Mathematical Reasoning on MATH 500 (Accuracy, Speed, Length, Modification Ratio)
93.8Accuracy14B model
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| 14B modelSpeculative Model=14B, Protocol=pass@12025.04 | 93.8 | — | 3,609 | — | 60.1 | — | — | |
| Speculative ThinkingSpeculative Model=7B, Target Model=32B, Protocol=pass@12025.04 | 93 | 0.2 | 3,767.8 | -5.2 | 46 | 42.9 | 18 | |
| 7B modelSpeculative Model=7B, Protocol=pass@12025.04 | 92.8 | — | 3,975.2 | — | 63.7 | — | — | |
| 32B modelSpeculative Model=32B, Protocol=pass@12025.04 | 92.8 | — | 3,802.2 | — | 32.2 | — | — | |
| Speculative ThinkingSpeculative Model=1.5B, Target Model=32B, Protocol=pass@12025.04 | 89.4 | 6.2 | 4,582.8 | -15.7 | 96.6 | 200 | 19 | |
| Speculative ThinkingSpeculative Model=1.5B, Target Model=14B, Protocol=pass@12025.04 | 89 | 5.8 | 4,527.4 | -16.8 | 134.6 | 124 | 19 | |
| Deepseek-Distilled Qwen-2.5-1.5BSpeculative Model=1.5B, Protocol=pass@12025.04 | 83.2 | — | 5,439.1 | — | 242.6 | — | — |