[forresti]

Heh, I think the part of the point of PlaidML is to avoid the speed/efficiency limitations of TensorFlow in deployment/production.