Skip to content

Latest commit

 

History

History
66 lines (57 loc) · 4.46 KB

README.md

File metadata and controls

66 lines (57 loc) · 4.46 KB

opus1m+bt-2021-04-10.zip

  • dataset: opus1m+bt
  • model: transformer-align
  • source language(s): eng
  • target language(s): afh avk bzt dws epo ido ile ina jbo ldn lfn nov qya sjn tlh tzl vol
  • model: transformer-align
  • pre-processing: normalization + SentencePiece (spm32k,spm32k)
  • a sentence initial language token is required in the form of >>id<< (id = valid target language ID)
  • valid language labels: >>afh<< >>afh_Latn<< >>avk<< >>avk_Latn<< >>bzt<< >>bzt_Latn<< >>dws<< >>dws_Latn<< >>epo<< >>ido<< >>ido_Latn<< >>igs<< >>ile<< >>ile_Latn<< >>ina<< >>ina_Latn<< >>jbo<< >>jbo_Cyrl<< >>jbo_Latn<< >>ldn<< >>ldn_Latn<< >>lfn<< >>lfn_Cyrl<< >>lfn_Latn<< >>neu<< >>nov<< >>nov_Latn<< >>qya<< >>qya_Latn<< >>rmv<< >>sjn<< >>sjn_Latn<< >>tlh<< >>tlh_Latn<< >>tzl<< >>tzl_Latn<< >>vol<< >>vol_Latn<< >>zbl<<
  • download: opus1m+bt-2021-04-10.zip
  • test set translations: opus1m+bt-2021-04-10.test.txt
  • test set scores: opus1m+bt-2021-04-10.eval.txt

Benchmarks

testset BLEU chr-F #sent #words BP
Tatoeba-test.eng-afh 1.5 0.108 10 46 1.000
Tatoeba-test.eng-avk 0.3 0.128 167 970 1.000
Tatoeba-test.eng-bzt 0.9 0.136 62 354 1.000
Tatoeba-test.eng-dws 0.9 0.107 10 40 1.000
Tatoeba-test.eng-epo 36.5 0.593 10000 76402 0.997
Tatoeba-test.eng-ido 5.4 0.309 1968 13078 1.000
Tatoeba-test.eng-ido_Latn 5.4 0.309 1967 13072 1.000
Tatoeba-test.eng-ile 0.7 0.115 1711 10655 0.832
Tatoeba-test.eng-ina 5.3 0.266 5000 44642 0.973
Tatoeba-test.eng-jbo 0.2 0.117 5000 35293 1.000
Tatoeba-test.eng-jbo_Cyrl 1.5 0.000 1 9 1.000
Tatoeba-test.eng-jbo_Latn 0.2 0.117 4996 35278 1.000
Tatoeba-test.eng-ldn 0.3 0.080 101 630 0.953
Tatoeba-test.eng-lfn 1.6 0.167 3297 24468 0.914
Tatoeba-test.eng-lfn_Cyrl 0.1 0.008 847 6075 0.975
Tatoeba-test.eng-lfn_Latn 2.0 0.220 2450 18393 0.893
Tatoeba-test.eng-multi 12.1 0.308 10000 69052 1.000
Tatoeba-test.eng-nov 1.7 0.263 198 1303 1.000
Tatoeba-test.eng-qya 0.8 0.114 116 485 1.000
Tatoeba-test.eng-qya_Latn 0.8 0.116 115 481 1.000
Tatoeba-test.eng-sjn 0.4 0.095 44 196 1.000
Tatoeba-test.eng-tlh 0.0 0.130 5000 21301 1.000
Tatoeba-test.eng-tzl 0.5 0.123 166 642 1.000
Tatoeba-test.eng-tzl_Latn 0.5 0.123 165 640 1.000
Tatoeba-test.eng-vol 0.3 0.128 1549 7884 1.000

opus4m+btTCv20210807-2021-09-30.zip

  • dataset: opus4m+btTCv20210807
  • model: transformer
  • source language(s): eng
  • target language(s): afh avk bzt dws epo ido ile ina jbo ldn lfn nov qya sjn tlh tzl vol
  • model: transformer
  • pre-processing: normalization + SentencePiece (spm32k,spm32k)
  • a sentence initial language token is required in the form of >>id<< (id = valid target language ID)
  • valid language labels: >>afh<< >>afh_Latn<< >>avk<< >>avk_Latn<< >>bzt<< >>bzt_Latn<< >>dws<< >>dws_Latn<< >>epo<< >>ido<< >>ido_Latn<< >>igs<< >>ile<< >>ile_Latn<< >>ina<< >>ina_Latn<< >>jbo<< >>jbo_Cyrl<< >>jbo_Latn<< >>ldn<< >>ldn_Latn<< >>lfn<< >>lfn_Cyrl<< >>lfn_Latn<< >>neu<< >>nov<< >>nov_Latn<< >>qya<< >>qya_Latn<< >>rmv<< >>sjn<< >>sjn_Latn<< >>tlh<< >>tlh_Latn<< >>tzl<< >>tzl_Latn<< >>vol<< >>vol_Latn<< >>zbl<<
  • download: opus4m+btTCv20210807-2021-09-30.zip
  • test set translations: opus4m+btTCv20210807-2021-09-30.test.txt
  • test set scores: opus4m+btTCv20210807-2021-09-30.eval.txt

Benchmarks

testset BLEU chr-F #sent #words BP
Tatoeba-test-v2021-08-07.eng-multi 22.2 0.430 10000 71346 1.000
Tatoeba-test-v2021-08-07.multi-multi 22.2 0.430 10000 71346 1.000