神经机器翻译模型,用于从英语(en)翻译到法语(fr)。
该模型是 OPUS-MT project 的一部分,该项目旨在为世界上的许多语言提供广泛可用且易于访问的神经机器翻译模型。所有模型最初是使用 Marian NMT 的出色框架进行训练的,后者是一个用纯C ++编写的高效NMT实现。这些模型已使用Huggingface的transformers库将其转换为pyTorch。训练数据取自 OPUS ,训练流程使用了 OPUS-MT-train 的程序。
@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",
}
简短的示例代码:
from transformers import MarianMTModel, MarianTokenizer
src_text = [
"The Portuguese teacher is very demanding.",
"When was your last hearing test?"
]
model_name = "pytorch-models/opus-mt-tc-big-en-fr"
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:
# Le professeur de portugais est très exigeant.
# Quand a eu lieu votre dernier test auditif ?
您还可以使用transformers pipelines使用OPUS-MT模型,例如:
from transformers import pipeline
pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-en-fr")
print(pipe("The Portuguese teacher is very demanding."))
# expected output: Le professeur de portugais est très exigeant.
| langpair | testset | chr-F | BLEU | #sent | #words |
|---|---|---|---|---|---|
| eng-fra | tatoeba-test-v2021-08-07 | 0.69621 | 53.2 | 12681 | 106378 |
| eng-fra | flores101-devtest | 0.72494 | 52.2 | 1012 | 28343 |
| eng-fra | multi30k_test_2016_flickr | 0.72361 | 52.4 | 1000 | 13505 |
| eng-fra | multi30k_test_2017_flickr | 0.72826 | 52.8 | 1000 | 12118 |
| eng-fra | multi30k_test_2017_mscoco | 0.73547 | 54.7 | 461 | 5484 |
| eng-fra | multi30k_test_2018_flickr | 0.66723 | 43.7 | 1071 | 15867 |
| eng-fra | newsdiscussdev2015 | 0.60471 | 33.4 | 1500 | 27940 |
| eng-fra | newsdiscusstest2015 | 0.64915 | 40.3 | 1500 | 27975 |
| eng-fra | newssyscomb2009 | 0.58903 | 30.7 | 502 | 12331 |
| eng-fra | news-test2008 | 0.55516 | 27.6 | 2051 | 52685 |
| eng-fra | newstest2009 | 0.57907 | 30.0 | 2525 | 69263 |
| eng-fra | newstest2010 | 0.60156 | 33.5 | 2489 | 66022 |
| eng-fra | newstest2011 | 0.61632 | 35.0 | 3003 | 80626 |
| eng-fra | newstest2012 | 0.59736 | 32.8 | 3003 | 78011 |
| eng-fra | newstest2013 | 0.59700 | 34.6 | 3000 | 70037 |
| eng-fra | newstest2014 | 0.66686 | 41.9 | 3003 | 77306 |
| eng-fra | tico19-test | 0.63022 | 40.6 | 2100 | 64661 |
该工作得到 European Language Grid 和 pilot project 2866 的支持,由 FoTran project 提供资助,该项目受欧洲研究理事会(ERC)在欧洲联盟的Horizon 2020研究与创新计划(授权号771113)下的支持,并且由 MeMAD project 在欧盟的Horizon 2020研究与创新计划下的资助协议号780069提供支持。我们还要感谢提供的慷慨计算资源和IT基础设施 CSC -- IT Center for Science ,芬兰。