An open source neural machine translation system.


January 2021 Release of OpenNMT-py 2.0 featuring a new dynamic data pipeline
May 2020 OpenNMT models are the most efficient on several metrics at WNGT 2020 using the CTranslate2 inference engine.
October 2019 Release of OpenNMT-tf 2.0 using TensorFlow 2.0.
October 2019 Publication of CTranslate2, a redesign of the original CTranslate to support fast execution of OpenNMT-py and OpenNMT-tf models.
August 2018 OpenNMT publishes the fastest model running on a single CPU core at WNMT 2018 using the CTranslate engine.
July 2018 Last version of the original LuaTorch implementation, now fully superseded by OpenNMT-py and PyTorch.
March 2018 The first OpenNMT workshop is held in Paris gathering more than 100 people from around the world.
November 2017 The TensorFlow version OpenNMT-tf is released.
July 2017 The project is awarded “Best Demonstration Paper Runner-Up” at ACL 2017.
March 2017 The PyTorch version OpenNMT-py is released in collaboration with the Facebook AI Research team.
January 2017 Release of CTranslate, a custom and lightweight inference engine for OpenNMT models.
December 2016 Initial release of OpenNMT, the original implementation using LuaTorch.
June 2016 Yoon Kim from the Harvard NLP group publishes the project seq2seq-attn that lays the foundation of the OpenNMT initiative.