Out-of-Domain Generalization on BBH, HumanEval, MMLU, and TruthfulQA Suite
59.1BBH ScorePACE
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PACEAlignment mechanism=Proximal Refinement2026.02 | 59.1 | 55.3 | 68.3 | 38.2 | 55.2 | |
| BoNSampling budget (N)=22026.02 | 58.1 | 53.9 | 68.4 | 38.8 | 54.8 | |
| BoNSampling budget (N)=82026.02 | 57 | 52.7 | 68.4 | 37.9 | 54 | |
| Llama3.1-8B2026.02 | 56.2 | 51.7 | 68.6 | 37.7 | 53.5 |