gmq-zle

opusTCv20210807+pbt_transformer-big_2022-03-14.zip

• dataset: opusTCv20210807+pbt
• model: transformer-big
• source language(s): dan isl nno nob swe
• target language(s): bel bel_Latn rus ukr
• raw source language(s): dan isl nno nob swe
• raw target language(s): bel rus ukr
• model: transformer-big
• 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:
• download: opusTCv20210807+pbt_transformer-big_2022-03-14.zip
• test set translations: opusTCv20210807+pbt_transformer-big_2022-03-14.test.txt
• test set scores: opusTCv20210807+pbt_transformer-big_2022-03-14.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
Tatoeba-test-v2021-08-07.dan-bel	49.3	0.83521	2	8	1.000
Tatoeba-test-v2021-08-07.dan-rus	52.1	0.71417	1713	10437	0.999
Tatoeba-test-v2021-08-07.dan-ukr	72.0	0.85449	10	47	1.000
Tatoeba-test-v2021-08-07.isl-rus	26.5	0.47607	139	897	1.000
Tatoeba-test-v2021-08-07.multi-multi	47.1	0.67550	5108	33857	0.994
Tatoeba-test-v2021-08-07.nno-rus	42.0	0.76344	2	11	1.000
Tatoeba-test-v2021-08-07.nob-rus	44.8	0.66354	1277	10631	0.987
Tatoeba-test-v2021-08-07.nor-bel	39.3	0.54588	9	84	1.000
Tatoeba-test-v2021-08-07.nor-rus	44.8	0.66362	1279	10642	0.987
Tatoeba-test-v2021-08-07.nor-ukr	40.7	0.64450	670	3966	1.000
Tatoeba-test-v2021-08-07.swe-rus	43.6	0.65169	1282	7621	0.991
Tatoeba-test-v2021-08-07.swe-ukr	19.1	0.48887	4	19	1.000

opusTCv20210807+pbt_transformer-big_2022-03-19.zip

• dataset: opusTCv20210807+pbt
• model: transformer-big
• source language(s): dan isl nno nob swe
• target language(s): bel bel_Latn rus ukr
• raw source language(s): dan isl nno nob swe
• raw target language(s): bel rus ukr
• model: transformer-big
• 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:
• download: opusTCv20210807+pbt_transformer-big_2022-03-19.zip
• test set translations: opusTCv20210807+pbt_transformer-big_2022-03-19.test.txt
• test set scores: opusTCv20210807+pbt_transformer-big_2022-03-19.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
Tatoeba-test-v2021-08-07.dan-bel	49.3	0.83521	2	8	1.000
Tatoeba-test-v2021-08-07.dan-rus	52.6	0.71760	1713	10437	0.999
Tatoeba-test-v2021-08-07.dan-ukr	66.4	0.81409	10	47	1.000
Tatoeba-test-v2021-08-07.isl-rus	25.8	0.47494	139	897	1.000
Tatoeba-test-v2021-08-07.multi-multi	46.8	0.67414	5108	33857	0.995
Tatoeba-test-v2021-08-07.nno-rus	42.0	0.76344	2	11	1.000
Tatoeba-test-v2021-08-07.nob-rus	44.5	0.66449	1277	10631	0.988
Tatoeba-test-v2021-08-07.nor-bel	36.6	0.57465	9	84	1.000
Tatoeba-test-v2021-08-07.nor-rus	44.5	0.66457	1279	10642	0.988
Tatoeba-test-v2021-08-07.nor-ukr	40.9	0.64550	670	3966	1.000
Tatoeba-test-v2021-08-07.swe-rus	44.9	0.65749	1282	7621	0.989
Tatoeba-test-v2021-08-07.swe-ukr	23.4	0.42115	4	19	1.000

opusTCv20210807+pbt_transformer-big_2022-03-23.zip

• dataset: opusTCv20210807+pbt
• model: transformer-big
• source language(s): dan isl nno nob swe
• target language(s): bel bel_Latn rus ukr
• raw source language(s): dan isl nno nob swe
• raw target language(s): bel rus ukr
• model: transformer-big
• 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:
• download: opusTCv20210807+pbt_transformer-big_2022-03-23.zip
• test set translations: opusTCv20210807+pbt_transformer-big_2022-03-23.test.txt
• test set scores: opusTCv20210807+pbt_transformer-big_2022-03-23.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
Tatoeba-test-v2021-08-07.dan-bel	49.3	0.83521	2	8	1.000
Tatoeba-test-v2021-08-07.dan-rus	52.9	0.72001	1713	10437	0.997
Tatoeba-test-v2021-08-07.dan-ukr	66.4	0.81409	10	47	1.000
Tatoeba-test-v2021-08-07.isl-rus	27.2	0.49824	139	897	1.000
Tatoeba-test-v2021-08-07.multi-multi	47.4	0.67644	5108	33857	0.991
Tatoeba-test-v2021-08-07.nno-rus	42.0	0.76344	2	11	1.000
Tatoeba-test-v2021-08-07.nob-rus	45.4	0.66602	1277	10631	0.984
Tatoeba-test-v2021-08-07.nor-bel	43.9	0.62109	9	84	1.000
Tatoeba-test-v2021-08-07.nor-rus	45.4	0.66610	1279	10642	0.984
Tatoeba-test-v2021-08-07.nor-ukr	40.8	0.63862	670	3966	1.000
Tatoeba-test-v2021-08-07.swe-rus	44.9	0.65799	1282	7621	0.989
Tatoeba-test-v2021-08-07.swe-ukr	26.7	0.43262	4	19	0.946

opusTCv20210807+xb+bt+pbt_transformer-big_2022-09-15.zip

• dataset: opusTCv20210807+xb+bt+pbt
• model: transformer-big
• source language(s): dan isl nno nob swe
• target language(s): bel bel_Latn rus ukr
• raw source language(s): dan isl nno nob swe
• raw target language(s): bel rus ukr
• model: transformer-big
• 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:
• download: opusTCv20210807+xb+bt+pbt_transformer-big_2022-09-15.zip
• test set translations: opusTCv20210807+xb+bt+pbt_transformer-big_2022-09-15.test.txt
• test set scores: opusTCv20210807+xb+bt+pbt_transformer-big_2022-09-15.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
flores101.dan-ukr	26.3	0.54752	1012	22810	0.992
flores101.swe-ukr	26.3	0.54726	1012	22810	0.980
Tatoeba-test-v2021-08-07.multi-multi	47.8	0.68059	5108	33857	0.994