Code Generation on HumanEval (Avg@64, Pass@64)
68.3Avg@64DARS-HW-Breadth
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DARS-HW-BreadthBase Model=Qwen2.5-Math-7B2025.08 | 68.3 | 94.9 | |
| DARS-HWBase Model=Qwen2.5-Math-7B2025.08 | 63.7 | 94.8 | |
| RLVR BaselineBase Model=Qwen2.5-Math-7B2025.08 | 63.2 | 93.9 | |
| DARS-HW-BreadthBase Model=Qwen2.5-Math-1.5B2025.08 | 49.5 | 85.6 | |
| RLVR BaselineBase Model=Qwen2.5-Math-1.5B2025.08 | 41.1 | 84.2 | |
| DARS-HWBase Model=Qwen2.5-Math-1.5B2025.08 | 40.7 | 85.8 |