Skip to content

Latest commit

 

History

History
408 lines (296 loc) · 18.7 KB

README.md

File metadata and controls

408 lines (296 loc) · 18.7 KB

opus-tuned4bel2deu-2021-01-15.zip

  • dataset: opus-tuned4bel2deu
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4bel2deu-2021-01-15.zip
  • test set translations: opus-tuned4bel2deu-2021-01-15.test.txt
  • test set scores: opus-tuned4bel2deu-2021-01-15.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.bel-deu.bel.deu 37.8 0.584
Tatoeba-test.multi.multi 46.9 0.636

opus-tuned4bel2nld-2021-01-15.zip

  • dataset: opus-tuned4bel2nld
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4bel2nld-2021-01-15.zip
  • test set translations: opus-tuned4bel2nld-2021-01-15.test.txt
  • test set scores: opus-tuned4bel2nld-2021-01-15.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.bel-nld.bel.nld 33.8 0.532
Tatoeba-test.multi.multi 47.9 0.642

opus-tuned4rus2nld-2021-01-17.zip

  • dataset: opus-tuned4rus2nld
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4rus2nld-2021-01-17.zip
  • test set translations: opus-tuned4rus2nld-2021-01-17.test.txt
  • test set scores: opus-tuned4rus2nld-2021-01-17.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.multi.multi 5.8 0.212
Tatoeba-test.rus-nld.rus.nld 47.3 0.658

opus-tuned4bel2eng-2021-01-17.zip

  • dataset: opus-tuned4bel2eng
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4bel2eng-2021-01-17.zip
  • test set translations: opus-tuned4bel2eng-2021-01-17.test.txt
  • test set scores: opus-tuned4bel2eng-2021-01-17.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.bel-eng.bel.eng 39.5 0.570
Tatoeba-test.multi.multi 38.9 0.550

opus-tuned4ukr2afr-2021-01-17.zip

  • dataset: opus-tuned4ukr2afr
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4ukr2afr-2021-01-17.zip
  • test set translations: opus-tuned4ukr2afr-2021-01-17.test.txt
  • test set scores: opus-tuned4ukr2afr-2021-01-17.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.multi.multi 1.4 0.165
Tatoeba-test.ukr-afr.ukr.afr 54.6 0.723

opus-tuned4ukr2deu-2021-01-17.zip

  • dataset: opus-tuned4ukr2deu
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4ukr2deu-2021-01-17.zip
  • test set translations: opus-tuned4ukr2deu-2021-01-17.test.txt
  • test set scores: opus-tuned4ukr2deu-2021-01-17.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.multi.multi 12.7 0.279
Tatoeba-test.ukr-deu.ukr.deu 50.7 0.678

opus-tuned4rus2eng-2021-01-17.zip

  • dataset: opus-tuned4rus2eng
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4rus2eng-2021-01-17.zip
  • test set translations: opus-tuned4rus2eng-2021-01-17.test.txt
  • test set scores: opus-tuned4rus2eng-2021-01-17.eval.txt

Benchmarks

testset BLEU chr-F
newstest2012-ruseng.rus.eng 34.2 0.601
newstest2013-ruseng.rus.eng 27.5 0.543
newstest2014-ruen-ruseng.rus.eng 30.9 0.588
newstest2015-enru-ruseng.rus.eng 29.7 0.564
newstest2016-enru-ruseng.rus.eng 29.2 0.562
newstest2017-enru-ruseng.rus.eng 32.8 0.588
newstest2018-enru-ruseng.rus.eng 28.6 0.561
newstest2019-ruen-ruseng.rus.eng 31.0 0.577
Tatoeba-test.multi.multi 46.2 0.627
Tatoeba-test.rus-eng.rus.eng 56.6 0.707

opus-tuned4ukr2eng-2021-01-17.zip

  • dataset: opus-tuned4ukr2eng
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4ukr2eng-2021-01-17.zip
  • test set translations: opus-tuned4ukr2eng-2021-01-17.test.txt
  • test set scores: opus-tuned4ukr2eng-2021-01-17.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.multi.multi 16.5 0.279
Tatoeba-test.ukr-eng.ukr.eng 53.5 0.683

opus-tuned4rus2afr-2021-01-17.zip

  • dataset: opus-tuned4rus2afr
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4rus2afr-2021-01-17.zip
  • test set translations: opus-tuned4rus2afr-2021-01-17.test.txt
  • test set scores: opus-tuned4rus2afr-2021-01-17.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.multi.multi 1.2 0.166
Tatoeba-test.rus-afr.rus.afr 51.7 0.693

opus-tuned4rus2deu-2021-01-17.zip

  • dataset: opus-tuned4rus2deu
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4rus2deu-2021-01-17.zip
  • test set translations: opus-tuned4rus2deu-2021-01-17.test.txt
  • test set scores: opus-tuned4rus2deu-2021-01-17.eval.txt

Benchmarks

testset BLEU chr-F
newstest2012-rusdeu.rus.deu 15.1 0.451
newstest2013-rusdeu.rus.deu 19.0 0.487
Tatoeba-test.multi.multi 13.1 0.290
Tatoeba-test.rus-deu.rus.deu 47.5 0.661

opus-tuned4ukr2nld-2021-01-18.zip

  • dataset: opus-tuned4ukr2nld
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm frr fry gos gsw ksh lim ltz nds nld ofs osx pdc pfl sco stq swg wae yid zea
  • 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)
  • download: opus-tuned4ukr2nld-2021-01-18.zip
  • test set translations: opus-tuned4ukr2nld-2021-01-18.test.txt
  • test set scores: opus-tuned4ukr2nld-2021-01-18.eval.txt

Benchmarks

testset BLEU chr-F
Tatoeba-test.multi.multi 10.2 0.238
Tatoeba-test.ukr-nld.ukr.nld 50.7 0.673

opus1m-2021-02-16.zip

  • dataset: opus1m
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang bar bis bzj deu djk drt eng enm frr fry gos gsw hrx jam kri ksh lim ltz nds nld ofs osx pcm pdc pdt pfl pih pis sco srm srn stq swg tpi vls wae wes yid zea
  • 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)
  • download: opus1m-2021-02-16.zip
  • test set translations: opus1m-2021-02-16.test.txt
  • test set scores: opus1m-2021-02-16.eval.txt

Benchmarks

testset BLEU chr-F
newstest2012-rusdeu.rus.deu 15.1 0.451
newstest2012-ruseng.rus.eng 34.0 0.601
newstest2013-rusdeu.rus.deu 18.6 0.484
newstest2013-ruseng.rus.eng 27.5 0.542
newstest2014-ruen-ruseng.rus.eng 30.9 0.589
newstest2015-enru-ruseng.rus.eng 29.6 0.564
newstest2016-enru-ruseng.rus.eng 29.1 0.562
newstest2017-enru-ruseng.rus.eng 32.7 0.589
newstest2018-enru-ruseng.rus.eng 28.4 0.561
newstest2019-ruen-ruseng.rus.eng 31.3 0.579
Tatoeba-test.bel-deu.bel.deu 34.9 0.550
Tatoeba-test.bel-eng.bel.eng 34.8 0.528
Tatoeba-test.bel-nld.bel.nld 31.5 0.509
Tatoeba-test.bel-yid.bel.yid 2.5 0.092
Tatoeba-test.multi.multi 48.9 0.655
Tatoeba-test.orv-deu.orv.deu 6.4 0.264
Tatoeba-test.orv-eng.orv.eng 7.2 0.216
Tatoeba-test.rue-eng.rue.eng 16.9 0.346
Tatoeba-test.rus-afr.rus.afr 47.1 0.666
Tatoeba-test.rus-ang.rus.ang 1.9 0.057
Tatoeba-test.rus-deu.rus.deu 46.7 0.655
Tatoeba-test.rus-eng.rus.eng 56.7 0.709
Tatoeba-test.rus-enm.rus.enm 6.4 0.356
Tatoeba-test.rus-fry.rus.fry 3.8 0.250
Tatoeba-test.rus-gos.rus.gos 3.7 0.203
Tatoeba-test.rus-ltz.rus.ltz 9.7 0.376
Tatoeba-test.rus-nds.rus.nds 7.3 0.308
Tatoeba-test.rus-nld.rus.nld 47.4 0.657
Tatoeba-test.rus-yid.rus.yid 0.2 0.070
Tatoeba-test.ukr-afr.ukr.afr 55.4 0.724
Tatoeba-test.ukr-ang.ukr.ang 12.7 0.279
Tatoeba-test.ukr-deu.ukr.deu 50.2 0.674
Tatoeba-test.ukr-eng.ukr.eng 52.7 0.676
Tatoeba-test.ukr-enm.ukr.enm 27.9 0.635
Tatoeba-test.ukr-fry.ukr.fry 3.5 0.228
Tatoeba-test.ukr-gos.ukr.gos 7.4 0.127
Tatoeba-test.ukr-nds.ukr.nds 6.1 0.350
Tatoeba-test.ukr-nld.ukr.nld 50.6 0.672
Tatoeba-test.ukr-yid.ukr.yid 0.6 0.069

opus1m-2021-02-18.zip

  • dataset: opus1m
  • model: transformer
  • source language(s): bel orv rue rus ukr
  • target language(s): afr ang deu eng enm fry gos ltz nds nld yid
  • 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: >>eng<< >>nld<< >>deu<< >>afr<< >>nds<< >>fry<< >>ang_Latn<< >>ltz<< >>yid<<
  • download: opus1m-2021-02-18.zip
  • test set translations: opus1m-2021-02-18.test.txt
  • test set scores: opus1m-2021-02-18.eval.txt

Benchmarks

testset BLEU chr-F #sent #words BP
newstest2012.rus-deu 15.1 0.451 3003 72886 0.959
newstest2012.rus-eng 34.0 0.601 3003 72812 0.984
newstest2013.rus-deu 18.6 0.484 3000 63737 0.982
newstest2013.rus-eng 27.5 0.542 3000 64505 0.998
newstest2014-ruen.rus-eng 30.9 0.589 3003 69190 0.993
newstest2015-enru.rus-eng 29.6 0.564 2818 64744 0.952
newstest2016-enru.rus-eng 29.1 0.562 2998 69278 0.987
newstest2017-enru.rus-eng 32.7 0.589 3001 69033 0.971
newstest2018-enru.rus-eng 28.5 0.561 3000 71723 0.978
newstest2019-ruen.rus-eng 31.3 0.579 2000 42875 0.976
Tatoeba-test.bel-deu 34.8 0.548 550 4171 0.994
Tatoeba-test.bel-eng 34.6 0.527 2500 18567 1.000
Tatoeba-test.bel_Latn-deu 2.6 0.089 3 21 0.951
Tatoeba-test.bel_Latn-eng 2.0 0.116 3 26 1.000
Tatoeba-test.bel-nld 31.3 0.507 606 4805 0.985
Tatoeba-test.bel-yid 2.5 0.092 5 29 1.000
Tatoeba-test.multi-multi 48.9 0.655 10000 68229 0.983
Tatoeba-test.orv-deu 6.4 0.264 28 197 1.000
Tatoeba-test.orv-eng 7.2 0.217 322 2102 0.974
Tatoeba-test.rue-eng 16.9 0.345 120 697 0.958
Tatoeba-test.rus-afr 46.9 0.664 228 1390 1.000
Tatoeba-test.rus-ang 1.9 0.057 10 48 0.662
Tatoeba-test.rus-deu 46.5 0.654 10000 76293 0.984
Tatoeba-test.rus-eng 56.6 0.708 10000 72891 0.972
Tatoeba-test.rus-enm 6.4 0.356 15 77 0.960
Tatoeba-test.rus-fry 3.8 0.250 28 189 0.912
Tatoeba-test.rus-gos 3.7 0.203 6 37 1.000
Tatoeba-test.rus-ltz 9.7 0.376 1 6 1.000
Tatoeba-test.rus-nds 7.2 0.309 894 6082 0.945
Tatoeba-test.rus-nld 47.1 0.656 2500 18368 0.964
Tatoeba-test.rus-yid 0.2 0.070 150 999 0.850
Tatoeba-test.ukr-afr 55.4 0.722 123 704 1.000
Tatoeba-test.ukr-ang 12.7 0.279 2 10 1.000
Tatoeba-test.ukr-deu 50.0 0.672 10000 62293 0.993
Tatoeba-test.ukr-eng 52.6 0.675 10000 66113 0.978
Tatoeba-test.ukr-enm 27.9 0.635 11 62 1.000
Tatoeba-test.ukr-fry 2.9 0.217 28 214 0.811
Tatoeba-test.ukr-gos 7.4 0.127 5 23 0.956
Tatoeba-test.ukr-nds 7.3 0.349 68 426 0.940
Tatoeba-test.ukr-nld 50.4 0.670 10000 59934 0.989
Tatoeba-test.ukr-yid 0.6 0.069 28 158 1.000
tico19-test.rus-eng 27.8 0.593 2100 56347 1.000