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 bal ben bho bpy ckb diq div dty fas gbm glk guj hif hin hne hns jdt kas kmr kok kur lah lrc mag mai mar mzn nep npi ori oss pal pan pes pli prs pus rhg rmy rom san sdh sin skr snd syl tgk tly urd zza
- 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-iir-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-iir-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	0.60240	40.8	279	1335
ben-eng	tatoeba-test-v2021-08-07	0.64471	49.3	2500	13978
fas-deu	tatoeba-test-v2021-08-07	0.58631	34.7	3185	25590
fas-eng	tatoeba-test-v2021-08-07	0.59868	41.8	3762	31480
fas-fra	tatoeba-test-v2021-08-07	0.57181	35.8	376	3377
hin-eng	tatoeba-test-v2021-08-07	0.65417	49.5	5000	33943
kur_Latn-deu	tatoeba-test-v2021-08-07	0.42694	27.0	223	1323
kur_Latn-eng	tatoeba-test-v2021-08-07	0.42721	25.6	290	1708
mar-eng	tatoeba-test-v2021-08-07	0.64493	48.3	10396	67527
pes-eng	tatoeba-test-v2021-08-07	0.59959	41.9	3757	31411
urd-eng	tatoeba-test-v2021-08-07	0.53679	35.4	1663	12029
ben-deu	flores101-devtest	0.46873	16.4	1012	25094
ben-eng	flores101-devtest	0.57508	30.0	1012	24721
ben-spa	flores101-devtest	0.44010	15.1	1012	29199
ckb-deu	flores101-devtest	0.41546	13.0	1012	25094
ckb-por	flores101-devtest	0.44178	17.6	1012	26519
fas-por	flores101-devtest	0.54077	26.1	1012	26519
guj-deu	flores101-devtest	0.45906	16.5	1012	25094
guj-spa	flores101-devtest	0.43928	15.2	1012	29199
hin-eng	flores101-devtest	0.62807	36.6	1012	24721
hin-por	flores101-devtest	0.52825	25.1	1012	26519
mar-deu	flores101-devtest	0.44767	14.8	1012	25094
npi-deu	flores101-devtest	0.46178	15.9	1012	25094
pan-fra	flores101-devtest	0.50909	23.4	1012	28343
pan-por	flores101-devtest	0.50634	23.0	1012	26519
pus-deu	flores101-devtest	0.42645	13.5	1012	25094
pus-fra	flores101-devtest	0.45719	18.0	1012	28343
urd-deu	flores101-devtest	0.46102	16.5	1012	25094
urd-eng	flores101-devtest	0.56356	28.4	1012	24721
asm-eng	flores200-devtest	0.48589	21.5	1012	24721
awa-deu	flores200-devtest	0.47071	16.0	1012	25094
awa-eng	flores200-devtest	0.53069	26.6	1012	24721
awa-fra	flores200-devtest	0.49700	21.1	1012	28343
awa-por	flores200-devtest	0.49950	21.8	1012	26519
awa-spa	flores200-devtest	0.43831	15.4	1012	29199
ben-deu	flores200-devtest	0.47434	17.0	1012	25094
ben-eng	flores200-devtest	0.58408	31.4	1012	24721
ben-fra	flores200-devtest	0.50930	23.2	1012	28343
ben-por	flores200-devtest	0.50661	22.4	1012	26519
ben-spa	flores200-devtest	0.44485	15.7	1012	29199
bho-deu	flores200-devtest	0.42463	12.8	1012	25094
bho-eng	flores200-devtest	0.50545	22.6	1012	24721
bho-fra	flores200-devtest	0.45264	17.4	1012	28343
bho-por	flores200-devtest	0.44737	17.0	1012	26519
bho-spa	flores200-devtest	0.40585	13.0	1012	29199
ckb-deu	flores200-devtest	0.42110	13.6	1012	25094
ckb-eng	flores200-devtest	0.50543	24.7	1012	24721
ckb-fra	flores200-devtest	0.45847	19.1	1012	28343
ckb-por	flores200-devtest	0.44567	17.8	1012	26519
guj-deu	flores200-devtest	0.46758	17.3	1012	25094
guj-eng	flores200-devtest	0.61139	34.4	1012	24721
guj-fra	flores200-devtest	0.50349	22.5	1012	28343
guj-por	flores200-devtest	0.49828	22.4	1012	26519
guj-spa	flores200-devtest	0.44472	15.5	1012	29199
hin-deu	flores200-devtest	0.50772	20.8	1012	25094
hin-eng	flores200-devtest	0.63234	37.3	1012	24721
hin-fra	flores200-devtest	0.53933	26.5	1012	28343
hin-por	flores200-devtest	0.53523	26.1	1012	26519
hin-spa	flores200-devtest	0.46183	17.4	1012	29199
hne-deu	flores200-devtest	0.49946	19.0	1012	25094
hne-eng	flores200-devtest	0.63640	38.1	1012	24721
hne-fra	flores200-devtest	0.53419	25.7	1012	28343
hne-por	flores200-devtest	0.53735	25.9	1012	26519
hne-spa	flores200-devtest	0.45610	16.9	1012	29199
mag-deu	flores200-devtest	0.50681	20.0	1012	25094
mag-eng	flores200-devtest	0.63966	38.0	1012	24721
mag-fra	flores200-devtest	0.53810	25.9	1012	28343
mag-por	flores200-devtest	0.54065	26.6	1012	26519
mag-spa	flores200-devtest	0.46131	17.1	1012	29199
mai-deu	flores200-devtest	0.47686	16.8	1012	25094
mai-eng	flores200-devtest	0.57552	30.2	1012	24721
mai-fra	flores200-devtest	0.50909	22.4	1012	28343
mai-por	flores200-devtest	0.51249	22.9	1012	26519
mai-spa	flores200-devtest	0.44694	15.9	1012	29199
mar-deu	flores200-devtest	0.45295	14.8	1012	25094
mar-eng	flores200-devtest	0.58203	31.0	1012	24721
mar-fra	flores200-devtest	0.48254	20.4	1012	28343
mar-por	flores200-devtest	0.48368	20.4	1012	26519
mar-spa	flores200-devtest	0.42799	14.7	1012	29199
npi-deu	flores200-devtest	0.47267	17.2	1012	25094
npi-eng	flores200-devtest	0.59559	32.5	1012	24721
npi-fra	flores200-devtest	0.50869	22.5	1012	28343
npi-por	flores200-devtest	0.50900	22.5	1012	26519
npi-spa	flores200-devtest	0.44304	15.6	1012	29199
pan-deu	flores200-devtest	0.48342	18.6	1012	25094
pan-eng	flores200-devtest	0.60328	33.4	1012	24721
pan-fra	flores200-devtest	0.51953	24.4	1012	28343
pan-por	flores200-devtest	0.51428	23.9	1012	26519
pan-spa	flores200-devtest	0.44615	16.3	1012	29199
pes-deu	flores200-devtest	0.51124	21.0	1012	25094
pes-eng	flores200-devtest	0.60538	33.7	1012	24721
pes-fra	flores200-devtest	0.55157	27.8	1012	28343
pes-por	flores200-devtest	0.54372	26.6	1012	26519
pes-spa	flores200-devtest	0.47561	18.8	1012	29199
prs-deu	flores200-devtest	0.50273	20.7	1012	25094
prs-eng	flores200-devtest	0.60144	34.5	1012	24721
prs-fra	flores200-devtest	0.54241	27.0	1012	28343
prs-por	flores200-devtest	0.53562	26.6	1012	26519
prs-spa	flores200-devtest	0.46497	18.1	1012	29199
sin-deu	flores200-devtest	0.45041	14.7	1012	25094
sin-eng	flores200-devtest	0.54060	26.3	1012	24721
sin-fra	flores200-devtest	0.48163	19.9	1012	28343
sin-por	flores200-devtest	0.47780	19.6	1012	26519
sin-spa	flores200-devtest	0.42546	14.2	1012	29199
tgk-deu	flores200-devtest	0.45203	15.6	1012	25094
tgk-eng	flores200-devtest	0.53740	25.3	1012	24721
tgk-fra	flores200-devtest	0.50153	22.1	1012	28343
tgk-por	flores200-devtest	0.49378	21.9	1012	26519
tgk-spa	flores200-devtest	0.44099	15.9	1012	29199
urd-deu	flores200-devtest	0.46894	17.2	1012	25094
urd-eng	flores200-devtest	0.56967	29.3	1012	24721
urd-fra	flores200-devtest	0.50616	22.6	1012	28343
urd-por	flores200-devtest	0.49398	21.7	1012	26519
urd-spa	flores200-devtest	0.43800	15.4	1012	29199
hin-eng	newstest2014	0.59024	30.3	2507	55571
guj-eng	newstest2019	0.53977	27.2	1016	17757
ben-deu	ntrex128	0.45551	15.0	1997	48761
ben-eng	ntrex128	0.56878	29.0	1997	47673
ben-fra	ntrex128	0.47077	18.6	1997	53481
ben-por	ntrex128	0.46049	17.1	1997	51631
ben-spa	ntrex128	0.48833	21.3	1997	54107
fas-deu	ntrex128	0.46991	16.1	1997	48761
fas-eng	ntrex128	0.55119	25.9	1997	47673
fas-fra	ntrex128	0.49626	21.2	1997	53481
fas-por	ntrex128	0.47499	18.6	1997	51631
fas-spa	ntrex128	0.50178	22.8	1997	54107
guj-deu	ntrex128	0.43998	14.3	1997	48761
guj-eng	ntrex128	0.58481	31.0	1997	47673
guj-fra	ntrex128	0.45468	17.3	1997	53481
guj-por	ntrex128	0.44223	15.8	1997	51631
guj-spa	ntrex128	0.47798	20.7	1997	54107
hin-deu	ntrex128	0.46580	15.0	1997	48761
hin-eng	ntrex128	0.59832	31.6	1997	47673
hin-fra	ntrex128	0.48328	19.5	1997	53481
hin-por	ntrex128	0.46833	17.8	1997	51631
hin-spa	ntrex128	0.49517	21.9	1997	54107
mar-deu	ntrex128	0.43713	13.5	1997	48761
mar-eng	ntrex128	0.55132	27.4	1997	47673
mar-fra	ntrex128	0.44797	16.9	1997	53481
mar-por	ntrex128	0.44342	16.1	1997	51631
mar-spa	ntrex128	0.46950	19.7	1997	54107
nep-deu	ntrex128	0.43568	13.5	1997	48761
nep-eng	ntrex128	0.55954	28.8	1997	47673
nep-fra	ntrex128	0.45083	16.9	1997	53481
nep-por	ntrex128	0.44458	16.0	1997	51631
nep-spa	ntrex128	0.46832	19.4	1997	54107
pan-deu	ntrex128	0.44327	14.6	1997	48761
pan-eng	ntrex128	0.57665	30.5	1997	47673
pan-fra	ntrex128	0.45815	17.7	1997	53481
pan-por	ntrex128	0.44608	16.3	1997	51631
pan-spa	ntrex128	0.47289	20.0	1997	54107
prs-deu	ntrex128	0.45067	14.6	1997	48761
prs-eng	ntrex128	0.54767	26.6	1997	47673
prs-fra	ntrex128	0.47453	19.3	1997	53481
prs-por	ntrex128	0.45843	17.1	1997	51631
prs-spa	ntrex128	0.48317	20.9	1997	54107
pus-eng	ntrex128	0.44698	17.6	1997	47673
pus-spa	ntrex128	0.41132	14.6	1997	54107
sin-deu	ntrex128	0.42541	12.5	1997	48761
sin-eng	ntrex128	0.51853	23.5	1997	47673
sin-fra	ntrex128	0.44099	15.9	1997	53481
sin-por	ntrex128	0.43010	14.4	1997	51631
sin-spa	ntrex128	0.46225	18.4	1997	54107
tgk_Cyrl-deu	ntrex128	0.40368	11.4	1997	48761
tgk_Cyrl-eng	ntrex128	0.47132	18.2	1997	47673
tgk_Cyrl-fra	ntrex128	0.43311	15.8	1997	53481
tgk_Cyrl-por	ntrex128	0.42095	13.8	1997	51631
tgk_Cyrl-spa	ntrex128	0.44279	17.3	1997	54107
urd-deu	ntrex128	0.45708	15.5	1997	48761
urd-eng	ntrex128	0.56560	28.5	1997	47673
urd-fra	ntrex128	0.47536	19.0	1997	53481
urd-por	ntrex128	0.45911	16.7	1997	51631
urd-spa	ntrex128	0.48986	21.6	1997	54107
ben-eng	tico19-test	0.64578	38.7	2100	56824
ben-fra	tico19-test	0.50165	22.8	2100	64661
ben-por	tico19-test	0.55662	27.7	2100	62729
ben-spa	tico19-test	0.56795	29.6	2100	66563
ckb-eng	tico19-test	0.51623	27.4	2100	56315
ckb-fra	tico19-test	0.42405	17.1	2100	64661
ckb-por	tico19-test	0.45405	19.0	2100	62729
ckb-spa	tico19-test	0.46976	21.7	2100	66563
fas-eng	tico19-test	0.62079	34.2	2100	56315
fas-fra	tico19-test	0.52041	24.4	2100	64661
fas-por	tico19-test	0.56780	29.2	2100	62729
fas-spa	tico19-test	0.58248	32.3	2100	66563
hin-eng	tico19-test	0.70535	46.8	2100	56323
hin-fra	tico19-test	0.53833	26.6	2100	64661
hin-por	tico19-test	0.60246	33.2	2100	62729
hin-spa	tico19-test	0.61504	35.7	2100	66563
mar-eng	tico19-test	0.59247	31.4	2100	56315
mar-fra	tico19-test	0.46895	19.3	2100	64661
mar-por	tico19-test	0.51945	23.8	2100	62729
mar-spa	tico19-test	0.52914	26.2	2100	66563
nep-eng	tico19-test	0.65865	40.1	2100	56824
nep-fra	tico19-test	0.50473	23.2	2100	64661
nep-por	tico19-test	0.56185	28.0	2100	62729
nep-spa	tico19-test	0.57270	30.2	2100	66563
prs-eng	tico19-test	0.59536	32.1	2100	56824
prs-fra	tico19-test	0.50044	23.1	2100	64661
prs-por	tico19-test	0.54448	27.3	2100	62729
prs-spa	tico19-test	0.56311	30.2	2100	66563
pus-eng	tico19-test	0.56711	31.4	2100	56315
pus-fra	tico19-test	0.45951	19.4	2100	64661
pus-por	tico19-test	0.50225	23.7	2100	62729
pus-spa	tico19-test	0.51246	25.4	2100	66563
urd-eng	tico19-test	0.57786	30.8	2100	56315
urd-fra	tico19-test	0.46807	20.1	2100	64661
urd-por	tico19-test	0.51567	24.1	2100	62729
urd-spa	tico19-test	0.52820	26.4	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 11:32:44 EEST 2024
port machine: LM0-400-22516.local

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

opus-mt-tc-bible-big-iir-deu_eng_fra_por_spa

Table of Contents

Model Details