Model hot update processing method and device, equipment, medium and product

Abstract

The invention discloses a model hot update processing method and device, equipment, a medium and a product, and the method comprises the steps: obtaining a model version number of a new neural network model in response to a new model deployment event acting on a current model service; according to a preset search field conversion rule, determining a model version search field corresponding to the model version number; maintaining the operation of the old neural network model in the current model service, continuously inputting the historical business data into the new neural network model, and correspondingly storing each new vector value output by the new neural network model into a model version search field of the vector search service; and after new vector values corresponding to all historical business data output by the new neural network model are stored in a vector search service, the offline old neural network model and the online new neural network model provide a reasoning function for the current model service. According to the method, vector data refreshing of the new model is realized in the hot updating process of the model, so that related online services can be quickly put and used.

Description

technical field [0001] The present application relates to the field of neural network model update, in particular to a model hot update processing method, and also relates to the corresponding device, equipment, non-volatile storage medium and computer program product of the method. Background technique [0002] There are various types of business services based on neural network models in the existing Internet platforms. For example, in the business services used to provide users with intelligent customer service, there are generally model services corresponding to neural network models used for semantic reasoning. In the business services that provide users with commodity classification, there are generally model services corresponding to the neural network models used for commodity classification, and with the development of neural network technology, each platform will also update the neural network model that provides reasoning functions in its business services. Networ...

AI Technical Summary

Problems solved by technology

[0003] However, when the existing platform updates the neural network model in the business service, it is often necessary to suspend the operation of the old version of the neural network model in the business service, and wait for the deployment of the new version of the neural network model to be completed before launching the new version of the neural network model. The business services mentioned above provide reasoning services, and during this process, because the business services need to be offline to wait for the deployment of the new model, users of the platform will not be able to use the business services

[0004] Secondly, although the existing platform also uses hot update to update the neural network model in the business service, after the new version of the neural network model is hot updated for the business service, the new version of the neural network model does not hold the corresponding new version. Eigenvectors, so that after the business service updates the model, it cannot provide complete services for platform users, affecting the user experience

Images