opus-2020-07-04.yml

release: gmq-eng/opus-2020-07-04.zip
release-date: 2020-07-04
dataset-name: opus
modeltype: transformer
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   source: spm32k
   target: spm32k
subword-models:
   source: source.spm
   target: target.spm
source-languages:
   - dan
   - fao
   - isl
   - non
   - swe
target-languages:
   - eng
test-data:
   Tatoeba-test.dan-eng: 10000/62015
   Tatoeba-test.fao-eng: 294/1631
   Tatoeba-test.isl-eng: 2500/16860
   Tatoeba-test.multi-eng: 10000/59565
   Tatoeba-test.non-eng: 15/112
   Tatoeba-test.swe-eng: 10000/54997
BLEU-scores:
   Tatoeba-test.dan-eng: 59.9
   Tatoeba-test.fao-eng: 22.0
   Tatoeba-test.isl-eng: 49.9
   Tatoeba-test.multi-eng: 58.9
   Tatoeba-test.non-eng: 29.8
   Tatoeba-test.swe-eng: 61.0
chr-F-scores:
   Tatoeba-test.dan-eng: 0.741
   Tatoeba-test.fao-eng: 0.419
   Tatoeba-test.isl-eng: 0.658
   Tatoeba-test.multi-eng: 0.728
   Tatoeba-test.non-eng: 0.462
   Tatoeba-test.swe-eng: 0.739