Math Reasoning on LIMR
96.59Relative Overall ScoreReverse KL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Reverse KLBackbone=Qwen2.5-Math-7B2026.02 | 96.59 | 2.4 | |
| PearsonBackbone=Qwen2.5-Math-7B2026.02 | 96.28 | 2.8 | |
| Jensen-ShannonBackbone=Qwen2.5-Math-7B2026.02 | 94.93 | 3.8 | |
| KLBackbone=Qwen2.5-Math-7B2026.02 | 93.64 | 3.4 | |
| GRPOBackbone=Qwen2.5-Math-7B2026.02 | 91.72 | 5.2 | |
| HellingerBackbone=Qwen2.5-Math-7B2026.02 | 91.54 | 3.8 | |
| Total VariationBackbone=Qwen2.5-Math-7B2026.02 | 87.79 | 5.4 | |
| PearsonBackbone=Qwen2.5-Math-1.5B2026.02 | 86.49 | 2.6 | |
| Total VariationBackbone=Qwen2.5-Math-1.5B2026.02 | 83.11 | 3.8 | |
| Reverse KLBackbone=Qwen2.5-Math-1.5B2026.02 | 82.47 | 3 | |
| KLBackbone=Qwen2.5-Math-1.5B2026.02 | 81.83 | 4.4 | |
| HellingerBackbone=Qwen2.5-Math-1.5B2026.02 | 81.11 | 4.2 | |
| Jensen-ShannonBackbone=Qwen2.5-Math-1.5B2026.02 | 78.99 | 4.2 | |
| GRPOBackbone=Qwen2.5-Math-1.5B2026.02 | 74.26 | 5.2 | |
| BaseBackbone=Qwen2.5-Math-1.5B2026.02 | 0 | 8 | |
| BaseBackbone=Qwen2.5-Math-7B2026.02 | 0 | 8 |