Multi-turn Dialogue Response Generation on MT-bench (test)
2.78Speedup Ratio (γ)CTC-drafter
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CTC-drafterBase Model=Vicuna-7B2024.11 | 2.78 | 3.56 | |
| CTC-drafterBase Model=Vicuna-13B2024.11 | 2.52 | 3.51 | |
| HydraBase Model=Vicuna-7B2024.11 | 2.36 | 3.04 | |
| CTC-drafterBase Model=Vicuna-33B2024.11 | 2.2 | 3.53 | |
| HydraBase Model=Vicuna-13B2024.11 | 2.17 | 3.06 | |
| HydraBase Model=Vicuna-33B2024.11 | 2.15 | 2.95 | |
| MedusaBase Model=Vicuna-7B2024.11 | 2.13 | 2.58 | |
| MedusaBase Model=Vicuna-13B2024.11 | 1.97 | 2.6 | |
| MedusaBase Model=Vicuna-33B2024.11 | 1.93 | 2.55 | |
| VanillaBase Model=Vicuna-7B2024.11 | 1 | 1 | |
| VanillaBase Model=Vicuna-13B2024.11 | 1 | 1 | |
| VanillaBase Model=Vicuna-33B2024.11 | 1 | 1 |