Math Reasoning on GSM8K (Speedup & Generation Length)
3.8SpeedupHyperDFlash
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HyperDFlashTemperature=0, Number of drafted steps=6, Mode=Non-thinking2026.06 | 3.8 | 4.04 | |
| BlockPilotModel=Qwen2.5-Coder-32B-Instruct, Temperature=12026.06 | 3.77 | 5.36 | |
| BlockPilotModel=Qwen2.5-Coder-32B-Instruct, Temperature=02026.06 | 3.75 | 5.36 | |
| DFlash (16)Model=Qwen2.5-Coder-32B-Instruct, Temperature=02026.06 | 3.58 | 5.12 | |
| DFlash (16)Model=Qwen2.5-Coder-32B-Instruct, Temperature=12026.06 | 3.41 | 4.84 | |
| DFlash (32)Model=Qwen2.5-Coder-32B-Instruct, Temperature=12026.06 | 3.41 | 4.84 | |
| DFlash (32)Model=Qwen2.5-Coder-32B-Instruct, Temperature=02026.06 | 3.38 | 4.83 | |
| BlockPilotModel=Llama-3.1-8B-Instruct, Temperature=02026.06 | 2.96 | 4.22 | |
| DFlash (8)Model=Qwen2.5-Coder-32B-Instruct, Temperature=02026.06 | 2.94 | 4.2 | |
| DFlash (16)Model=Llama-3.1-8B-Instruct, Temperature=02026.06 | 2.91 | 4.16 | |
| DFlash (8)Model=Llama-3.1-8B-Instruct, Temperature=02026.06 | 2.89 | 4.13 | |
| DFlash (8)Model=Qwen2.5-Coder-32B-Instruct, Temperature=12026.06 | 2.89 | 4.1 | |
| DFlash (32)Model=Llama-3.1-8B-Instruct, Temperature=02026.06 | 2.87 | 4.09 | |
| MTPTemperature=0, Number of drafted steps=3, Mode=Non-thinking2026.06 | 2.76 | 3.06 | |
| BlockPilotModel=Llama-3.1-8B-Instruct, Temperature=12026.06 | 2.73 | 4.29 | |
| HyperDFlashTemperature=1, Number of drafted steps=6, Mode=Non-thinking2026.06 | 2.59 | 3.81 | |
| DFlash (16)Model=Llama-3.1-8B-Instruct, Temperature=12026.06 | 2.45 | 3.85 | |
| Vanilla DFlashTemperature=0, Number of drafted steps=6, Mode=Non-thinking2026.06 | 2.41 | 2.41 | |
| MTPTemperature=0, Number of drafted steps=6, Mode=Non-thinking2026.06 | 2.38 | 3.25 | |
| DFlash (4)Model=Llama-3.1-8B-Instruct, Temperature=02026.06 | 2.21 | 3.16 | |
| DFlash (8)Model=Llama-3.1-8B-Instruct, Temperature=12026.06 | 2.16 | 3.4 | |
| DFlash (4)Model=Qwen2.5-Coder-32B-Instruct, Temperature=02026.06 | 2.13 | 3.04 | |
| DFlash (4)Model=Qwen2.5-Coder-32B-Instruct, Temperature=12026.06 | 2.06 | 2.93 | |
| MTPTemperature=1, Number of drafted steps=3, Mode=Non-thinking2026.06 | 2.01 | 2.93 | |
| DFlash (4)Model=Llama-3.1-8B-Instruct, Temperature=12026.06 | 1.84 | 2.89 | |
| EAGLE-3Model=Llama-3.1-8B-Instruct, Temperature=02026.06 | 1.81 | 3.06 | |
| Vanilla DFlashTemperature=1, Number of drafted steps=6, Mode=Non-thinking2026.06 | 1.8 | 2.39 | |
| DFlash (32)Model=Llama-3.1-8B-Instruct, Temperature=12026.06 | 1.76 | 2.76 | |
| EAGLE-3Model=Qwen2.5-Coder-32B-Instruct, Temperature=02026.06 | 1.75 | 2.94 | |
| EAGLE-3Model=Qwen2.5-Coder-32B-Instruct, Temperature=12026.06 | 1.69 | 2.83 | |
| MTPTemperature=1, Number of drafted steps=6, Mode=Non-thinking2026.06 | 1.61 | 3.09 | |
| EAGLE-3Model=Llama-3.1-8B-Instruct, Temperature=12026.06 | 1.51 | 2.79 |