Edit model card
YAML Metadata Error: "language[6]" must only contain lowercase characters
YAML Metadata Error: "language[6]" with value "sr_Cyrl" is not valid. It must be an ISO 639-1, 639-2 or 639-3 code (two/three letters), or a special value like "code", "multilingual". If you want to use BCP-47 identifiers, you can specify them in language_bcp47.
YAML Metadata Error: "language[7]" must only contain lowercase characters
YAML Metadata Error: "language[7]" with value "sr_Latn" is not valid. It must be an ISO 639-1, 639-2 or 639-3 code (two/three letters), or a special value like "code", "multilingual". If you want to use BCP-47 identifiers, you can specify them in language_bcp47.

opus-mt-tc-big-zls-zle

Neural machine translation model for translating from South Slavic languages (zls) to East Slavic languages (zle).

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.

@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): bul hbs hrv slv srp_Cyrl srp_Latn
  • target language(s): bel rus ukr
  • valid target language labels: >>bel<< >>rus<< >>ukr<<
  • 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 zls-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:

from transformers import MarianMTModel, MarianTokenizer

src_text = [
    ">>rus<< Gdje je brigadir?",
    ">>ukr<< Zovem se Seli."
]

model_name = "pytorch-models/opus-mt-tc-big-zls-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:
#     Где бригадир?
#     Мене звати Саллі.

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

from transformers import pipeline
pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-zls-zle")
print(pipe(">>rus<< Gdje je brigadir?"))

# expected output: Где бригадир?

Benchmarks

langpair testset chr-F BLEU #sent #words
bul-rus tatoeba-test-v2021-08-07 0.71467 52.6 1247 7870
bul-ukr tatoeba-test-v2021-08-07 0.71757 53.3 1020 4932
hbs-rus tatoeba-test-v2021-08-07 0.74593 58.5 2500 14213
hbs-ukr tatoeba-test-v2021-08-07 0.70244 52.3 942 4961
hrv-ukr tatoeba-test-v2021-08-07 0.68931 50.0 389 2232
slv-rus tatoeba-test-v2021-08-07 0.42255 27.3 657 4056
srp_Cyrl-rus tatoeba-test-v2021-08-07 0.74112 56.2 881 5117
srp_Cyrl-ukr tatoeba-test-v2021-08-07 0.68915 51.8 205 1061
srp_Latn-rus tatoeba-test-v2021-08-07 0.75340 60.1 1483 8311
srp_Latn-ukr tatoeba-test-v2021-08-07 0.73106 55.8 348 1668
bul-rus flores101-devtest 0.54226 24.6 1012 23295
bul-ukr flores101-devtest 0.53382 22.9 1012 22810
hrv-rus flores101-devtest 0.51726 23.5 1012 23295
hrv-ukr flores101-devtest 0.51011 21.9 1012 22810
mkd-bel flores101-devtest 0.40885 10.7 1012 24829
mkd-rus flores101-devtest 0.52509 24.3 1012 23295
mkd-ukr flores101-devtest 0.52021 22.5 1012 22810
slv-rus flores101-devtest 0.50349 22.0 1012 23295
slv-ukr flores101-devtest 0.49156 20.2 1012 22810
srp_Cyrl-rus flores101-devtest 0.53656 25.7 1012 23295
srp_Cyrl-ukr flores101-devtest 0.53623 24.4 1012 22810

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 04:08:51 EET 2022
  • port machine: LM0-400-22516.local
Downloads last month
1
Hosted inference API
This model can be loaded on the Inference API on-demand.

Evaluation results