zlw-zle

opusTCv20210807+bt_transformer-big_2022-03-07.zip

• dataset: opusTCv20210807+bt
• model: transformer-big
• source language(s): ces dsb hsb pol
• target language(s): bel bel_Latn orv_Cyrl rus ukr
• raw source language(s): ces dsb hsb pol
• raw target language(s): bel orv 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+bt_transformer-big_2022-03-07.zip
• test set translations: opusTCv20210807+bt_transformer-big_2022-03-07.test.txt
• test set scores: opusTCv20210807+bt_transformer-big_2022-03-07.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newstest2012.ces-rus	20.6	0.49166	3003	64830	0.997
newstest2013.ces-rus	26.8	0.53763	3000	58560	0.973
Tatoeba-test-v2021-08-07.ces-bel	33.8	0.51349	31	181	0.966
Tatoeba-test-v2021-08-07.ces-rus	55.0	0.72626	2934	17743	0.985
Tatoeba-test-v2021-08-07.ces-ukr	51.0	0.68280	1787	8854	0.997
Tatoeba-test-v2021-08-07.dsb-rus	26.9	0.48948	24	124	1.000
Tatoeba-test-v2021-08-07.dsb-ukr	8.8	0.34208	3	13	1.000
Tatoeba-test-v2021-08-07.hsb-rus	19.4	0.40929	38	281	0.859
Tatoeba-test-v2021-08-07.hsb-ukr	3.5	0.16605	8	126	0.585
Tatoeba-test-v2021-08-07.multi-multi	52.8	0.70695	10000	58091	0.987
Tatoeba-test-v2021-08-07.pol-bel	28.7	0.51885	287	1730	1.000
Tatoeba-test-v2021-08-07.pol-bel_Latn	3.8	0.847	2	16	0.794
Tatoeba-test-v2021-08-07.pol-orv	4.5	0.24322	7	31	1.000
Tatoeba-test-v2021-08-07.pol-rus	54.5	0.72518	3543	21971	0.992
Tatoeba-test-v2021-08-07.pol-ukr	48.1	0.67885	2519	13493	0.998

opusTCv20210807+bt_transformer-big_2022-03-19.zip

• dataset: opusTCv20210807+bt
• model: transformer-big
• source language(s): ces dsb hsb pol
• target language(s): bel bel_Latn orv_Cyrl rus ukr
• raw source language(s): ces dsb hsb pol
• raw target language(s): bel orv 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+bt_transformer-big_2022-03-19.zip
• test set translations: opusTCv20210807+bt_transformer-big_2022-03-19.test.txt
• test set scores: opusTCv20210807+bt_transformer-big_2022-03-19.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newstest2012.ces-rus	21.0	0.49476	3003	64830	0.989
newstest2013.ces-rus	27.2	0.54197	3000	58560	0.963
Tatoeba-test-v2021-08-07.ces-bel	33.7	0.51728	31	181	1.000
Tatoeba-test-v2021-08-07.ces-rus	55.6	0.72736	2934	17743	0.979
Tatoeba-test-v2021-08-07.ces-ukr	52.5	0.69628	1787	8854	0.997
Tatoeba-test-v2021-08-07.dsb-rus	23.4	0.48769	24	124	1.000
Tatoeba-test-v2021-08-07.dsb-ukr	9.7	0.29332	3	13	1.000
Tatoeba-test-v2021-08-07.hsb-rus	26.9	0.45080	38	281	0.907
Tatoeba-test-v2021-08-07.hsb-ukr	7.8	0.36314	8	126	0.701
Tatoeba-test-v2021-08-07.multi-multi	53.3	0.71039	10000	58091	0.985
Tatoeba-test-v2021-08-07.pol-bel	28.9	0.50513	287	1730	0.994
Tatoeba-test-v2021-08-07.pol-bel_Latn	4.0	1.122	2	16	0.936
Tatoeba-test-v2021-08-07.pol-orv	4.1	0.25181	7	31	1.000
Tatoeba-test-v2021-08-07.pol-rus	54.3	0.72718	3543	21971	0.989
Tatoeba-test-v2021-08-07.pol-ukr	48.4	0.68147	2519	13493	0.997

opusTCv20210807+bt_transformer-big_2022-03-23.zip

• dataset: opusTCv20210807+bt
• model: transformer-big
• source language(s): ces dsb hsb pol
• target language(s): bel bel_Latn orv_Cyrl rus ukr
• raw source language(s): ces dsb hsb pol
• raw target language(s): bel orv 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+bt_transformer-big_2022-03-23.zip
• test set translations: opusTCv20210807+bt_transformer-big_2022-03-23.test.txt
• test set scores: opusTCv20210807+bt_transformer-big_2022-03-23.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newstest2012.ces-rus	21.2	0.49627	3003	64830	0.987
newstest2013.ces-rus	27.5	0.54297	3000	58560	0.962
Tatoeba-test-v2021-08-07.ces-bel	32.7	0.50195	31	181	1.000
Tatoeba-test-v2021-08-07.ces-rus	55.6	0.72859	2934	17743	0.979
Tatoeba-test-v2021-08-07.ces-ukr	52.3	0.69750	1787	8854	0.996
Tatoeba-test-v2021-08-07.dsb-rus	23.9	0.47404	24	124	1.000
Tatoeba-test-v2021-08-07.dsb-ukr	9.4	0.27363	3	13	1.000
Tatoeba-test-v2021-08-07.hsb-rus	36.2	0.47720	38	281	0.945
Tatoeba-test-v2021-08-07.hsb-ukr	23.0	0.38722	8	126	0.800
Tatoeba-test-v2021-08-07.multi-multi	53.5	0.71203	10000	58091	0.985
Tatoeba-test-v2021-08-07.pol-bel	28.2	0.49950	287	1730	0.999
Tatoeba-test-v2021-08-07.pol-bel_Latn	3.5	4.508	2	16	0.717
Tatoeba-test-v2021-08-07.pol-orv	4.0	0.24981	7	31	1.000
Tatoeba-test-v2021-08-07.pol-rus	54.6	0.72814	3543	21971	0.988
Tatoeba-test-v2021-08-07.pol-ukr	49.0	0.68496	2519	13493	0.997

opusTCv20210807+xb+bt_transformer-big_2022-05-08.zip

• dataset: opusTCv20210807+xb+bt
• model: transformer-big
• source language(s): ces dsb hsb pol slk
• target language(s): bel bel_Latn orv_Cyrl rus ukr
• raw source language(s): ces dsb hsb pol slk
• raw target language(s): bel orv 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_transformer-big_2022-05-08.zip
• test set translations: opusTCv20210807+xb+bt_transformer-big_2022-05-08.test.txt
• test set scores: opusTCv20210807+xb+bt_transformer-big_2022-05-08.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
flores101.ces-ukr	23.5	0.53099	1012	22810	0.967
flores101.slk-ukr	22.7	0.52645	1012	22810	0.987
newstest2012.ces-rus	21.3	0.49883	3003	64830	0.992
newstest2013.ces-rus	27.9	0.54568	3000	58560	0.965
Tatoeba-test-v2021-08-07.multi-multi	53.5	0.71063	10000	58091	0.984