v2 published (v15 internally) longer training time with better accuracy.

First import of functional embedding model.