opus-2021-02-23.yml

release: ukr-hbs/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
use-target-labels:
   - ">>bos_Cyrl<<"
   - ">>bos_Latn<<"
   - ">>hbs_Cyrl<<"
   - ">>hrv<<"
   - ">>srp_Cyrl<<"
   - ">>srp_Latn<<"
source-languages:
   - ukr
target-languages:
   - bos
   - hbs
   - hrv
   - srp
training-data:
   ukr-bos_Cyrl: Tatoeba-train (2) 
   ukr-bos_Latn: Tatoeba-train (204582) 
   ukr-hbs_Cyrl: Tatoeba-train (11) 
   ukr-hrv: Tatoeba-train (1398103) 
   ukr-srp_Cyrl: Tatoeba-train (152997) 
   ukr-srp_Latn: 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-shuffled: 998
   devset-selected: top 998  lines of Tatoeba-dev.src.shuffled!
test-data:
   Tatoeba-test.ukr-hbs: 941/5128
   Tatoeba-test.ukr-hrv: 389/2302
   Tatoeba-test.ukr-srp_Cyrl: 204/1110
   Tatoeba-test.ukr-srp_Latn: 348/1716
BLEU-scores:
   Tatoeba-test.ukr-hbs: 44.5
   Tatoeba-test.ukr-hrv: 43.8
   Tatoeba-test.ukr-srp_Cyrl: 47.3
   Tatoeba-test.ukr-srp_Latn: 42.4
chr-F-scores:
   Tatoeba-test.ukr-hbs: 0.643
   Tatoeba-test.ukr-hrv: 0.635
   Tatoeba-test.ukr-srp_Cyrl: 0.641
   Tatoeba-test.ukr-srp_Latn: 0.651