Reasoning on GSM8K (Accuracy, Speed, and Length)
14.71AVGLENOnline-LR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Online-LRTarget Model=Qwen / Qwen3-8B, Draft Model=Qwen / Qwen3-0.6B-Base, framework=OnlineSPEC2026.03 | 14.71 | 1.41 | 92.88 | |
| LRTarget Model=Qwen / Qwen3-8B, Draft Model=Qwen / Qwen3-0.6B-Base2026.03 | 13.25 | 1.26 | 91.04 | |
| OSD-LRTarget Model=Qwen / Qwen3-8B, Draft Model=Qwen / Qwen3-0.6B-Base2026.03 | 12.57 | 1.2 | 93.84 | |
| DFlash+DDTreeModel=Qwen3-8B, Temperature=0.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 9.54 | 6.75 | — | |
| DFlash+DDTreeModel=Qwen3-4B, Temperature=0.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 9.33 | 6.58 | — | |
| DFlash+DDTreeModel=Qwen3-8B, Temperature=1.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 8.95 | 6.27 | — | |
| DFlash+DDTreeModel=Qwen3-4B, Temperature=1.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 8.87 | 6.17 | — | |
| DFlash+DDTreeModel=Qwen3-Coder-30B-A3B-Instruct, Temperature=0.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 7.55 | 5.83 | — | |
| DFlash+DDTreeModel=Qwen3-Coder-30B-A3B-Instruct, Temperature=1.0, Node Budget=Best from {16, 32, 64, 128, 256, 512, 1024}2026.04 | 7.38 | 5.66 | — | |
| DFlashModel=Qwen3-8B, Temperature=0.02026.04 | 6.57 | 4.78 | — | |
| DFlashModel=Qwen3-4B, Temperature=0.02026.04 | 6.51 | 4.77 | — | |
| DFlashModel=Qwen3-4B, Temperature=1.02026.04 | 5.99 | 4.35 | — | |
| DFlashModel=Qwen3-8B, Temperature=1.02026.04 | 5.93 | 4.33 | — | |
| DFlashModel=Qwen3-Coder-30B-A3B-Instruct, Temperature=0.02026.04 | 5.18 | 4 | — | |
| DFlashModel=Qwen3-Coder-30B-A3B-Instruct, Temperature=1.02026.04 | 5.08 | 3.94 | — | |
| Target ModelTarget Model=Qwen / Qwen3-8B2026.03 | 1 | 1 | 94.32 | |
| Draft ModelDraft Model=Qwen / Qwen3-0.6B-Base2026.03 | 1 | 3.6 | 53.84 |