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. Model Description:

Developed by: Language Technology Research Group at the University of Helsinki
Model Type: Translation (transformer-big)
Release: 2024-05-30
License: Apache-2.0
Language(s):
- Source Language(s): anp asm awa ben bho bpy div dty gbm guj hif hin hne hns kas kok lah mag mai mar nep npi ori pan pli rhg rmy rom san sin skr snd syl urd
- Target Language(s): deu eng fra por spa
- Valid Target Language Labels: >>deu<< >>eng<< >>fra<< >>por<< >>spa<< >>xxx<<
Original Model: opusTCv20230926max50+bt+jhubc_transformer-big_2024-05-30.zip
Resources for more information:

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. >>deu<<

Uses

This model can be used for translation and text-to-text generation.

Risks, Limitations and Biases

CONTENT WARNING: Readers should be aware that the model is trained on various public data sets that may contain content that is disturbing, offensive, and can propagate historical and current stereotypes.

Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)).

How to Get Started With the Model

A short example code:

python

from transformers import MarianMTModel, MarianTokenizer

src_text = [
    ">>deu<< Replace this with text in an accepted source language.",
    ">>spa<< This is the second sentence."
]

model_name = "pytorch-models/opus-mt-tc-bible-big-inc-deu_eng_fra_por_spa"
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) )

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-bible-big-inc-deu_eng_fra_por_spa")
print(pipe(">>deu<< Replace this with text in an accepted source language."))

Training

Data: opusTCv20230926max50+bt+jhubc (source)
Pre-processing: SentencePiece (spm32k,spm32k)
Model Type: transformer-big
Original MarianNMT Model: opusTCv20230926max50+bt+jhubc_transformer-big_2024-05-30.zip
Training Scripts: GitHub Repo

Evaluation

Model scores at the OPUS-MT dashboard
test set translations: opusTCv20230926max50+bt+jhubc_transformer-big_2024-05-29.test.txt
test set scores: opusTCv20230926max50+bt+jhubc_transformer-big_2024-05-29.eval.txt
benchmark results: benchmark_results.txt
benchmark output: benchmark_translations.zip

langpair	testset	chr-F	BLEU	#sent	#words
awa-eng	tatoeba-test-v2021-08-07	0.60390	40.8	279	1335
ben-eng	tatoeba-test-v2021-08-07	0.64078	49.4	2500	13978
hin-eng	tatoeba-test-v2021-08-07	0.64929	49.1	5000	33943
mar-eng	tatoeba-test-v2021-08-07	0.64074	48.0	10396	67527
urd-eng	tatoeba-test-v2021-08-07	0.52963	35.0	1663	12029
ben-eng	flores101-devtest	0.57906	30.4	1012	24721
ben-fra	flores101-devtest	0.50109	21.9	1012	28343
guj-spa	flores101-devtest	0.44065	15.2	1012	29199
mar-deu	flores101-devtest	0.44067	13.8	1012	25094
mar-por	flores101-devtest	0.46685	18.6	1012	26519
mar-spa	flores101-devtest	0.41662	14.0	1012	29199
pan-eng	flores101-devtest	0.59922	33.0	1012	24721
pan-por	flores101-devtest	0.49373	21.9	1012	26519
pan-spa	flores101-devtest	0.43910	15.4	1012	29199
asm-eng	flores200-devtest	0.48584	21.9	1012	24721
awa-deu	flores200-devtest	0.47173	16.5	1012	25094
awa-eng	flores200-devtest	0.50582	24.5	1012	24721
awa-fra	flores200-devtest	0.49682	21.4	1012	28343
awa-por	flores200-devtest	0.49663	21.5	1012	26519
awa-spa	flores200-devtest	0.43740	15.1	1012	29199
ben-deu	flores200-devtest	0.47330	16.6	1012	25094
ben-eng	flores200-devtest	0.58077	30.5	1012	24721
ben-fra	flores200-devtest	0.50884	22.6	1012	28343
ben-por	flores200-devtest	0.50054	21.4	1012	26519
ben-spa	flores200-devtest	0.44159	15.2	1012	29199
bho-deu	flores200-devtest	0.42660	12.6	1012	25094
bho-eng	flores200-devtest	0.50609	22.7	1012	24721
bho-fra	flores200-devtest	0.44889	16.8	1012	28343
bho-por	flores200-devtest	0.44582	16.9	1012	26519
bho-spa	flores200-devtest	0.40581	13.1	1012	29199
guj-deu	flores200-devtest	0.46665	16.8	1012	25094
guj-eng	flores200-devtest	0.61383	34.3	1012	24721
guj-fra	flores200-devtest	0.50410	22.3	1012	28343
guj-por	flores200-devtest	0.49257	21.3	1012	26519
guj-spa	flores200-devtest	0.44565	15.6	1012	29199
hin-deu	flores200-devtest	0.50226	20.4	1012	25094
hin-eng	flores200-devtest	0.63336	37.3	1012	24721
hin-fra	flores200-devtest	0.53701	25.9	1012	28343
hin-por	flores200-devtest	0.53448	25.5	1012	26519
hin-spa	flores200-devtest	0.46171	17.2	1012	29199
hne-deu	flores200-devtest	0.49698	19.0	1012	25094
hne-eng	flores200-devtest	0.63936	38.5	1012	24721
hne-fra	flores200-devtest	0.52835	25.3	1012	28343
hne-por	flores200-devtest	0.52788	25.0	1012	26519
hne-spa	flores200-devtest	0.45443	16.7	1012	29199
mag-deu	flores200-devtest	0.50359	19.7	1012	25094
mag-eng	flores200-devtest	0.63906	38.0	1012	24721
mag-fra	flores200-devtest	0.53616	25.8	1012	28343
mag-por	flores200-devtest	0.53537	25.9	1012	26519
mag-spa	flores200-devtest	0.45822	16.9	1012	29199
mai-deu	flores200-devtest	0.46791	16.2	1012	25094
mai-eng	flores200-devtest	0.57461	30.4	1012	24721
mai-fra	flores200-devtest	0.50585	22.1	1012	28343
mai-por	flores200-devtest	0.50490	22.0	1012	26519
mai-spa	flores200-devtest	0.44366	15.3	1012	29199
mar-deu	flores200-devtest	0.44725	14.5	1012	25094
mar-eng	flores200-devtest	0.58500	31.4	1012	24721
mar-fra	flores200-devtest	0.47027	19.5	1012	28343
mar-por	flores200-devtest	0.47216	19.3	1012	26519
mar-spa	flores200-devtest	0.42178	14.2	1012	29199
npi-deu	flores200-devtest	0.46631	16.4	1012	25094
npi-eng	flores200-devtest	0.59776	32.3	1012	24721
npi-fra	flores200-devtest	0.50548	22.5	1012	28343
npi-por	flores200-devtest	0.50202	21.7	1012	26519
npi-spa	flores200-devtest	0.43804	15.3	1012	29199
pan-deu	flores200-devtest	0.48421	18.7	1012	25094
pan-eng	flores200-devtest	0.60676	33.8	1012	24721
pan-fra	flores200-devtest	0.51368	23.5	1012	28343
pan-por	flores200-devtest	0.50586	22.7	1012	26519
pan-spa	flores200-devtest	0.44653	16.5	1012	29199
sin-deu	flores200-devtest	0.44676	14.2	1012	25094
sin-eng	flores200-devtest	0.54777	26.8	1012	24721
sin-fra	flores200-devtest	0.47283	19.0	1012	28343
sin-por	flores200-devtest	0.46935	18.4	1012	26519
sin-spa	flores200-devtest	0.42143	13.7	1012	29199
urd-deu	flores200-devtest	0.46542	17.1	1012	25094
urd-eng	flores200-devtest	0.56935	29.3	1012	24721
urd-fra	flores200-devtest	0.50276	22.3	1012	28343
urd-por	flores200-devtest	0.48010	20.3	1012	26519
urd-spa	flores200-devtest	0.43032	14.7	1012	29199
hin-eng	newstest2014	0.59329	30.3	2507	55571
guj-eng	newstest2019	0.53383	26.9	1016	17757
ben-deu	ntrex128	0.45180	14.6	1997	48761
ben-eng	ntrex128	0.57247	29.5	1997	47673
ben-fra	ntrex128	0.46475	18.0	1997	53481
ben-por	ntrex128	0.45486	16.8	1997	51631
ben-spa	ntrex128	0.48738	21.1	1997	54107
guj-deu	ntrex128	0.43539	13.9	1997	48761
guj-eng	ntrex128	0.58894	31.6	1997	47673
guj-fra	ntrex128	0.45075	16.9	1997	53481
guj-por	ntrex128	0.43567	15.2	1997	51631
guj-spa	ntrex128	0.47525	20.2	1997	54107
hin-deu	ntrex128	0.46336	15.0	1997	48761
hin-eng	ntrex128	0.59842	31.5	1997	47673
hin-fra	ntrex128	0.48208	19.2	1997	53481
hin-por	ntrex128	0.46509	17.6	1997	51631
hin-spa	ntrex128	0.49436	21.8	1997	54107
mar-deu	ntrex128	0.43119	12.8	1997	48761
mar-eng	ntrex128	0.55151	27.3	1997	47673
mar-fra	ntrex128	0.43957	16.2	1997	53481
mar-por	ntrex128	0.43555	15.4	1997	51631
mar-spa	ntrex128	0.46271	19.1	1997	54107
nep-deu	ntrex128	0.42940	13.0	1997	48761
nep-eng	ntrex128	0.56277	29.1	1997	47673
nep-fra	ntrex128	0.44663	16.5	1997	53481
nep-por	ntrex128	0.43686	15.4	1997	51631
nep-spa	ntrex128	0.46553	19.3	1997	54107
pan-deu	ntrex128	0.44036	14.1	1997	48761
pan-eng	ntrex128	0.58427	31.6	1997	47673
pan-fra	ntrex128	0.45593	17.3	1997	53481
pan-por	ntrex128	0.44264	15.9	1997	51631
pan-spa	ntrex128	0.47199	20.0	1997	54107
sin-deu	ntrex128	0.42280	12.4	1997	48761
sin-eng	ntrex128	0.52576	24.6	1997	47673
sin-fra	ntrex128	0.43594	15.6	1997	53481
sin-por	ntrex128	0.42751	14.4	1997	51631
sin-spa	ntrex128	0.45890	18.3	1997	54107
urd-deu	ntrex128	0.45737	15.6	1997	48761
urd-eng	ntrex128	0.56781	28.6	1997	47673
urd-fra	ntrex128	0.47298	18.9	1997	53481
urd-por	ntrex128	0.45273	16.2	1997	51631
urd-spa	ntrex128	0.48644	21.0	1997	54107
ben-eng	tico19-test	0.64568	38.2	2100	56824
ben-fra	tico19-test	0.49799	22.0	2100	64661
ben-por	tico19-test	0.55115	27.2	2100	62729
ben-spa	tico19-test	0.56847	29.9	2100	66563
hin-eng	tico19-test	0.70694	46.6	2100	56323
hin-fra	tico19-test	0.53932	26.7	2100	64661
hin-por	tico19-test	0.60581	33.4	2100	62729
hin-spa	tico19-test	0.61585	35.7	2100	66563
mar-eng	tico19-test	0.59329	31.8	2100	56315
mar-fra	tico19-test	0.46574	19.3	2100	64661
mar-por	tico19-test	0.51463	23.6	2100	62729
mar-spa	tico19-test	0.52551	25.7	2100	66563
nep-eng	tico19-test	0.66283	40.7	2100	56824
nep-fra	tico19-test	0.50397	22.8	2100	64661
nep-por	tico19-test	0.55951	28.1	2100	62729
nep-spa	tico19-test	0.57272	30.3	2100	66563
urd-eng	tico19-test	0.57473	30.5	2100	56315
urd-fra	tico19-test	0.46725	19.6	2100	64661
urd-por	tico19-test	0.50913	23.5	2100	62729
urd-spa	tico19-test	0.52387	25.8	2100	66563

Citation Information

Publications: Democratizing neural machine translation with OPUS-MT and 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.)

bibtex

@article{tiedemann2023democratizing,
  title={Democratizing neural machine translation with {OPUS-MT}},
  author={Tiedemann, J{\"o}rg and Aulamo, Mikko and Bakshandaeva, Daria and Boggia, Michele and Gr{\"o}nroos, Stig-Arne and Nieminen, Tommi and Raganato, Alessandro and Scherrer, Yves and Vazquez, Raul and Virpioja, Sami},
  journal={Language Resources and Evaluation},
  number={58},
  pages={713--755},
  year={2023},
  publisher={Springer Nature},
  issn={1574-0218},
  doi={10.1007/s10579-023-09704-w}
}

@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",
}

Acknowledgements

The work is supported by the HPLT project, funded by the European Union’s Horizon Europe research and innovation programme under grant agreement No 101070350. We are also grateful for the generous computational resources and IT infrastructure provided by CSC -- IT Center for Science, Finland, and the EuroHPC supercomputer LUMI.

Model conversion info

transformers version: 4.45.1
OPUS-MT git hash: 0882077
port time: Tue Oct 8 11:39:25 EEST 2024
port machine: LM0-400-22516.local

Helsinki-NLP/opus-mt-tc-bible-big-inc-deu_eng_fra_por_spa

opus-mt-tc-bible-big-inc-deu_eng_fra_por_spa

Table of Contents

Model Details