Model compression method and system and computer readable medium

Abstract

The invention relates to a model compression method and system and a computer readable medium. The model compression method comprises the steps of 1, obtaining a trained original model; 2, searching the optimal pruning rate of each layer in the original model, and then performing filter-level pruning on the model; 3, quantifying the pruned model; and 4, performing knowledge migration in a knowledge distillation mode, and then performing fine tuning on the model after knowledge migration on the training data set to obtain a compression model. Compared with the prior art, the invention has the advantages of automatic compression, adaptability to hardware environment, small model size, good performance, convenience, rapidness and the like.

Description

technical field [0001] The present invention relates to the technical field of model compression, in particular to a model compression method, system and computer-readable medium based on neural architecture search. Background technique [0002] With the rapid development of deep learning technology in recent years, deep neural networks have shown excellent performance in many tasks, such as computer vision, speech recognition, natural language processing, etc., deep learning models rely on their powerful The learning ability has achieved the best performance on many mainstream data sets. However, deep learning models often have a large number of parameters, which take up a lot of computing resources during the training and inference stages, so they cannot be deployed on some resource-constrained devices, such as mobile phones, embedded devices, etc. . [0003] The goal of model compression is to achieve a model that is simplified from the original model without significan...

AI Technical Summary

Problems solved by technology

[0005] The model compression method in the prior art needs to manually set the pruning rate of each layer, and manual intervention is required in the process of model compression, and the degree of automation is not high

In addition, the specific environment needs to be considered when compressing the model to determine the extent to which the model needs to be compressed. However, the existing model compression methods do not consider whether the model compression is suitable for the hardware environment. Although the compression effect is good, they do not consider Is it suitable for the current hardware environment, or is it overcompressed

Images