Machine Translation on WMT English-German '14 (val)
3.69PerplexityEvolved Transformer
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Evolved TransformerSize=DEEP, Parameters=218.1M2019.01 | 3.69 | — | — | — | |
| Evolved TransformerSize=BIG, Parameters=221.7M2019.01 | 3.77 | — | — | — | |
| TransformerSize=DEEP, Parameters=224.0M2019.01 | 3.86 | — | — | — | |
| TransformerSize=BIG, Parameters=210.4M2019.01 | 3.87 | — | — | — | |
| Evolved TransformerSize=BASE, Parameters=64.1M2019.01 | 4.03 | — | — | — | |
| TransformerSize=BASE, Parameters=61.1M2019.01 | 4.24 | — | — | — | |
| Transformer (+SRU)number of layers=4, Model Size=79m2017.09 | — | 26.7 | 22,000 | 1.8 | |
| Transformer (+SRU)number of layers=5, Model Size=90m2017.09 | — | 27.1 | 19,000 | 2.1 | |
| Transformer (base)number of layers=6, Model Size=76m2017.09 | — | 26.6 | 20,000 | 2 |