Heterogeneous intelligent processing quantization device, quantization method, electronic equipment and storage medium

Abstract

The invention provides a heterogeneous intelligent processing quantification device. The heterogeneous intelligent processing device is applied to the heterogeneous intelligent processor. The heterogeneous intelligent processor comprises a general processor and an intelligent processor; the intelligent processor comprises a control circuit and a plurality of operational circuits. The heterogeneousintelligent processing quantization device comprises a weight quantization module used for quantizing weights in an original neural network to enable the weights in a shift quantization neural network to be power data, and retraining the shift quantization neural network to enable the precision of the shift quantization neural network to be the same as that of the original neural network; and theneuron quantization module is used for quantizing neurons in the original neural network, so that the neurons in the shift quantization neural network are quantized into fixed-point number data. Thestorage overhead of the neural network can be reduced, and the data processing efficiency of the neural network is improved.

Description

technical field [0001] The present disclosure relates to the technical field of mobile terminals, and in particular to a heterogeneous intelligent processing quantization device, a neural network quantization method, electronic equipment, and a storage medium. Background technique [0002] A neural network is a mathematical model that imitates the structure and function of a biological neural network. Neural networks are widely used in image recognition, speech recognition, natural speech processing and other fields. However, with the increasing complexity of the neural network, the data in the neural network is constantly increasing, and data processing is a necessary step in the neural network. [0003] In mobile terminals, a large number of hardware computing units such as GPUs and CPUs currently in use in mobile terminal scenarios still use neural networks represented by floating-point numbers or fixed-point numbers, and data processing still requires a large number of ...

AI Technical Summary

Problems solved by technology

[0003] In mobile terminals, a large number of hardware computing units such as GPUs and CPUs currently in use in mobile terminal scenarios still use neural networks represented by floating-point numbers or fixed-point numbers, and data processing still requires a large number of floating-point number multiplications or fixed-point numbers. Point number multiplication calculation, but the calculation adopts floating-point number calculation or fixed-point number multiplication calculation with huge power consumption, and the data processing efficiency is low and the speed is slow

Images