A method for processing the interactive context of a language model, an electronic device, and a storage medium.

Abstract

This invention discloses a method, electronic device, and storage medium for processing the interactive context of a language model. The method includes: real-time monitoring of the length of the dialogue context and the access frequency of each data fragment in the session context cache; if there are data fragments with an access frequency below a threshold, they are migrated to external storage, and step transformation data containing semantic summaries, key findings, and file indexes is generated; if there are no such data fragments and the context length exceeds the limit, historical content is summarized and compressed based on preset anchor points, retaining recent content; then, the processed content is reorganized into a structured context with type labels according to a preset structure; finally, a context state snapshot associated with a version identifier is generated and stored. This method achieves adaptive optimization management of the context, significantly reduces computational load and cost, improves response speed and decision quality, and enhances the traceability of the state.

Images