Mathematical Reasoning on GSM8K (Acc, Tok, ReL)
90.9AccuracyOurs
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OursBackbone=Qwen2.5-7B2026.02 | 90.9 | 2,603 | 87.2 | |
| Ours (Base)Backbone=Qwen2.5-7B2026.02 | 89.4 | 2,649 | 96.4 | |
| INFTYTHINKBackbone=Qwen2.5-7B2026.02 | 87.9 | 3,237 | 117 | |
| LightThinkerBackbone=Qwen2.5-7B2026.02 | 85.6 | 2,779 | 90 | |
| SepLLMBackbone=Qwen2.5-7B2026.02 | 84.9 | 2,582 | 92.3 | |
| H2OBackbone=Qwen2.5-7B2026.02 | 83.5 | 2,730 | 92 | |
| OursBackbone=Qwen2.5-1.5B2026.02 | 82.1 | 2,116 | 69.7 | |
| Ours (Base)Backbone=Qwen2.5-1.5B2026.02 | 80.1 | 2,086 | 76.2 | |
| INFTYTHINKBackbone=Qwen2.5-1.5B2026.02 | 79.5 | 2,503 | 89.3 | |
| LightThinkerBackbone=Qwen2.5-1.5B2026.02 | 78.8 | 2,109 | 69.6 | |
| SepLLMBackbone=Qwen2.5-1.5B2026.02 | 77.3 | 2,230 | 75.8 | |
| H2OBackbone=Qwen2.5-1.5B2026.02 | 76.1 | 2,116 | 71.7 |