Mathematical Reasoning on MATH (Avg Acceptance Length & Speedup)
8.03Average Acceptance Length (τ)FlexDraft
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FlexDraftModel=Qwen3-8B, Scaled training=true2026.05 | 8.03 | 5.79 | |
| dFlashModel=Qwen3-8B, Scaled training=true2026.05 | 7.93 | 5.5 | |
| DSparktarget_model=Qwen3-8B2026.07 | 5.78 | — | |
| DSparktarget_model=Gemma4-12B2026.07 | 5.78 | — | |
| DSparktarget_model=Qwen3-14B2026.07 | 5.74 | — | |
| DSparktarget_model=Qwen3-4B2026.07 | 5.7 | — | |
| Eagle3target_model=Gemma4-12B2026.07 | 5.46 | — | |
| DFlashtarget_model=Gemma4-12B2026.07 | 5.04 | — | |
| DFlashtarget_model=Qwen3-8B2026.07 | 4.91 | — | |
| DFlashtarget_model=Qwen3-4B2026.07 | 4.85 | — | |
| DFlashtarget_model=Qwen3-14B2026.07 | 4.84 | — | |
| Eagle3target_model=Qwen3-8B2026.07 | 4.77 | — | |
| Eagle3target_model=Qwen3-4B2026.07 | 4.62 | — | |
| Eagle3target_model=Qwen3-14B2026.07 | 4.6 | — | |
| EAGLE-3Model=Qwen3-8B, Scaled training=true2026.05 | 3.61 | 2.44 | |
| Apple MTPModel=Qwen3-8B, Scaled training=true2026.05 | 3.14 | 2.61 | |
| DARTModel=Qwen3-8B, Scaled training=true2026.05 | 2.7 | 2.29 |