Method, apparatus and computer program product for neural network

Abstract

This disclosure discloses methods, apparatuses, and computer program products for neural networks. The method includes: receiving an input tensor at a linear layer in the neural network during neural network training; and performing mixed-precision quantization on the input tensor to obtain a quantized tensor, wherein the quantized tensor contains data of different precision types. Apparatus, computer program products, and non-volatile computer-readable storage media corresponding to this method are also provided.

Images