Method and device for reducing memory occupation, electronic equipment and readable storage medium

Abstract

The invention provides a method and a device for reducing memory occupation, electronic equipment and a readable storage medium, and the method comprises the steps that terminal equipment firstly obtains a plurality of target network models when executing the method, and then splices the target network models meeting a splicing condition to obtain at least one spliced network model; and for each spliced network model, obtaining the memory usage amount required to be occupied when each layer of network in the spliced network model runs, and then determining the maximum value in the plurality ofmemory usage amounts as the total memory usage amount of the spliced network model. When the memory is distributed to the spliced network model, only the memory with the same numerical value as the total memory occupancy amount can be distributed, and compared with the prior art, the memory occupancy amount required for running of the spliced network model is greatly reduced.

Description

technical field [0001] The present invention relates to the field of data processing, in particular to a method, device, electronic equipment, and readable storage medium for reducing memory usage. Background technique [0002] When the neural network model is running, it often needs to occupy a large amount of memory. However, the current smart terminals usually have strict memory limitations. Therefore, when the neural network model runs on the smart terminal, the realization effect of the neural network model will be limited. [0003] In order to solve this problem, the traditional method is to realize memory multiplexing between different layers in a single neural network model, so as to reduce the memory usage of the neural network model. However, when multiple neural network models need to be run on the same device, traditional methods are of little help. Therefore, there is a need for a method of reducing the memory footprint of a neural network model at runtime to a...

AI Technical Summary

Problems solved by technology

However, when multiple neural network models need to be run on the same device, traditional methods are of little use

Images