Text compression encoding method and device for large language model and computer device

By using a text compression encoding method for large language models, data is collected and segmented, candidate encoding sequences are generated, and numbering and encoding packaging are performed. This solves the problems of high cost, limited window, and cache invalidation when large language models process highly redundant data, and achieves a synergistic improvement in cost, efficiency, and accuracy.

CN122394564APending Publication Date: 2026-07-14BEIJING RISING NETWORK SECURITY TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
BEIJING RISING NETWORK SECURITY TECH CO LTD
Filing Date
2026-03-26
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Large language models face challenges such as high token costs, limited windows, and cache invalidation when processing highly redundant log data. Existing compression technologies suffer from lossy compression and wasted computing resources, failing to effectively reduce input costs.

Method used

By collecting input data, performing extraction and segmentation processing, generating a candidate coding sequence list, assigning numbers and packaging codes, ensuring lossless data compression and reusing historical symbols, a rigorous closed-loop process is formed to avoid redundant calculations and ambiguity.

Benefits of technology

It reduces the input cost of large language models, improves data compression efficiency and accuracy, meets the data originality requirements of different scenarios, and solves the triangular problem of high cost, limited window and cache invalidation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122394564A_ABST
    Figure CN122394564A_ABST
Patent Text Reader

Abstract

The application relates to a text compression encoding method, device and computer equipment for a large language model. The method comprises the following steps: when the data amount of the current dialogue data exceeds a preset length threshold, receiving an original text block corresponding to the current dialogue data; performing extraction and segmentation processing on the original text block to obtain old encoding data and new text data, and generating a candidate encoding sequence list of the new text data through a benefit evaluation strategy; performing number allocation processing on the new text data based on the candidate encoding sequence list and the old encoding data to obtain text number information of the original text block, and performing text encoding packaging processing on the original text block based on the text number information of the original text block to obtain compressed text encoding data. The method can improve the input cost optimization practicality effect of the large language model.
Need to check novelty before this filing date? Find Prior Art