fra-zle

opusTCv20210807_transformer-big_2022-03-13.zip

• dataset: opusTCv20210807
• model: transformer-big
• source language(s): fra
• target language(s): bel bel_Latn orv_Cyrl rus ukr
• raw source language(s): fra
• 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_transformer-big_2022-03-13.zip
• test set translations: opusTCv20210807_transformer-big_2022-03-13.test.txt
• test set scores: opusTCv20210807_transformer-big_2022-03-13.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newstest2012.fra-rus	22.1	0.50232	3003	64830	0.983
newstest2013.fra-rus	23.9	0.51508	3000	58560	0.980
Tatoeba-test-v2021-08-07.fra-bel	25.0	0.46858	283	1702	0.983
Tatoeba-test-v2021-08-07.fra-multi	42.0	0.63408	10000	58050	1.000
Tatoeba-test-v2021-08-07.fra-orv	0.5	0.16772	37	217	1.000
Tatoeba-test-v2021-08-07.fra-rus	41.5	0.62797	11490	69903	1.000
Tatoeba-test-v2021-08-07.fra-ukr	36.2	0.58410	10035	54232	0.998

opusTCv20210807_transformer-big_2022-03-19.zip

• dataset: opusTCv20210807
• model: transformer-big
• source language(s): fra
• target language(s): bel bel_Latn orv_Cyrl rus ukr
• raw source language(s): fra
• 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_transformer-big_2022-03-19.zip
• test set translations: opusTCv20210807_transformer-big_2022-03-19.test.txt
• test set scores: opusTCv20210807_transformer-big_2022-03-19.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newstest2012.fra-rus	23.2	0.51203	3003	64830	0.976
newstest2013.fra-rus	24.9	0.52311	3000	58560	0.973
Tatoeba-test-v2021-08-07.fra-bel	25.9	0.49382	283	1702	0.985
Tatoeba-test-v2021-08-07.fra-multi	42.7	0.63939	10000	58050	1.000
Tatoeba-test-v2021-08-07.fra-orv	0.5	0.17402	37	217	1.000
Tatoeba-test-v2021-08-07.fra-rus	43.5	0.64456	11490	69903	0.998
Tatoeba-test-v2021-08-07.fra-ukr	38.3	0.60294	10035	54232	0.992

opusTCv20210807_transformer-big_2022-03-23.zip

• dataset: opusTCv20210807
• model: transformer-big
• source language(s): fra
• target language(s): bel bel_Latn orv_Cyrl rus ukr
• raw source language(s): fra
• 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_transformer-big_2022-03-23.zip
• test set translations: opusTCv20210807_transformer-big_2022-03-23.test.txt
• test set scores: opusTCv20210807_transformer-big_2022-03-23.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newstest2012.fra-rus	23.1	0.51258	3003	64830	0.973
newstest2013.fra-rus	24.8	0.52343	3000	58560	0.970
Tatoeba-test-v2021-08-07.fra-bel	30.3	0.51730	283	1702	0.979
Tatoeba-test-v2021-08-07.fra-multi	43.3	0.64353	10000	58050	1.000
Tatoeba-test-v2021-08-07.fra-orv	0.5	0.17487	37	217	1.000
Tatoeba-test-v2021-08-07.fra-rus	44.2	0.64851	11490	69903	0.998
Tatoeba-test-v2021-08-07.fra-ukr	38.7	0.60570	10035	54232	0.992