opusTCv20210807_transformer-big_2022-06-24.yml

release: zle-fra/opusTCv20210807_transformer-big_2022-06-24.zip
release-date: 2022-06-24
dataset-name: opusTCv20210807
modeltype: transformer-big
vocabulary:
   source: opusTCv20210807.spm32k-spm32k.vocab.yml
   target: opusTCv20210807.spm32k-spm32k.vocab.yml
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   source: spm32k
   target: spm32k
subword-models:
   source: source.spm
   target: target.spm
source-languages:
   - bel
   - bel_Latn
   - orv_Cyrl
   - rus
   - ukr
target-languages:
   - fra
raw-source-languages:
   - bel
   - orv
   - rus
   - ukr
raw-target-languages:
   - fra
training-data:
   bel-fra: Tatoeba-train-v2021-08-07.bel-fra.strict (666499) 
   bel_Latn-fra: Tatoeba-train-v2021-08-07.bel_Latn-fra.strict (1665) 
   rus-fra: Tatoeba-train-v2021-08-07.fra-rus.strict (104720380) 
   ukr-fra: Tatoeba-train-v2021-08-07.fra-ukr.strict (11234086) 
validation-data:
   bel-fra: Tatoeba-dev-v2021-08-07, 954
   bel_Latn-fra: Tatoeba-dev-v2021-08-07, 4
   fra-rus: Tatoeba-dev-v2021-08-07, 176856
   fra-ukr: Tatoeba-dev-v2021-08-07, 17548
   total-size-shuffled: 2958
   devset-selected: top 2958  lines of Tatoeba-dev-v2021-08-07.src.shuffled
test-data:
   newstest2012.rus-fra: 3003/78011
   newstest2013.rus-fra: 3000/70037
   Tatoeba-test-v2021-08-07.bel-fra: 283/2005
   Tatoeba-test-v2021-08-07.multi-fra: 10000/66671
   Tatoeba-test-v2021-08-07.orv-fra: 37/290
   Tatoeba-test-v2021-08-07.rus-fra: 11490/80573
   Tatoeba-test-v2021-08-07.ukr-fra: 10035/63222
BLEU-scores:
   newstest2012.rus-fra: 25.4
   newstest2013.rus-fra: 29.9
   Tatoeba-test-v2021-08-07.bel-fra: 46.4
   Tatoeba-test-v2021-08-07.multi-fra: 52.3
   Tatoeba-test-v2021-08-07.orv-fra: 6.3
   Tatoeba-test-v2021-08-07.rus-fra: 51.8
   Tatoeba-test-v2021-08-07.ukr-fra: 50.7
chr-F-scores:
   newstest2012.rus-fra: 0.53678
   newstest2013.rus-fra: 0.56394
   Tatoeba-test-v2021-08-07.bel-fra: 0.65415
   Tatoeba-test-v2021-08-07.multi-fra: 0.68348
   Tatoeba-test-v2021-08-07.orv-fra: 0.22337
   Tatoeba-test-v2021-08-07.rus-fra: 0.68699
   Tatoeba-test-v2021-08-07.ukr-fra: 0.67887