The release of the machine translation system OpenNMT 0.28.0 (Open Neural Machine Translation), which uses machine learning methods, has been published. To build a neural network, the project uses the capabilities of the TensorFlow deep learning library. The code of the modules developed by the OpenNMT project is written in Python and distributed under the MIT license. Ready-made models are prepared for English, German and Catalan languages, for other languages you can independently form a model based on a data set from the OPUS project (two files are transferred to the system for training - one with sentences in the source language, and the second with a high-quality translation of these sentences into the target language ).
The project is being developed with the participation of SYSTRAN, a company specializing in the creation of machine translation tools, and a group of researchers from Harvard, which develops human language models for machine learning systems. The user interface is maximally simplified and requires only specifying the input file with the text and the file to save the translation result. The extension system makes it possible to implement additional functionality on the basis of OpenNMT, for example, autoabstracting, text classification and subtitle generation.
The use of TensorFlow allows you to use the capabilities of the GPU (to speed up the learning process of the neural network. To simplify the distribution of the product, the project also develops a self-sufficient version of the C++ translator - CTranslate2, which uses pre-trained models without being tied to additional dependencies.
The new version adds the initial_learning_rate parameter and implements several new arguments (mha_bias and output_layer_bias) to customize the Transformer model generator. The rest is marked by bug fixes.
Source: opennet.ru