Code Generation on LiveCodeBench (Avg, V5, V6)
34.02Average ScoreHAPO
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HAPOModel=Qwen2.5-7B-Instruct-1M2025.09 | 34.02 | 32.55 | 35.48 | |
| DAPO w/ Forking TokensModel=Qwen2.5-7B-Instruct-1M2025.09 | 31.46 | 29.72 | 33.19 | |
| EDGE-GRPOModel=Qwen2.5-7B-Instruct-1M2025.09 | 30.4 | 29.27 | 31.53 | |
| ArcherModel=Qwen2.5-7B-Instruct-1M2025.09 | 29.75 | 28.63 | 30.86 | |
| Vanilla DAPOModel=Qwen2.5-7B-Instruct-1M2025.09 | 29.08 | 28.22 | 29.94 |