Mathematical Problem Solving on OlympiadBench (675 samples)
19.37AccuracyHAPO
Evaluation Results
| Method | Links | |
|---|---|---|
| HAPOBackbone=LLaMA3.2-3B Instruct2025.09 | 19.37 | |
| HAPOBackbone=LLaMA3.1-8B-Instruct2025.09 | 17.66 | |
| DAPO w/ Forking TokensBackbone=LLaMA3.2-3B Instruct2025.09 | 16.89 | |
| Vanilla DAPOBackbone=LLaMA3.2-3B Instruct2025.09 | 16.63 | |
| Vanilla DAPOBackbone=LLaMA3.1-8B-Instruct2025.09 | 16.03 | |
| DAPO w/ Forking TokensBackbone=LLaMA3.1-8B-Instruct2025.09 | 15.89 | |
| Vanilla GRPOBackbone=LLaMA3.2-3B Instruct2025.09 | 14.35 | |
| Vanilla GRPOBackbone=LLaMA3.1-8B-Instruct2025.09 | 13.81 |