I suggest publish basics primitives like adapters, serializers and some usefull abstract features as separate library on pipy.
This will allow use and enhance core for all developers. I already have similar ideas for standartization model serializing and loading. And it will be great to "make ML great again" together without copy-paste :)