Mathematical Problem Solving on MATH 500 samples (Accuracy)
47.83Accuracy (MATH 500)HAPO
Evaluation Results
| Method | Links | |
|---|---|---|
| HAPOBackbone=LLaMA3.1-8B-Instruct2025.09 | 47.83 | |
| DAPO w/ Forking TokensBackbone=LLaMA3.1-8B-Instruct2025.09 | 46.55 | |
| HAPOBackbone=LLaMA3.2-3B Instruct2025.09 | 44.86 | |
| Vanilla DAPOBackbone=LLaMA3.1-8B-Instruct2025.09 | 44.05 | |
| DAPO w/ Forking TokensBackbone=LLaMA3.2-3B Instruct2025.09 | 43.34 | |
| Vanilla DAPOBackbone=LLaMA3.2-3B Instruct2025.09 | 41.95 | |
| Vanilla GRPOBackbone=LLaMA3.1-8B-Instruct2025.09 | 40.78 | |
| Vanilla GRPOBackbone=LLaMA3.2-3B Instruct2025.09 | 38.62 |