Math Reasoning on AMC '23 (Acc@16)
82.18Acc@16Ctrl-R
Evaluation Results
| Method | Links | |
|---|---|---|
| Ctrl-RBackbone=Qwen3-8B-Base, beta=0.22026.03 | 82.18 | |
| Reward ShapingBackbone=Qwen3-8B-Base2026.03 | 81.87 | |
| NL GuidanceBackbone=Qwen3-8B-Base2026.03 | 81.25 | |
| DAPOBackbone=Qwen3-8B-Base2026.03 | 80.62 | |
| Ctrl-RBackbone=Qwen3-1.7B-Base, beta=0.22026.03 | 49.69 | |
| Reward ShapingBackbone=Qwen3-1.7B-Base2026.03 | 47.96 | |
| DAPOBackbone=Qwen3-1.7B-Base2026.03 | 47.81 | |
| NL GuidanceBackbone=Qwen3-1.7B-Base2026.03 | 47.65 | |
| Ctrl-RBackbone=Qwen3-1.7B-Base, beta=1, Importance Sampling (IS)=true2026.03 | 46.71 | |
| Ctrl-RBackbone=Qwen3-1.7B-Base, beta=0, Importance Sampling (IS)=false2026.03 | 45.46 |