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: 2022-08-10
License: CC-BY-4.0
Language(s):
- Source Language(s): bos_Latn bul hbs hrv mkd slv srp_Cyrl srp_Latn
- Target Language(s): fra ita por ron spa
- Language Pair(s): bul-fra bul-ita bul-por bul-ron bul-spa hbs-fra hbs-ita hbs-spa hrv-fra hrv-ita hrv-por hrv-ron hrv-spa mkd-fra mkd-ita mkd-por mkd-ron mkd-spa slv-fra slv-ita slv-por slv-ron slv-spa srp_Cyrl-fra srp_Cyrl-ita srp_Cyrl-por srp_Cyrl-ron srp_Cyrl-spa srp_Latn-ita
- 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<< >>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_2022-08-10.zip
Resources for more information:
- OPUS-MT-train GitHub Repo
- More information about released models for this language pair: OPUS-MT zls-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 = [
    ">>fra<< Dobar dan, kako si?",
    ">>spa<< Znam da je ovo čudno."
]

model_name = "pytorch-models/opus-mt-tc-big-zls-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:
#     Bonjour, comment allez-vous ?
#     Sé que esto es raro.

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-zls-itc")
print(pipe(">>fra<< Dobar dan, kako si?"))

# expected output: Bonjour, comment allez-vous ?

Training

Data: opusTCv20210807 (source)
Pre-processing: SentencePiece (spm32k,spm32k)
Model Type: transformer-big
Original MarianNMT Model: opusTCv20210807_transformer-big_2022-08-10.zip
Training Scripts: GitHub Repo

Evaluation

test set translations: opusTCv20210807_transformer-big_2022-08-10.test.txt
test set scores: opusTCv20210807_transformer-big_2022-08-10.eval.txt
benchmark results: benchmark_results.txt
benchmark output: benchmark_translations.zip

langpair	testset	chr-F	BLEU	#sent	#words
bul-fra	tatoeba-test-v2021-08-07	0.68971	52.9	446	3669
bul-ita	tatoeba-test-v2021-08-07	0.66412	45.1	2500	16951
bul-spa	tatoeba-test-v2021-08-07	0.66672	49.7	286	1783
hbs-fra	tatoeba-test-v2021-08-07	0.66434	48.1	474	3370
hbs-ita	tatoeba-test-v2021-08-07	0.72381	53.5	534	3208
hbs-spa	tatoeba-test-v2021-08-07	0.73105	58.0	607	3766
hrv-fra	tatoeba-test-v2021-08-07	0.62800	44.3	258	1943
hrv-spa	tatoeba-test-v2021-08-07	0.71370	57.5	254	1702
mkd-spa	tatoeba-test-v2021-08-07	0.75366	62.1	217	1121
srp_Latn-ita	tatoeba-test-v2021-08-07	0.76045	59.6	212	1292
bul-fra	flores101-devtest	0.60640	34.4	1012	28343
bul-ita	flores101-devtest	0.54135	24.0	1012	27306
bul-por	flores101-devtest	0.59322	32.4	1012	26519
bul-ron	flores101-devtest	0.55558	27.1	1012	26799
bul-spa	flores101-devtest	0.50962	22.4	1012	29199
hrv-fra	flores101-devtest	0.59349	33.1	1012	28343
hrv-ita	flores101-devtest	0.52980	23.5	1012	27306
hrv-por	flores101-devtest	0.57402	30.2	1012	26519
hrv-ron	flores101-devtest	0.53650	25.9	1012	26799
hrv-spa	flores101-devtest	0.50161	21.5	1012	29199
mkd-fra	flores101-devtest	0.60801	35.2	1012	28343
mkd-ita	flores101-devtest	0.53543	23.9	1012	27306
mkd-por	flores101-devtest	0.59648	33.9	1012	26519
mkd-ron	flores101-devtest	0.54998	28.0	1012	26799
mkd-spa	flores101-devtest	0.51079	22.8	1012	29199
slv-fra	flores101-devtest	0.58233	31.5	1012	28343
slv-ita	flores101-devtest	0.52390	22.4	1012	27306
slv-por	flores101-devtest	0.56436	29.0	1012	26519
slv-ron	flores101-devtest	0.53116	25.0	1012	26799
slv-spa	flores101-devtest	0.49621	21.1	1012	29199
srp_Cyrl-fra	flores101-devtest	0.62110	36.0	1012	28343
srp_Cyrl-ita	flores101-devtest	0.54083	23.9	1012	27306
srp_Cyrl-por	flores101-devtest	0.61248	34.9	1012	26519
srp_Cyrl-ron	flores101-devtest	0.56235	28.8	1012	26799
srp_Cyrl-spa	flores101-devtest	0.51698	22.8	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: Fri Aug 12 23:59:29 EEST 2022
port machine: LM0-400-22516.local

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

opus-mt-tc-big-zls-itc

Table of Contents

Model Details