用于将法语(fra)翻译为英语(eng)的神经机器翻译模型。
该模型是 OPUS-MT project 的一部分,这是一个旨在为世界上许多语言提供广泛可用和可访问的神经机器翻译模型的努力。所有模型最初都是使用 Marian NMT 的出色框架进行训练的,该框架是用纯C++编写的高效NMT实现。这些模型使用转换器库(huggingface)将其转换为pyTorch。训练数据来自 OPUS ,并且训练流程使用 OPUS-MT-train 的步骤。
@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 = [
"J'ai adoré l'Angleterre.",
"C'était la seule chose à faire."
]
model_name = "pytorch-models/opus-mt-tc-big-fr-en"
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:
# I loved England.
# It was the only thing to do.
您还可以使用transformers工具包的pipelines功能,例如:
from transformers import pipeline
pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-fr-en")
print(pipe("J'ai adoré l'Angleterre."))
# expected output: I loved England.
| langpair | testset | chr-F | BLEU | #sent | #words |
|---|---|---|---|---|---|
| fra-eng | tatoeba-test-v2021-08-07 | 0.73772 | 59.8 | 12681 | 101754 |
| fra-eng | flores101-devtest | 0.69350 | 46.0 | 1012 | 24721 |
| fra-eng | multi30k_test_2016_flickr | 0.68005 | 49.7 | 1000 | 12955 |
| fra-eng | multi30k_test_2017_flickr | 0.70596 | 52.0 | 1000 | 11374 |
| fra-eng | multi30k_test_2017_mscoco | 0.69356 | 50.6 | 461 | 5231 |
| fra-eng | multi30k_test_2018_flickr | 0.65751 | 44.9 | 1071 | 14689 |
| fra-eng | newsdiscussdev2015 | 0.59008 | 34.4 | 1500 | 27759 |
| fra-eng | newsdiscusstest2015 | 0.62603 | 40.2 | 1500 | 26982 |
| fra-eng | newssyscomb2009 | 0.57488 | 31.1 | 502 | 11818 |
| fra-eng | news-test2008 | 0.54316 | 26.5 | 2051 | 49380 |
| fra-eng | newstest2009 | 0.56959 | 30.4 | 2525 | 65399 |
| fra-eng | newstest2010 | 0.59561 | 33.4 | 2489 | 61711 |
| fra-eng | newstest2011 | 0.60271 | 33.8 | 3003 | 74681 |
| fra-eng | newstest2012 | 0.59507 | 33.6 | 3003 | 72812 |
| fra-eng | newstest2013 | 0.59691 | 34.8 | 3000 | 64505 |
| fra-eng | newstest2014 | 0.64533 | 39.4 | 3003 | 70708 |
| fra-eng | tico19-test | 0.63326 | 41.3 | 2100 | 56323 |
该工作得到 European Language Grid 和 pilot project 2866 的支持,由 FoTran project 提供资金支持,该项目由欧盟的Horizon 2020研究和创新计划(授权协议号:771113)下的欧洲研究委员会(ERC)资助,以及 MeMAD project 项目,该项目由欧盟的Horizon 2020研究和创新计划资助,协议编号为780069。我们还感谢 CSC -- IT Center for Science 提供的慷慨计算资源和IT基础设施,芬兰。