Machine Translation on WMT English-French (test)
43.8BLEUPrevious best
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Previous best2019.10 | 43.8 | — | — | |
| T5-11BParameters=11 billion2019.10 | 43.4 | — | — | |
| T5-3BParameters=3 billion2019.10 | 42.6 | — | — | |
| T5-LargeParameters=770 million2019.10 | 41.5 | — | — | |
| T5-BaseParameters=220 million2019.10 | 41.2 | — | — | |
| T5-SmallParameters=60 million2019.10 | 36 | — | — | |
| Extended Neural GPUm=5122016.10 | 29.6 | 3.3 | 1.19 | |
| GRU-Attentiondecoder=full beam-search decoder, beam size=12, length normalization=true, attention coverage penalty=true2016.10 | 26.4 | 3.4 | 1.22 | |
| Markovian Neural GPU2016.10 | — | 11.8 | 2.5 | |
| Neural GPU2016.10 | — | 30.1 | 3.5 |