Mathematical Reasoning on Olympiad (Pass@1, Pass@32)
31.6Pass@1DynaMO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DynaMOBackbone=Qwen2.5-Math-7B2026.02 | 31.6 | 50.1 | |
| Fork TokensBackbone=Qwen2.5-Math-7B2026.02 | 30.9 | 47.3 | |
| W-REINFORCEBackbone=Qwen2.5-Math-7B2026.02 | 30.9 | 46.7 | |
| Entropy AdvantagesBackbone=Qwen2.5-Math-7B2026.02 | 30.4 | 47.3 | |
| Entropy LossBackbone=Qwen2.5-Math-7B2026.02 | 30.2 | 41.1 | |
| KL-COVBackbone=Qwen2.5-Math-7B2026.02 | 30.2 | 44.2 | |
| Clip-COVBackbone=Qwen2.5-Math-7B2026.02 | 29.5 | 44.6 | |
| GRPOBackbone=Qwen2.5-Math-7B2026.02 | 28.6 | 44.7 | |
| Clip-HigherBackbone=Qwen2.5-Math-7B2026.02 | 28.1 | 46.3 | |
| DynaMOBackbone=Qwen2.5-Math-1.5B2026.02 | 27.2 | 47.1 | |
| Clip-HigherBackbone=Qwen2.5-Math-1.5B2026.02 | 26.4 | 46.8 | |
| Entropy AdvantagesBackbone=Qwen2.5-Math-1.5B2026.02 | 25.9 | 44.9 | |
| Clip-COVBackbone=Qwen2.5-Math-1.5B2026.02 | 25.8 | 47.6 | |
| GRPOBackbone=Qwen2.5-Math-1.5B2026.02 | 25.6 | 47 | |
| Fork TokensBackbone=Qwen2.5-Math-1.5B2026.02 | 25.5 | 45.2 | |
| Entropy LossBackbone=Qwen2.5-Math-1.5B2026.02 | 25.4 | 46.4 | |
| KL-COVBackbone=Qwen2.5-Math-1.5B2026.02 | 25.4 | 48.1 | |
| W-REINFORCEBackbone=Qwen2.5-Math-1.5B2026.02 | 24.4 | 46.2 |