language:
- da
- is
- nb
- nn
- sv
- multilingual
license: cc-by-4.0
tags:
- translation
- opus-mt-tc
model-index:
- name: opus-mt-tc-big-gmq-gmq
results:
- task:
type: translation
name: Translation isl-swe
dataset:
name: europeana2021
type: europeana2021
args: isl-swe
metrics:
- type: bleu
value: 22.2
name: BLEU
- type: chrf
value: 0.45562
name: chr-F
- type: bleu
value: 29.7
name: BLEU
- type: chrf
value: 0.54171
name: chr-F
- type: bleu
value: 54
name: BLEU
- type: chrf
value: 0.73891
name: chr-F
- task:
type: translation
name: Translation dan-isl
dataset:
name: flores101-devtest
type: flores_101
args: dan isl devtest
metrics:
- type: bleu
value: 22.2
name: BLEU
- type: chrf
value: 0.50227
name: chr-F
- type: bleu
value: 28.6
name: BLEU
- type: chrf
value: 0.58445
name: chr-F
- type: bleu
value: 38.5
name: BLEU
- type: chrf
value: 0.65
name: chr-F
- type: bleu
value: 27.2
name: BLEU
- type: chrf
value: 0.5363
name: chr-F
- type: bleu
value: 20.5
name: BLEU
- type: chrf
value: 0.49434
name: chr-F
- type: bleu
value: 26
name: BLEU
- type: chrf
value: 0.53373
name: chr-F
- type: bleu
value: 31.7
name: BLEU
- type: chrf
value: 0.59657
name: chr-F
- type: bleu
value: 18.9
name: BLEU
- type: chrf
value: 0.47432
name: chr-F
- type: bleu
value: 31.3
name: BLEU
- type: chrf
value: 0.6003
name: chr-F
- type: bleu
value: 39
name: BLEU
- type: chrf
value: 0.6434
name: chr-F
- type: bleu
value: 21.7
name: BLEU
- type: chrf
value: 0.4959
name: chr-F
- type: bleu
value: 28.9
name: BLEU
- type: chrf
value: 0.58336
name: chr-F
- task:
type: translation
name: Translation dan-nob
dataset:
name: tatoeba-test-v2021-08-07
type: tatoeba_mt
args: dan-nob
metrics:
- type: bleu
value: 78.2
name: BLEU
- type: chrf
value: 0.87556
name: chr-F
- type: bleu
value: 72.5
name: BLEU
- type: chrf
value: 0.83556
name: chr-F
- type: bleu
value: 78.9
name: BLEU
- type: chrf
value: 0.88349
name: chr-F
- type: bleu
value: 73.9
name: BLEU
- type: chrf
value: 0.85345
name: chr-F
- type: bleu
value: 55.2
name: BLEU
- type: chrf
value: 0.74571
name: chr-F
- type: bleu
value: 73.9
name: BLEU
- type: chrf
value: 0.84747
name: chr-F
- type: bleu
value: 72.6
name: BLEU
- type: chrf
value: 0.83392
name: chr-F
- type: bleu
value: 76.3
name: BLEU
- type: chrf
value: 0.85815
name: chr-F
opus-mt-tc-big-gmq-gmq
Table of Contents
- Model Details
- Uses
- Risks, Limitations and Biases
- How to Get Started With the Model
- Training
- Evaluation
- Citation Information
- Acknowledgements
Model Details
Neural machine translation model for translating from North Germanic languages (gmq) to North Germanic languages (gmq).
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: 2022-07-29
- License: CC-BY-4.0
- Language(s):
- Source Language(s): dan fao isl nno nob nor swe
- Target Language(s): dan isl nno nob nor swe
- Language Pair(s): dan-isl dan-nob dan-swe isl-dan isl-nob isl-swe nno-nob nob-dan nob-isl nob-nno nob-swe swe-dan swe-isl swe-nob
- Valid Target Language Labels: >>dan<< >>fao<< >>isl<< >>jut<< >>nno<< >>nob<< >>non<< >>nrn<< >>ovd<< >>qer<< >>rmg<< >>swe<<
- Original Model: opusTCv20210807_transformer-big_2022-07-29.zip
- Resources for more information:
- OPUS-MT-train GitHub Repo
- More information about released models for this language pair: OPUS-MT gmq-gmq 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. >>dan<<
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:
from transformers import MarianMTModel, MarianTokenizer
src_text = [
">>fao<< Jeg er bange for kakerlakker.",
">>nob<< Vladivostok �r en stad i Ryssland."
]
model_name = "pytorch-models/opus-mt-tc-big-gmq-gmq"
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:
# Ta� eru uml.
# Vladivostok er en by i Russland.
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-gmq-gmq")
print(pipe(">>fao<< Jeg er bange for kakerlakker."))
# expected output: Ta� eru uml.
Training
- Data: opusTCv20210807 (source)
- Pre-processing: SentencePiece (spm32k,spm32k)
- Model Type: transformer-big
- Original MarianNMT Model: opusTCv20210807_transformer-big_2022-07-29.zip
- Training Scripts: GitHub Repo
Evaluation
- test set translations: opusTCv20210807_transformer-big_2022-07-29.test.txt
- test set scores: opusTCv20210807_transformer-big_2022-07-29.eval.txt
- benchmark results: benchmark_results.txt
- benchmark output: benchmark_translations.zip
| langpair | testset | chr-F | BLEU | #sent | #words |
|---|---|---|---|---|---|
| dan-nob | tatoeba-test-v2021-08-07 | 0.87556 | 78.2 | 1299 | 9620 |
| dan-swe | tatoeba-test-v2021-08-07 | 0.83556 | 72.5 | 1549 | 10060 |
| nno-nob | tatoeba-test-v2021-08-07 | 0.88349 | 78.9 | 467 | 3129 |
| nob-dan | tatoeba-test-v2021-08-07 | 0.85345 | 73.9 | 1299 | 9794 |
| nob-nno | tatoeba-test-v2021-08-07 | 0.74571 | 55.2 | 466 | 3141 |
| nob-swe | tatoeba-test-v2021-08-07 | 0.84747 | 73.9 | 563 | 3698 |
| swe-dan | tatoeba-test-v2021-08-07 | 0.83392 | 72.6 | 1549 | 10239 |
| swe-nob | tatoeba-test-v2021-08-07 | 0.85815 | 76.3 | 563 | 3708 |
| isl-swe | europeana2021 | 0.45562 | 22.2 | 563 | 10293 |
| nob-isl | europeana2021 | 0.54171 | 29.7 | 538 | 9932 |
| nob-swe | europeana2021 | 0.73891 | 54.0 | 538 | 9885 |
| dan-isl | flores101-devtest | 0.50227 | 22.2 | 1012 | 22834 |
| dan-nob | flores101-devtest | 0.58445 | 28.6 | 1012 | 23873 |
| dan-swe | flores101-devtest | 0.65000 | 38.5 | 1012 | 23121 |
| isl-dan | flores101-devtest | 0.53630 | 27.2 | 1012 | 24638 |
| isl-nob | flores101-devtest | 0.49434 | 20.5 | 1012 | 23873 |
| isl-swe | flores101-devtest | 0.53373 | 26.0 | 1012 | 23121 |
| nob-dan | flores101-devtest | 0.59657 | 31.7 | 1012 | 24638 |
| nob-isl | flores101-devtest | 0.47432 | 18.9 | 1012 | 22834 |
| nob-swe | flores101-devtest | 0.60030 | 31.3 | 1012 | 23121 |
| swe-dan | flores101-devtest | 0.64340 | 39.0 | 1012 | 24638 |
| swe-isl | flores101-devtest | 0.49590 | 21.7 | 1012 | 22834 |
| swe-nob | flores101-devtest | 0.58336 | 28.9 | 1012 | 23873 |
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 23:59:02 EEST 2022
- port machine: LM0-400-22516.local