Chat Evaluation on MT-Bench (Speedup and Acceptance Length)
2.14SpeedupHyperDFlash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HyperDFlashTemperature=0, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.06 | 2.14 | 2.79 | |
| HyperDFlashTemperature=0, Number of drafted steps=6, Mode=Non-thinking2026.06 | 2.07 | 2.72 | |
| MTPTemperature=0, Drafted steps per verification round=3, Decoding mode=Think-high mode2026.06 | 1.94 | 2.51 | |
| HyperDFlashTemperature=1, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.06 | 1.93 | 2.52 | |
| MTPTemperature=0, Number of drafted steps=3, Mode=Non-thinking2026.06 | 1.91 | 2.49 | |
| HyperDFlashTemperature=1, Number of drafted steps=6, Mode=Non-thinking2026.06 | 1.91 | 2.51 | |
| MTPTemperature=1, Drafted steps per verification round=3, Decoding mode=Think-high mode2026.06 | 1.81 | 2.33 | |
| MTPTemperature=1, Number of drafted steps=3, Mode=Non-thinking2026.06 | 1.78 | 2.31 | |
| Vanilla DFlashTemperature=0, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.06 | 1.43 | 1.82 | |
| MTPTemperature=0, Number of drafted steps=6, Mode=Non-thinking2026.06 | 1.39 | 2.53 | |
| Vanilla DFlashTemperature=0, Number of drafted steps=6, Mode=Non-thinking2026.06 | 1.39 | 1.76 | |
| MTPTemperature=0, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.06 | 1.39 | 2.58 | |
| Vanilla DFlashTemperature=1, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.06 | 1.36 | 1.73 | |
| Vanilla DFlashTemperature=1, Number of drafted steps=6, Mode=Non-thinking2026.06 | 1.33 | 1.7 | |
| MTPTemperature=1, Drafted steps per verification round=6, Decoding mode=Think-high mode2026.06 | 1.28 | 2.39 | |
| MTPTemperature=1, Number of drafted steps=6, Mode=Non-thinking2026.06 | 1.26 | 2.34 |