Machine Translation on WMT (en-de, en-fr BLEU)
28.6BLEU (en-de)GHT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GHTParam=44M, Inference Speed=1016.4 sent/s, FLOPs=1996M2023.05 | 28.6 | 40.7 | |
| GHT-PSParam=30M, Inference Speed=1122.1 sent/s, FLOPs=1558M2023.05 | 28.2 | 40.5 | |
| Cordonnier et al. (2020)Param=44M, Inference Speed=416.6 sent/s, FLOPs=2054M2023.05 | 27.6 | 38.5 | |
| Li et al. (2018)Param=44M, Inference Speed=1011.2 sent/s, FLOPs=1996M2023.05 | 27.3 | 39.3 | |
| Lite TransformerParam=30M, Inference Speed=1096.6 sent/s, FLOPs=1809M2023.05 | 24.9 | 37.4 | |
| Dynamic convParam=30M, Inference Speed=1050.2 sent/s, FLOPs=1615M2023.05 | 24 | 36.5 | |
| Voita et al. (2019)Param=30M, Inference Speed=1099.1 sent/s, FLOPs=1558M2023.05 | 22 | 34 | |
| Li et al. (2021)Param=30M, Inference Speed=1116.9 sent/s, FLOPs=1558M2023.05 | 20.5 | 33.6 |