Mathematical Reasoning on SVAMP (Accuracy, Token Count)
83.33AccuracyBRIDGE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BRIDGEBackbone=Qwen 2.5-3B2026.02 | 83.33 | 105 | |
| Std-CoT KDBackbone=Qwen 2.5-3B2026.02 | 82 | 161 | |
| Base ModelBackbone=Qwen 2.5-3B2026.02 | 79.33 | 162 | |
| SuperRLBackbone=Qwen 2.5-3B2026.02 | 79.33 | 194 | |
| Instruct ModelBackbone=Llama 3.2-3B2026.02 | 69 | 127 | |
| SuperRLBackbone=Llama 3.2-3B2026.02 | 61.33 | 173 | |
| Mix Distill.Backbone=Qwen 2.5-3B2026.02 | 58.2 | 170 | |
| BRIDGEBackbone=Llama 3.2-3B2026.02 | 50.33 | 52 | |
| Instruct ModelBackbone=Qwen 2.5-3B2026.02 | 45.33 | 214 | |
| Std-CoT KDBackbone=Llama 3.2-3B2026.02 | 45 | 149 | |
| Short-CoTBackbone=Qwen 2.5-3B2026.02 | 42.1 | 140 | |
| Short-CoTBackbone=Llama 3.2-3B2026.02 | 38.33 | 130 | |
| Mix Distill.Backbone=Llama 3.2-3B2026.02 | 24.67 | 105 | |
| Base ModelBackbone=Llama 3.2-3B2026.02 | 20.67 | 192 |