Model quantification method and device, electronic equipment and storage medium

Abstract

The invention relates to a model quantification method and device, electronic equipment and a storage medium. The method comprises the steps that a quantification model set is acquired; the quantization model set comprises at least two quantization models; the quantization model is obtained by quantizing at least one target network layer in a to-be-quantized model; determining fitness corresponding to each quantitative model; the fitness is used for representing performance improvement of the quantitative model; the performance improvement of the quantitative model is determined according to the difference between the model performance of the quantitative model and the model performance of a preset model; the preset model is obtained by quantizing each network layer in the to-be-quantized model; the model performance comprises model speed and model precision; and performing genetic algorithm optimization on the quantization strategy parameters corresponding to the quantization models according to the fitness corresponding to the quantization models to obtain a target quantization model. According to the invention, the quantification effect of the neural network model can be improved.

Description

technical field [0001] The present disclosure relates to the technical field of artificial intelligence, and in particular to a model quantification method, device, electronic equipment and storage medium. Background technique [0002] Artificial intelligence algorithms are widely used in short video special effects, content understanding, recommendation systems and other fields. The application of artificial intelligence algorithms in the industry requires three elements: algorithms, data, and computing power; among them, computing power occupies expensive operating costs, so it is becoming more and more important to improve the reasoning speed of algorithms. [0003] In the inference stage of deep learning algorithms, related technologies often map all the weights and activation values ​​expressed in floating-point data format in the neural network model to integer data format, so as to realize the quantization of the neural network model and improve the accuracy of the ne...

AI Technical Summary

Problems solved by technology

[0005] The present disclosure provides a neural network model quantification method, device, electronic equipment and storage medium to at least solve the problem of poor quantification effect in the neural network model quantification method in the related art

Images