Reasoning on MMLU (AVGLEN, SPEEDUP, ACC %)
10.62Avg LengthOnline-LR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Online-LRTarget Model=Qwen / Qwen3-8B, Draft Model=Qwen / Qwen3-0.6B-Base, framework=OnlineSPEC2026.03 | 10.62 | 1.26 | 84.52 | |
| LRTarget Model=Qwen / Qwen3-8B, Draft Model=Qwen / Qwen3-0.6B-Base2026.03 | 9.4 | 1.21 | 82.86 | |
| OSD-LRTarget Model=Qwen / Qwen3-8B, Draft Model=Qwen / Qwen3-0.6B-Base2026.03 | 6.67 | 1.14 | 82.14 | |
| Target ModelTarget Model=Qwen / Qwen3-8B2026.03 | 1 | 1 | 83.81 | |
| Draft ModelDraft Model=Qwen / Qwen3-0.6B-Base2026.03 | 1 | 3.56 | 55.71 |