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