Actually the docs around this weren’t great. Took the train-in-python & inference-in-go approach. And only for versions greater than tf2