Submission date: 
04 November 2021

This document describes the packaging and release of the CUNI machine translation (MT) systems trained within the Tensor2tensor framework for sequence-to-sequence learning that was developed for task T4.5. The MT backend, together with a simple command-line interface (CLI) is released separately from the models. The MT backend is released as a separate Docker image as a platform-independent solution. The compilation of the docker images is separated into two parts: first, a base image containing the Tensor2tensor installation is prepared and second, a language-specific derivative of the base image is compiled for the chosen MT model. The independent release of the MT models also allows the user to download the models separately and execute them using their local Tensor2tensor installation if necessary.

Publication type: 
Deliverable