Speculative Decoding on Fineweb-edu distillation 8B to 300M
62Spec. Accept %Random Sampling KD
Evaluation results
| Method | Links | |
|---|---|---|
| Random Sampling KD2025.03 | 62 | |
| FullKD2025.03 | 62 | |
| Peng et al. (2024)sampling method only=true2025.03 | 61.9 | |
| Raman et al. (2023)sampling method only=true2025.03 | 61.9 | |
| Peng et al. (2024)sampling method only=false2025.03 | 61.5 | |
| Raman et al. (2023)sampling method only=false2025.03 | 61 | |
| CE2025.03 | 60 |