OpenNMT

An open-source ecosystem for neural machine translation and sequence modeling maintained by SYSTRAN and Harvard NLP.

Launched as a collaboration between Harvard NLP and SYSTRAN: OpenNMT provides industrial-strength frameworks for neural machine translation and sequence generation. The project maintains two primary implementations (OpenNMT-py for PyTorch and OpenNMT-tf for TensorFlow) alongside CTranslate2: a custom inference engine designed for high-speed deployment. Developers use these tools to train production-grade models for tasks like document summarization: text-to-speech: and multilingual translation. The ecosystem supports advanced features including Transformer architectures: multi-GPU training: and dynamic data loading to handle massive datasets efficiently.

https://opennmt.net/