STEM Reasoning on Minerva (Avg@32, Pass@16)
56.57Avg@32Training-time reweighting
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Training-time reweightingBackbone=Qwen3-8B-Base2026.03 | 56.57 | 76.78 | |
| DAPOBackbone=Qwen3-8B-Base2026.03 | 55.04 | 76.98 | |
| Training-time reweightingBackbone=Qwen2.5-Math-7B2026.03 | 49.72 | 70.37 | |
| PPLBackbone=Qwen2.5-Math-7B2026.03 | 48.68 | 68.69 | |
| DominateBackbone=Qwen2.5-Math-7B2026.03 | 47.01 | 64.59 | |
| DAPOBackbone=Qwen2.5-Math-7B2026.03 | 46.43 | 69.44 | |
| BaseBackbone=Qwen3-8B-Base2026.03 | 29.8 | 70.43 | |
| BaseBackbone=Qwen2.5-Math-7B2026.03 | 18.35 | 61.04 |