Today, Harvard NLP (Harvard University, Natural Language Processing research group) announced its open source development Machine Translation neural system OpenNMT, the system uses the Torch mathematical toolkit, industrial-strength has reached the level of production.
OpenNMT system design is simple and easy to use, easy to expand, while maintaining efficiency and the most advanced translation accuracy.
Simple generic interface requires only source / destination files.
Fast high performance GPU training and memory optimization.
Recent achievements in improving translation performance.
Pre training model for multiple languages (upcoming).
Allows expansion of other sequence generation tasks, such as sum of graphs.