High-efficiency floating-point neural network inference operators for mobile, server, and Web - View it on GitHub