Mathematical Reasoning on GSM8K (test) (Pass@1, Token Count, Ratio)
95.7Pass@1SimPOShortest
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SimPOShortestBackbone=Qwen3-8B2026.01 | 95.7 | 1,247 | 61.8 | |
| OriginalBackbone=Qwen3-8B2026.01 | 95.5 | 2,017 | 100 | |
| BeliefPromptBackbone=Qwen3-8B2026.01 | 95.2 | 1,109 | 55 | |
| RELIEFBackbone=Qwen3-8B2026.01 | 95 | 757 | 37.5 | |
| SFTbeliefBackbone=Qwen3-8B2026.01 | 94.9 | 1,934 | 95.9 | |
| SFTShortestBackbone=Qwen3-8B2026.01 | 94.4 | 2,162 | 107.2 | |
| SimPOShortestBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 91.3 | 1,377 | 69.8 | |
| RELIEFBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 89.8 | 1,400 | 71 | |
| OriginalBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 88.7 | 1,973 | 100 | |
| SFTShortestBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 88.3 | 897 | 45.5 | |
| SFTbeliefBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 87.7 | 1,428 | 72.4 | |
| BeliefPromptBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 87.1 | 495 | 25.1 |