Math Reasoning on Minerva (Acc@1)
36.76Acc@1DAPO
Evaluation Results
| Method | Links | |
|---|---|---|
| DAPOBackbone=Qwen3-8B-Base2026.03 | 36.76 | |
| NL GuidanceBackbone=Qwen3-8B-Base2026.03 | 36.39 | |
| Ctrl-RBackbone=Qwen3-8B-Base, beta=0.22026.03 | 36.39 | |
| Reward ShapingBackbone=Qwen3-8B-Base2026.03 | 34.92 | |
| Ctrl-RBackbone=Qwen3-1.7B-Base, beta=1, Importance Sampling (IS)=true2026.03 | 24.26 | |
| NL GuidanceBackbone=Qwen3-1.7B-Base2026.03 | 23.52 | |
| DAPOBackbone=Qwen3-1.7B-Base2026.03 | 23.16 | |
| Reward ShapingBackbone=Qwen3-1.7B-Base2026.03 | 23.16 | |
| Ctrl-RBackbone=Qwen3-1.7B-Base, beta=0.22026.03 | 23.16 | |
| Ctrl-RBackbone=Qwen3-1.7B-Base, beta=0, Importance Sampling (IS)=false2026.03 | 22.42 |