Data processing method and apparatus, and computer device and storage medium

By conducting structural search and automatic quantification of AI models, the problems of high code invasiveness and poor universality caused by manual intervention of traditional quantization methods are solved, and the efficient deployment of AI models on terminal devices is achieved.

WO2024259853A1PCT designated stage expired Publication Date: 2024-12-26TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2023/128556
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-06-19
Filing Date
2023-10-31
Publication Date
2024-12-26

AI Technical Summary

Technical Problem

Traditional quantization methods require manual modification of AI models, resulting in high code invasiveness, slow quantization speed and lack of versatility, making it difficult to effectively deploy complex AI models on end devices with high requirements for real-time performance.

Method used

By conducting structural search on the initial model, the quantitative forgery operator is automatically selected according to the target structure type in the model structure chart for quantization, and a business model is generated to realize the automation and universality of model quantification.

Benefits of technology

It improves the speed of model quantization, reduces labor costs, realizes the versatility of model quantization, is suitable for various AI model structures, and reduces code intrusion.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2023128556_26122024_PF_FP_ABST
    Figure CN2023128556_26122024_PF_FP_ABST
Patent Text Reader

Abstract

Disclosed in the embodiments of the present application are a data processing method and apparatus, and a computer device and a storage medium, which can be applied to an artificial intelligence scenario. The method comprise: when an initial model is acquired, performing a structure search on the initial model, so as to obtain a model structure diagram of the initial model; according to the type of a target structure in the model structure diagram, quantizing the target structure by using a forgery operator for quantization, wherein the forgery operator comprises a quantization operator and a dequantization operator; and on the basis of the quantized initial model, obtaining a service model for service processing. By using the embodiments of the present application, the speed of model quantization can be increased, and universality of model quantization is achieved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Network model quantification method and device, computer equipment and storage medium

    CN113610232A

  • Model quantification method and device, equipment, storage medium and computer program product

    CN113780551A

  • Model training method and device, computer equipment and storage medium

    CN115983365A