opus1m-2021-05-15.yml

release: fiu-fin/opus1m-2021-05-15.zip
release-date: 2021-05-15
dataset-name: opus1m
modeltype: transformer-align
vocabulary:
   source: opus1m.spm12k-spm12k.vocab.yml
   target: opus1m.spm12k-spm12k.vocab.yml
pre-processing: normalization + SentencePiece (spm12k,spm12k)
subwords:
   source: spm12k
   target: spm12k
subword-models:
   source: source.spm
   target: target.spm
source-languages:
   - est
target-languages:
   - fin
training-data:
   est-fin: Tatoeba-train (1000000) 
validation-data:
   est-fin: Tatoeba-dev, 1000
   total-size-shuffled: 993
   devset-selected: top 993  lines of Tatoeba-dev.src.shuffled
test-data:
   Tatoeba-test.est-fin: 189/966
   Tatoeba-test.fkv-fin: 297/1664
   Tatoeba-test.hun-fin: 1297/6499
   Tatoeba-test.izh-fin: 3/12
   Tatoeba-test.krl-fin: 29/153
   Tatoeba-test.multi-fin: 189/966
BLEU-scores:
   Tatoeba-test.est-fin: 51.5
   Tatoeba-test.fkv-fin: 5.3
   Tatoeba-test.hun-fin: 0.4
   Tatoeba-test.izh-fin: 6.4
   Tatoeba-test.krl-fin: 6.6
   Tatoeba-test.multi-fin: 51.5
chr-F-scores:
   Tatoeba-test.est-fin: 0.689
   Tatoeba-test.fkv-fin: 0.326
   Tatoeba-test.hun-fin: 0.106
   Tatoeba-test.izh-fin: 0.408
   Tatoeba-test.krl-fin: 0.402
   Tatoeba-test.multi-fin: 0.689