[imranq]

Is it possible to know the minimum model size / data set size it takes to train a model given certain efficiency parameters (latency, etc.)?