CytonMT: an Efficient Neural Machine Translation Open-source Toolkit Implemented in C++

Xiaolin Wang, Masao Utiyama, Eiichiro Sumita

Abstract: This paper presented an open-source neural machine translation toolkit named CytonMT\footnote{this https URL}. The toolkit was built from scratch using C++ and Nvidia's GPU-accelerated libraries. The toolkit featured training efficiency, code simplicity and translation quality. Benchmarks showed that cytonMT accelerated the training speed by 64.5\% to 110.8\% and achieved a high translation quality only lower than the Google's production engine among the NMT systems of attention-based RNN encoder-decoder.