Machine Translation on WMT En-Ro
35.28BLEURK2-block-Big (learnable gamma_i)
Evaluation Results
| Method | Links | |
|---|---|---|
| RK2-block-Big (learnable gamma_i)Params=226M, Epochs=202022.03 | 35.28 | |
| RK2-block (learnable gamma_i)Params=69M, Epochs=202022.03 | 34.94 | |
| DeLight (Mehta et al., 2020)Params=53M, Epochs=1702022.03 | 34.7 | |
| Transformer (Mehta et al., 2020)Params=62M, Epochs=1702022.03 | 34.3 | |
| Transformer (Our impl.)Params=69M, Epochs=202022.03 | 33.49 | |
| Int Transformer (Lin et al., 2020)2022.03 | 32.6 |