Math Reasoning on MATH500 (avg@16, pass@16)
94.8Pass@16KPO-clipped
Evaluation results
| Method | Links | ||
|---|---|---|---|
| KPO-clippedClipping strategy=clipped, Number of candidate solutions=162026.02 | 94.8 | 89.42 | |
| GSPONumber of candidate solutions=162026.02 | 94 | 87.41 | |
| GMPONumber of candidate solutions=162026.02 | 93.4 | 86.62 | |
| KPO-unclippedClipping strategy=unclipped, Number of candidate solutions=162026.02 | 93 | 88.24 | |
| GEB-1/πBackbone=Qwen2.5-7B2025.09 | 93 | — | |
| GRPONumber of candidate solutions=162026.02 | 92.8 | 85.66 | |
| GEB-πBackbone=Qwen2.5-7B2025.09 | 92.8 | — | |
| GEB-arctanhπBackbone=Qwen2.5-7B2025.09 | 92.8 | — | |
| DPOBackbone=Qwen2.5-7B2025.09 | 89.8 | — |