This model is part of the OPUS-MT project, an effort to make neural machine translation models widely available and accessible for many languages in the world. All models are originally trained using the amazing framework of Marian NMT, an efficient NMT implementation written in pure C++. The models have been converted to pyTorch using the transformers library by huggingface. Training data is taken from OPUS and training pipelines use the procedures of OPUS-MT-train.

Publications: OPUS-MT – Building open translation services for the World and The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT (Please, cite if you use this model.)

@inproceedings{tiedemann-thottingal-2020-opus,
    title = "{OPUS}-{MT} {--} Building open translation services for the World",
    author = {Tiedemann, J{\"o}rg  and Thottingal, Santhosh},
    booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation",
    month = nov,
    year = "2020",
    address = "Lisboa, Portugal",
    publisher = "European Association for Machine Translation",
    url = "https://aclanthology.org/2020.eamt-1.61",
    pages = "479--480",
}

@inproceedings{tiedemann-2020-tatoeba,
    title = "The Tatoeba Translation Challenge {--} Realistic Data Sets for Low Resource and Multilingual {MT}",
    author = {Tiedemann, J{\"o}rg},
    booktitle = "Proceedings of the Fifth Conference on Machine Translation",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2020.wmt-1.139",
    pages = "1174--1182",
}

Model info

Release: 2022-03-23
source language(s): deu
target language(s): bel rus ukr
valid target language labels: >>bel<< >>rus<< >>ukr<<
model: transformer-big
data: opusTCv20210807 (source)
tokenization: SentencePiece (spm32k,spm32k)
original model: opusTCv20210807_transformer-big_2022-03-23.zip
more information released models: OPUS-MT deu-zle README
more information about the model: MarianMT

This is a multilingual translation model with multiple target languages. A sentence initial language token is required in the form of >>id<< (id = valid target language ID), e.g. >>bel<<

Usage

A short example code:

python

from transformers import MarianMTModel, MarianTokenizer

src_text = [
    ">>ukr<< Der Soldat hat mir Wasser gegeben.",
    ">>ukr<< Ich will hier nicht essen."
]

model_name = "pytorch-models/opus-mt-tc-big-de-zle"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)
translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True))

for t in translated:
    print( tokenizer.decode(t, skip_special_tokens=True) )

# expected output:
#     Солдат дав мені воду.
#     Я не хочу тут їсти.

You can also use OPUS-MT models with the transformers pipelines, for example:

python

from transformers import pipeline
pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-de-zle")
print(pipe(">>ukr<< Der Soldat hat mir Wasser gegeben."))

# expected output: Солдат дав мені воду.

Benchmarks

test set translations: opusTCv20210807_transformer-big_2022-03-23.test.txt
test set scores: opusTCv20210807_transformer-big_2022-03-23.eval.txt
benchmark results: benchmark_results.txt
benchmark output: benchmark_translations.zip

langpair	testset	chr-F	BLEU	#sent	#words
deu-bel	tatoeba-test-v2021-08-07	0.53128	29.5	551	3601
deu-rus	tatoeba-test-v2021-08-07	0.67143	46.1	12800	87296
deu-ukr	tatoeba-test-v2021-08-07	0.62737	40.7	10319	56287
deu-rus	flores101-devtest	0.54152	26.3	1012	23295
deu-ukr	flores101-devtest	0.53286	24.2	1012	22810
deu-rus	newstest2012	0.49409	20.8	3003	64790
deu-rus	newstest2013	0.52631	24.9	3000	58560

Acknowledgements

The work is supported by the European Language Grid as pilot project 2866, by the FoTran project, funded by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 771113), and the MeMAD project, funded by the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement No 780069. We are also grateful for the generous computational resources and IT infrastructure provided by CSC -- IT Center for Science, Finland.

Model conversion info

transformers version: 4.16.2
OPUS-MT git hash: 1bdabf7
port time: Thu Mar 24 01:29:09 EET 2022
port machine: LM0-400-22516.local