opusTCv20210807+bt_transformer-big_2022-03-09.yml

release: hun-eng/opusTCv20210807+bt_transformer-big_2022-03-09.zip
release-date: 2022-03-09
dataset-name: opusTCv20210807+bt
modeltype: transformer-big
vocabulary:
   source: opusTCv20210807+bt.spm32k-spm32k.vocab.yml
   target: opusTCv20210807+bt.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:
   - hun
target-languages:
   - eng
raw-source-languages:
   - hun
raw-target-languages:
   - eng
training-data:
   eng-hun: Tatoeba-train-v2021-08-07.eng-hun.strict (107829720) wiki.aa.hun-eng (973640) wiki.ab.hun-eng (973430) wiki.ac.hun-eng (973452) wiki.ad.hun-eng (973459) wiki.ae.hun-eng (973446) wiki.af.hun-eng (973808) wiki.ag.hun-eng (973751) wiki.ah.hun-eng (847756) wikibooks.aa.hun-eng (172375) wikinews.aa.hun-eng (8746) wikiquote.aa.hun-eng (61771) wikisource.aa.hun-eng (653918) 
validation-data:
   eng-hun: Tatoeba-dev-v2021-08-07, 92504
   total-size-shuffled: 92504
   devset-selected: top 5000  lines of Tatoeba-dev-v2021-08-07.src.shuffled
test-data:
   newssyscomb2009.hun-eng: 502/11821
   newstest2009.hun-eng: 2525/65402
   Tatoeba-test-v2021-08-07.hun-eng: 10000/72821
BLEU-scores:
   newssyscomb2009.hun-eng: 24.7
   newstest2009.hun-eng: 23.4
   Tatoeba-test-v2021-08-07.hun-eng: 50.8
chr-F-scores:
   newssyscomb2009.hun-eng: 0.52550
   newstest2009.hun-eng: 0.51701
   Tatoeba-test-v2021-08-07.hun-eng: 0.67001