Mathematical Reasoning on GSM8K (Accuracy, Token Count, ReL)
91.6AccuracyMixed Attention Module
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Mixed Attention ModuleSeries=Qwen2.5-7B, beta=0.4, alpha=0.32026.02 | 91.6 | 2,580 | 86.4 | |
| Qwen2.5-7B (Base)Series=Qwen2.5-7B, beta=0.4, alpha=0.32026.02 | 89.4 | 2,649 | 96.4 |