Code Generation on APPS+ Competition
2.67Pass@1WITH GRPO ADVANTAGE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| WITH GRPO ADVANTAGEBackbone=Qwen-2.5-Coder, Parameters=7B2026.05 | 2.67 | 11.54 | |
| ADVATNAGE WITH EXP()Backbone=Qwen-2.5-Coder, Parameters=7B2026.05 | 2.38 | 9.97 | |
| RLOOBackbone=Qwen-2.5-Coder, Parameters=7B2026.05 | 2.06 | 8.04 | |
| BASEBackbone=Qwen-2.5-Coder, Parameters=7B2026.05 | 1.42 | 8.22 | |
| SFTBackbone=Qwen-2.5-Coder, Parameters=7B2026.05 | 1.38 | 6.29 |