Not sure I see a lot of benefits here as compared to just running the model on a centralised highend system. I’m interested to know more about this approach though.