Text classification model packaging method, text classification method and related equipment

Abstract

The invention relates to the technical field of natural language processing, and discloses a text classification model packaging method, a text classification method and related device. The text classification model packaging method comprises the steps: extracting a Transform part in a language model, and and packaging the extracted Transform part to generate a vector conversion module; in response to a text classification model training request, determining a classification scene of a text classification model corresponding to the text classification model training request; obtaining a training text corresponding to the classification scene; inputting the training text into the vector conversion module to obtain a sentence vector corresponding to the training text; training a full-connection shallow neural network based on the sentence vector corresponding to the training text to obtain a text classification model; and packaging the text classification model to generate a docker mirror image corresponding to the text classification model. The language recognition precision of the language model can be ensured while the volume of the language model is reduced.

Description

technical field [0001] The present application relates to the technical field of natural language processing, and in particular to a text classification model packaging method, a text classification method, a text classification model packaging device, computer equipment, and a storage medium. Background technique [0002] Language models greatly improve the accuracy of overall natural language processing. However, the reasoning speed and model size of language models have always been criticized. For example, a model that has completed basic pre-training has nearly 10 million parameters. Although the language model can be compressed, the accuracy of the model is lost after the compression model. In addition, if you want to use the model as a text classifier, you need to fine-tune it in each scene. When deploying each scene, you need a high-performance GPU server for deployment, which wastes a lot of computing resources and significantly increases the computing cost. [000...

AI Technical Summary

Problems solved by technology

Although the language model can be compressed, the accuracy of the model is lost after the compression model

In addition, if you want to use the model as a text classifier, you need to fine-tune it in each scene. When deploying each scene, you need a high-performance GPU server for deployment, which wastes a lot of computing resources and significantly increases the computing cost.

Images