Code Generation on HumanEval (Accuracy@4)
70Accuracy@4PAPO
Evaluation results
| Method | Links | |
|---|---|---|
| PAPOModel=Qwen2.5-14B2026.03 | 70 | |
| ORM(GRPO)Model=Qwen2.5-14B2026.03 | 66 | |
| PAPOModel=Qwen2.5-7B2026.03 | 63.9 | |
| PAPOModel=Qwen3-4B-Base2026.03 | 63.9 | |
| ORM(GRPO)Model=Qwen2.5-7B2026.03 | 61.6 | |
| BaseModel=Qwen2.5-14B2026.03 | 56.9 | |
| BaseModel=Qwen2.5-7B2026.03 | 54.6 | |
| ORM(DAPO)Model=Qwen3-4B-Base2026.03 | 53.8 | |
| PAPOModel=Qwen2.5-3B2026.03 | 52.5 | |
| BaseModel=Qwen3-4B-Base2026.03 | 49.1 | |
| ORM(GRPO)Model=Qwen2.5-3B2026.03 | 39.3 | |
| BaseModel=Qwen2.5-3B2026.03 | 35.3 |