opus-2021-02-23.yml

release: hbs-ukr/opus-2021-02-23.zip
release-date: 2021-02-23
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:
   - bos
   - hbs
   - hrv
   - srp
target-languages:
   - ukr
training-data:
   - bos_Cyrl-ukr: Tatoeba-train (2) 
   - bos_Latn-ukr: Tatoeba-train (204582) 
   - hbs_Cyrl-ukr: Tatoeba-train (11) 
   - hrv-ukr: Tatoeba-train (1398103) 
   - srp_Cyrl-ukr: Tatoeba-train (152997) 
   - srp_Latn-ukr: Tatoeba-train (381969) 
validation-data:
   - bos_Latn-ukr: Tatoeba-dev, 95
   - hrv-ukr: Tatoeba-dev, 628
   - srp_Cyrl-ukr: Tatoeba-dev, 74
   - srp_Latn-ukr: Tatoeba-dev, 202
   - total size of shuffled dev data: 997
   - devset = top 997  lines of Tatoeba-dev.src.shuffled!
test-data:
   - Tatoeba-test.hrv-ukr: 389/2217
   - Tatoeba-test.hbs-ukr: 941/4959
   - Tatoeba-test.srp_Cyrl-ukr: 204/1051
   - Tatoeba-test.srp_Latn-ukr: 348/1658
BLEU-scores:
   - Tatoeba-test.hrv-ukr: 41.4
   - Tatoeba-test.hbs-ukr: 46.4
   - Tatoeba-test.srp_Cyrl-ukr: 48.8
   - Tatoeba-test.srp_Latn-ukr: 50.8
chr-F-scores:
   - Tatoeba-test.hrv-ukr: 0.624
   - Tatoeba-test.hbs-ukr: 0.653
   - Tatoeba-test.srp_Cyrl-ukr: 0.652
   - Tatoeba-test.srp_Latn-ukr: 0.691