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): bel rus ukr
target language(s): bul hbs hrv slv srp_Cyrl srp_Latn
valid target language labels: >>bul<< >>hbs<< >>hrv<< >>slv<< >>srp_Cyrl<< >>srp_Latn<<
model: transformer-big
data: opusTCv20210807+bt (source)
tokenization: SentencePiece (spm32k,spm32k)
original model: opusTCv20210807+bt_transformer-big_2022-03-23.zip
more information released models: OPUS-MT zle-zls 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. >>bul<<

Usage

A short example code:

python

from transformers import MarianMTModel, MarianTokenizer

src_text = [
    ">>bul<< Новы каранавірус вельмі заразны.",
    ">>srp_Latn<< Моє ім'я — Саллі."
]

model_name = "pytorch-models/opus-mt-tc-big-zle-zls"
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:
#     Короната е силно заразна.
#     Zovem se Sali.

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-zle-zls")
print(pipe(">>bul<< Новы каранавірус вельмі заразны."))

# expected output: Короната е силно заразна.

Benchmarks

test set translations: opusTCv20210807+bt_transformer-big_2022-03-23.test.txt
test set scores: opusTCv20210807+bt_transformer-big_2022-03-23.eval.txt
benchmark results: benchmark_results.txt
benchmark output: benchmark_translations.zip

langpair	testset	chr-F	BLEU	#sent	#words
rus-bul	tatoeba-test-v2021-08-07	0.71515	53.7	1247	8272
rus-hbs	tatoeba-test-v2021-08-07	0.69192	49.4	2500	14736
rus-slv	tatoeba-test-v2021-08-07	0.38051	21.5	657	3969
rus-srp_Cyrl	tatoeba-test-v2021-08-07	0.66622	46.1	881	5407
rus-srp_Latn	tatoeba-test-v2021-08-07	0.70990	51.7	1483	8552
ukr-bul	tatoeba-test-v2021-08-07	0.77283	61.3	1020	5181
ukr-hbs	tatoeba-test-v2021-08-07	0.69401	52.1	942	5130
ukr-hrv	tatoeba-test-v2021-08-07	0.67202	50.1	389	2302
ukr-srp_Cyrl	tatoeba-test-v2021-08-07	0.70064	54.7	205	1112
ukr-srp_Latn	tatoeba-test-v2021-08-07	0.72405	53.4	348	1716
bel-bul	flores101-devtest	0.49528	16.1	1012	24700
bel-hrv	flores101-devtest	0.46308	12.4	1012	22423
bel-mkd	flores101-devtest	0.48608	13.5	1012	24314
bel-slv	flores101-devtest	0.44452	12.2	1012	23425
bel-srp_Cyrl	flores101-devtest	0.44424	12.6	1012	23456
rus-bul	flores101-devtest	0.58653	28.9	1012	24700
rus-hrv	flores101-devtest	0.53494	23.2	1012	22423
rus-mkd	flores101-devtest	0.55184	24.3	1012	24314
rus-slv	flores101-devtest	0.52201	23.1	1012	23425
rus-srp_Cyrl	flores101-devtest	0.53038	24.1	1012	23456
ukr-bul	flores101-devtest	0.59625	30.8	1012	24700
ukr-hrv	flores101-devtest	0.54530	24.6	1012	22423
ukr-mkd	flores101-devtest	0.56822	26.2	1012	24314
ukr-slv	flores101-devtest	0.53092	24.2	1012	23425
ukr-srp_Cyrl	flores101-devtest	0.54618	26.2	1012	23456

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 00:46:26 EET 2022
port machine: LM0-400-22516.local