opusTCv20210807_transformer-big_2022-03-23.yml

release: zle-por/opusTCv20210807_transformer-big_2022-03-23.zip
release-date: 2022-03-23
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
   - rus
   - ukr
target-languages:
   - pob
   - por
raw-source-languages:
   - bel
   - rus
   - ukr
raw-target-languages:
   - pob
   - por
use-target-labels:
training-data:
   bel-pob: Tatoeba-train-v2021-08-07.bel-pob.strict (30436) 
   bel-por: Tatoeba-train-v2021-08-07.bel-por.strict (672750) 
   bel_Latn-pob: Tatoeba-train-v2021-08-07.bel_Latn-pob.strict (61) 
   bel_Latn-por: Tatoeba-train-v2021-08-07.bel_Latn-por.strict (2155) 
   rus-pob: Tatoeba-train-v2021-08-07.pob-rus.strict (20596244) 
   rus-por: Tatoeba-train-v2021-08-07.por-rus.strict (50778621) 
   ukr-pob: Tatoeba-train-v2021-08-07.pob-ukr.strict (691987) 
   ukr-por: Tatoeba-train-v2021-08-07.por-ukr.strict (8426806) 
validation-data:
   bel-pob: Tatoeba-dev-v2021-08-07, 50
   bel-por: Tatoeba-dev-v2021-08-07, 943
   bel_Latn-pob: Tatoeba-dev-v2021-08-07, 1
   bel_Latn-por: Tatoeba-dev-v2021-08-07, 6
   por-rus: Tatoeba-dev-v2021-08-07, 9945
   por-ukr: Tatoeba-dev-v2021-08-07, 1015
   total-size-shuffled: 3000
   devset-selected: top 3000  lines of Tatoeba-dev-v2021-08-07.src.shuffled
test-data:
   Tatoeba-test-v2021-08-07.bel-por: 3/21
   Tatoeba-test-v2021-08-07.multi-multi: 10000/71519
   Tatoeba-test-v2021-08-07.rus-por: 10000/74705
   Tatoeba-test-v2021-08-07.ukr-por: 3372/21301
BLEU-scores:
   Tatoeba-test-v2021-08-07.bel-por: 65.6
   Tatoeba-test-v2021-08-07.multi-multi: 43.1
   Tatoeba-test-v2021-08-07.rus-por: 42.3
   Tatoeba-test-v2021-08-07.ukr-por: 44.6
chr-F-scores:
   Tatoeba-test-v2021-08-07.bel-por: 0.74422
   Tatoeba-test-v2021-08-07.multi-multi: 0.63939
   Tatoeba-test-v2021-08-07.rus-por: 0.63500
   Tatoeba-test-v2021-08-07.ukr-por: 0.65006