98 | 4 | |
| SGGenerator=GPT-OSS-20B2025.10 | 97 | 3.1 | |
| Threshold filterGenerator=GPT-OSS-20B2025.10 | 96 | 2.1 |