Quantization method, device and equipment of text feature extraction model and storage medium

Abstract

The embodiment of the invention discloses a quantification method and device of a text feature extraction model, equipment and a storage medium. The method comprises the steps that in the training process of a text feature extraction model, a quantization interval corresponding to a target quantization parameter is determined according to the target quantization parameter; wherein the quantization interval comprises a part of floating point values in the target quantization parameter; constructing a mapping relationship between a floating point value and a fixed point value of the target quantization parameter according to the quantization interval; wherein the floating point value, smaller than the left end point of the quantization interval, in the target quantization parameter is mapped into a quantized minimum fixed point value, and the floating point value, larger than the right end point of the quantization interval, in the target quantization parameter is mapped into a quantized maximum fixed point value; and performing quantization operation on the target quantization parameter based on the mapping relationship. According to the quantification method of the text feature extraction model provided by the embodiment of the invention, the accuracy of the text feature extraction model can be effectively improved while the running speed of the text feature extraction model is improved.

Description

technical field [0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular, to a quantification method, device, device, and storage medium of a text feature extraction model. Background technique [0002] In natural language processing, text feature extraction models have been more and more widely used. At present, the popular text feature extraction models are the Transformer model and its various variant models. These models have been achieved in various natural language processing tasks. for the best results. However, the text feature extraction model has a large amount of model parameters and calculations, and is limited by computer storage and computing capabilities, which brings great problems to the fast reasoning of the text feature extraction model and the deployment of the model on the mobile terminal. [0003] At present, the most direct solution to this problem is to quantify the text feature extraction model, that...

AI Technical Summary

Problems solved by technology

However, after quantifying the text feature extraction model, the accuracy of the text feature extraction model will be greatly reduced. Therefore, it is very important to optimize the quantification method of the text feature extraction model and reduce the loss of accuracy.

Images