WMT
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
WMT En-Ro 2016 (test)
35.4BLEU
39
WMT14 English-French (newstest2014)
45.9BLEU
39
WMT16 German-English (test)
40.6BLEU
39
WMT Ro-En '16
38.5BLEU Score
37
WMT En-De 2019 (test)
93SacreBLEU
37
WMT-23
2.81Speedup
32
WMT En-Fr 14
45.4BLEU
32
WMT newstest (test)
27.5Error Rate
32
WMT newstest 2016 (test)
40.4BLEU
31
WMT En-Ro newstest2016 (test)
35.2BLEU
31
WMT newstest 2015 (test)
34.93BLEU
31
WMT newstest 2014
37.5Tokenized BLEU
30
WMT De-En 22 (test)
88.14COMET
29
WMT English-German (EN-DE) 2014 (newstest2014)
30.77BLEU
29
WMT 14
27.5BLEU
28
WMT 19
0.608COMET AUC
28
WMT19
0.346COMET Score
28
WMT MQM (En-De, En-Es, Ja-Zh) 24
87.9SPA
28
WMT newstest (test)
35.41BLEU
28
WMT14 DE-EN (test)
32.7BLEU
28
WMT EN-RO 2016
38.3BLEU
28
WMT16 Ro-En (test)
38.1BLEU
27
WMT de-en
34.04BLEU
27
WMT German-English 16 (test)
2.496Speedup ratio
26
WMT EN-ZH 24
80COMETkiwi
26