Machine Translation on WMT en-de 14 (Operational Metrics)
86.3Acceptance RateGSD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GSDTarget Model=LLaMA-2-7b, Draft Model=LLaMA-160m, k=4, tau=1, theta_prob=0.4, theta_sib=0.12024.07 | 86.3 | 1.36 | |
| TSDTarget Model=LLaMA-2-7b, Draft Model=LLaMA-160m, k=4, tau=1, theta_prob=0.4, theta_sib=0.12024.07 | 85.1 | 0.56 | |
| SSDTarget Model=LLaMA-2-7b, Draft Model=LLaMA-160m, k=4, tau=1, theta_prob=0.4, theta_sib=0.12024.07 | 78.3 | 1.29 | |
| GSDTarget Model=OPT-13b, Draft Model=OPT-350m, k (maximum out-degree)=4, tau (redundant node threshold)=1, theta_prob=0.4, theta_sib=0.1, Decoding Setting=Deterministic2024.07 | 73.3 | 1.28 | |
| TSDTarget Model=OPT-13b, Draft Model=OPT-350m, k (maximum out-degree)=4, tau (redundant node threshold)=1, theta_prob=0.4, theta_sib=0.1, Decoding Setting=Deterministic2024.07 | 70.5 | 0.38 | |
| SSDTarget Model=OPT-13b, Draft Model=OPT-350m, k (maximum out-degree)=4, tau (redundant node threshold)=1, theta_prob=0.4, theta_sib=0.1, Decoding Setting=Deterministic2024.07 | 62.1 | 1.16 |