A decade ago, when traditional machine learning techniques were first being commercialized, training was incredibly hard and expensive, but because models were relatively small, inference – running ...