OpenNMT is an open-source toolkit for neural machine translation (NMT) developed by Harvard University and SYSTRAN. The toolkit aims to provide an efficient, modular, and extensible framework for NMT research and production use. It supports a wide range of NMT tasks, including model architectures, feature representations, and source modalities, while maintaining competitive performance and reasonable training requirements. The toolkit includes modeling and translation support, as well as detailed pedagogical documentation. It is designed to be efficient and accurate enough for production contexts. OpenNMT is a complete NMT implementation, with three main aims: (a) prioritize training and test efficiency, (b) maintain model modularity and readability, (c) support significant research extensibility. OpenNMT is implemented in the Lua/Torch framework and has been extended to support Python/PyTorch. It is available on GitHub under an MIT license and has been widely adopted by the research community. The system is efficient, with a compact codebase and minimal dependencies. It supports multiple languages and has been used in various research projects. The toolkit includes several different translation deployments, including a batched CPU/GPU implementation, a simple single-instance implementation for mobile devices, and a specialized C implementation. It also includes additional tools for tokenization, word embeddings, and visualization. OpenNMT has been benchmarked on several translation tasks, including English-to-German (EN→DE) using the WMT 2015 dataset. It has shown competitive performance compared to the publicly available Nematus system. Additionally, it has been adapted for non-standard MT tasks such as sentence document summarization and dialogue response generation. The toolkit is designed to be extensible, allowing for future variants of NMT models. It supports multiple modalities, including image-to-text generation and speech-to-text recognition. OpenNMT provides a stable and flexible framework for both research and production use.OpenNMT is an open-source toolkit for neural machine translation (NMT) developed by Harvard University and SYSTRAN. The toolkit aims to provide an efficient, modular, and extensible framework for NMT research and production use. It supports a wide range of NMT tasks, including model architectures, feature representations, and source modalities, while maintaining competitive performance and reasonable training requirements. The toolkit includes modeling and translation support, as well as detailed pedagogical documentation. It is designed to be efficient and accurate enough for production contexts. OpenNMT is a complete NMT implementation, with three main aims: (a) prioritize training and test efficiency, (b) maintain model modularity and readability, (c) support significant research extensibility. OpenNMT is implemented in the Lua/Torch framework and has been extended to support Python/PyTorch. It is available on GitHub under an MIT license and has been widely adopted by the research community. The system is efficient, with a compact codebase and minimal dependencies. It supports multiple languages and has been used in various research projects. The toolkit includes several different translation deployments, including a batched CPU/GPU implementation, a simple single-instance implementation for mobile devices, and a specialized C implementation. It also includes additional tools for tokenization, word embeddings, and visualization. OpenNMT has been benchmarked on several translation tasks, including English-to-German (EN→DE) using the WMT 2015 dataset. It has shown competitive performance compared to the publicly available Nematus system. Additionally, it has been adapted for non-standard MT tasks such as sentence document summarization and dialogue response generation. The toolkit is designed to be extensible, allowing for future variants of NMT models. It supports multiple modalities, including image-to-text generation and speech-to-text recognition. OpenNMT provides a stable and flexible framework for both research and production use.