Mathematical Reasoning on GSM8K (MAT, Token/s)
6.77MAT (Accuracy)HedgeSpec
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HedgeSpecTarget Model=Llama-3.1-8B2025.10 | 6.77 | 84.32 | |
| HedgeSpecTarget Model=Qwen-3-8B2025.10 | 6.6 | 73.59 | |
| HedgeSpecTarget Model=Qwen-3-32B2025.10 | 6.57 | 42.37 | |
| EAGLETarget Model=Llama-3.1-8B2025.10 | 6.38 | 78.32 | |
| UCBSpecTarget Model=Llama-3.1-8B2025.10 | 5.05 | 66.11 | |
| EAGLETarget Model=Qwen-3-8B2025.10 | 5.05 | 60.59 | |
| EXP3SpecTarget Model=Llama-3.1-8B2025.10 | 5.03 | 63.28 | |
| UCBSpecTarget Model=Qwen-3-8B2025.10 | 5.02 | 57.42 | |
| UCBSpecTarget Model=Qwen-3-32B2025.10 | 5 | 37.5 | |
| EXP3SpecTarget Model=Qwen-3-8B2025.10 | 4.9 | 53.57 | |
| EXP3SpecTarget Model=Qwen-3-32B2025.10 | 4.86 | 36.38 | |
| EAGLETarget Model=Qwen-3-32B2025.10 | 3.44 | 24.86 | |
| Llama-3.1-8BTarget Model=Llama-3.1-8B2025.10 | 1 | 18.26 | |
| Qwen-3-8BTarget Model=Qwen-3-8B2025.10 | 1 | 14.82 | |
| Qwen-3-32BTarget Model=Qwen-3-32B2025.10 | 1 | 8.24 |