[jeffbee]

I can't think of any case where inference doesn't care about latency.

[cyanydeez]

I cant thinl of any reason training isnt going to become real time with a significant cpu budget.