gsarti/opus-mt-tc-base-en-ru

Opus Tatoeba English-Russian

This model was obtained by running the script convert_marian_to_pytorch.py. The original models were trained by J�rg Tiedemann using the MarianNMT library. See all available MarianMTModel models on the profile of the Helsinki NLP group.

• dataset: opus+bt
• model: transformer-align
• source language(s): eng
• target language(s): rus
• model: transformer-align
• pre-processing: normalization + SentencePiece (spm32k,spm32k)
• download: opus+bt-2021-04-14.zip
• test set translations: opus+bt-2021-04-14.test.txt
• test set scores: opus+bt-2021-04-14.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newstest2012.eng-rus	31.2	0.588	3003	64830	0.980
newstest2013.eng-rus	23.2	0.518	3000	58560	0.974
newstest2015-enru.eng-rus	29.1	0.581	2818	55915	1.000
newstest2016-enru.eng-rus	27.4	0.565	2998	62018	0.993
newstest2017-enru.eng-rus	30.8	0.592	3001	60255	0.998
newstest2018-enru.eng-rus	27.4	0.572	3000	61920	1.000
newstest2019-enru.eng-rus	27.1	0.540	1997	48153	0.927
Tatoeba-test.eng-rus	45.6	0.652	10000	66872	0.987
tico19-test.eng-rus	27.5	0.556	2100	55837	0.927