As part of the commitment of our unit to open source the code of our scientific contributions, the code and models used in recent papers can be found in our public GitHub repository. The main README contains a list of all our works whose code is included in the repository, with links to a dedicated README for each work.
The repository is a fork of a 2021 version of fairseq maintained by us, following open source guidelines to ensure software quality:
- for every new commit, unit tests (UTs) are added and executed with a standard continuous integration (CI) pipeline ensuring that new contributions do not break existing features and cause regressions;
- templates for merge requests (MRs) are required to provide context and documentation regarding the code added, and these descriptions are included in commit messages;
- all contributions are internally peer-reviewed by another member of the group to improve the readability of
the code and style coherence of the codebase.
The adoption of these practices is a guarantee of the functionality, code quality, and stability of the codebase, both for the MT unit at FBK and for people interested in using and/or contributing to our repository, and distinguish it from the original fairseq repository where breaking changes are frequently introduced without notification and documentation, due to the lack of working and complete UTs and CI.
If you experience any issue with the software or you have questions about it, open an issue in the repository.
The repository licensed under the permissive Apache License, Version 2.0, and is a fork of fairseq that is released under the MIT license. If using the repository, please consider citing the related papers.