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: 2022-08-09
License: CC-BY-4.0
Language(s):
- Source Language(s): dan isl nno nob nor swe
- Target Language(s): cat fra glg ita lat por ron spa
- Language Pair(s): dan-cat dan-fra dan-glg dan-ita dan-por dan-ron dan-spa isl-cat isl-fra isl-ita isl-por isl-ron isl-spa nob-cat nob-fra nob-glg nob-ita nob-por nob-ron nob-spa swe-cat swe-fra swe-glg swe-ita swe-por swe-ron swe-spa
- Valid Target Language Labels: >>acf<< >>aoa<< >>arg<< >>ast<< >>cat<< >>cbk<< >>ccd<< >>cks<< >>cos<< >>cri<< >>crs<< >>dlm<< >>drc<< >>egl<< >>ext<< >>fab<< >>fax<< >>fra<< >>frc<< >>frm<< >>fro<< >>frp<< >>fur<< >>gcf<< >>gcr<< >>glg<< >>hat<< >>idb<< >>ist<< >>ita<< >>itk<< >>kea<< >>kmv<< >>lad<< >>lad_Latn<< >>lat<< >>lat_Latn<< >>lij<< >>lld<< >>lmo<< >>lou<< >>mcm<< >>mfe<< >>mol<< >>mwl<< >>mxi<< >>mzs<< >>nap<< >>nrf<< >>oci<< >>osc<< >>osp<< >>osp_Latn<< >>pap<< >>pcd<< >>pln<< >>pms<< >>pob<< >>por<< >>pov<< >>pre<< >>pro<< >>qbb<< >>qhr<< >>rcf<< >>rgn<< >>roh<< >>ron<< >>ruo<< >>rup<< >>ruq<< >>scf<< >>scn<< >>sdc<< >>sdn<< >>spa<< >>spq<< >>spx<< >>src<< >>srd<< >>sro<< >>tmg<< >>tvy<< >>vec<< >>vkp<< >>wln<< >>xfa<< >>xum<<
Original Model: opusTCv20210807_transformer-big_2022-08-09.zip
Resources for more information:
- OPUS-MT-train GitHub Repo
- More information about released models for this language pair: OPUS-MT gmq-itc README
- More information about MarianNMT models in the transformers library
- [Tatoeba Translation Challenge](https://github.com/Helsinki-NLP/Tatoeba-Challenge/

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. >>fra<<

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 = [
    ">>spa<< Jag är inte religiös.",
    ">>por<< Livet er for kort til å lære seg tysk."
]

model_name = "pytorch-models/opus-mt-tc-big-gmq-itc"
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:
#     No soy religioso.
#     A vida é muito curta para aprender alemão.

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-gmq-itc")
print(pipe(">>spa<< Jag är inte religiös."))

# expected output: No soy religioso.

Training

Data: opusTCv20210807 (source)
Pre-processing: SentencePiece (spm32k,spm32k)
Model Type: transformer-big
Original MarianNMT Model: opusTCv20210807_transformer-big_2022-08-09.zip
Training Scripts: GitHub Repo

Evaluation

test set translations: opusTCv20210807_transformer-big_2022-08-09.test.txt
test set scores: opusTCv20210807_transformer-big_2022-08-09.eval.txt
benchmark results: benchmark_results.txt
benchmark output: benchmark_translations.zip

langpair	testset	chr-F	BLEU	#sent	#words
dan-fra	tatoeba-test-v2021-08-07	0.76671	63.8	1731	11882
dan-ita	tatoeba-test-v2021-08-07	0.74658	56.2	284	2226
dan-por	tatoeba-test-v2021-08-07	0.74944	57.8	873	5360
dan-spa	tatoeba-test-v2021-08-07	0.72328	54.8	5000	35528
isl-ita	tatoeba-test-v2021-08-07	0.69354	51.0	236	1450
isl-spa	tatoeba-test-v2021-08-07	0.66008	49.2	238	1229
nob-fra	tatoeba-test-v2021-08-07	0.70854	54.4	323	2269
nob-spa	tatoeba-test-v2021-08-07	0.73672	55.9	885	6866
swe-fra	tatoeba-test-v2021-08-07	0.73014	59.2	1407	9580
swe-ita	tatoeba-test-v2021-08-07	0.73211	56.6	715	4711
swe-por	tatoeba-test-v2021-08-07	0.68146	48.7	320	2032
swe-spa	tatoeba-test-v2021-08-07	0.71373	55.3	1351	8235
dan-cat	flores101-devtest	0.59224	33.4	1012	27304
dan-fra	flores101-devtest	0.63387	38.3	1012	28343
dan-glg	flores101-devtest	0.54446	26.4	1012	26582
dan-ita	flores101-devtest	0.55237	25.7	1012	27306
dan-por	flores101-devtest	0.62233	36.9	1012	26519
dan-ron	flores101-devtest	0.58235	31.8	1012	26799
dan-spa	flores101-devtest	0.52453	24.3	1012	29199
isl-cat	flores101-devtest	0.48930	22.7	1012	27304
isl-fra	flores101-devtest	0.52704	26.2	1012	28343
isl-glg	flores101-devtest	0.45387	18.0	1012	26582
isl-ita	flores101-devtest	0.47303	18.6	1012	27306
isl-por	flores101-devtest	0.51381	24.9	1012	26519
isl-ron	flores101-devtest	0.48224	21.6	1012	26799
isl-spa	flores101-devtest	0.45786	18.1	1012	29199
nob-cat	flores101-devtest	0.55984	28.9	1012	27304
nob-fra	flores101-devtest	0.60102	33.8	1012	28343
nob-glg	flores101-devtest	0.52145	23.4	1012	26582
nob-ita	flores101-devtest	0.52619	22.2	1012	27306
nob-por	flores101-devtest	0.58836	32.2	1012	26519
nob-ron	flores101-devtest	0.54845	27.6	1012	26799
nob-spa	flores101-devtest	0.50661	21.8	1012	29199
swe-cat	flores101-devtest	0.58542	32.4	1012	27304
swe-fra	flores101-devtest	0.63688	39.3	1012	28343
swe-glg	flores101-devtest	0.53989	26.0	1012	26582
swe-ita	flores101-devtest	0.55232	25.9	1012	27306
swe-por	flores101-devtest	0.61882	36.5	1012	26519
swe-ron	flores101-devtest	0.57419	31.0	1012	26799
swe-spa	flores101-devtest	0.52175	23.8	1012	29199

Citation Information

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",
}

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: 8b9f0b0
port time: Sat Aug 13 00:00:00 EEST 2022
port machine: LM0-400-22516.local

Helsinki-NLP/opus-mt-tc-big-gmq-itc

opus-mt-tc-big-gmq-itc

Table of Contents

Model Details