Training is not open source (yet?).

Among other questions, there is an issue with the copyrighted dataset that needs discussion.