I’m pleased to announce the release of OpenNMT-tf 2.0!
The primary goal of this first major update is to use the new features and practices introduced by TensorFlow 2.0.
OpenNMT-tf 2.0 is the result of several months of planning and work. In my (biased) opinion, the changes are largely positive: the implementation is simpler, more consistent, and more modular. In particular, we carefully defined and documented a stable public API with many components that can be easily reused in other projects, from dataset utilities to beam search decoding.
I invite you to take a look at the changes and give it a try!
Note: As a major release, several breaking changes were introduced. If you don’t want to make the transition just yet, you can always freeze your version to OpenNMT-tf v1 (e.g. with
OpenNMT-tf==1.* in your
requirements.txt file). We will still push important bug fixes to the v1 line if needed.