Data processing method of large language model, training method and device of decision network, equipment, storage medium and program product
By employing iterative evaluation and adaptive selection of target actions, the problem of inflexible processing flow in the generation of large language model responses is solved, thereby improving the accuracy and quality of response data.
CN122132509APending Publication Date: 2026-06-02TENCENT TECHNOLOGY (SHENZHEN) CO LTD
0 Cites 0 Cited by
Patent Information
- Application Number
- CN202610225707.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-25
- Publication Date
- 2026-06-02
AI Technical Summary
Technical Problem
In the response generation process of large language models in the existing technology, fixed retrieval strategies make it difficult to flexibly adjust the processing flow according to the complexity of the input information or the generation status, which affects the quality of the generated response data.
Method used
By iteratively evaluating the current state, adaptively selecting the target action, and using multiple iterations to process the context data to be processed for state feature encoding and feature mapping, the target action is determined until the response data is generated.
Benefits of technology
It improves the flexibility and accuracy of data processing, ensuring the quality and accuracy of response data.
✦ Generated by Eureka AI based on patent content.
Abstract
The invention provides a data processing method of a large language model, a training method and device of a decision network, equipment, a storage medium and a program product. The method comprises the steps of constructing initialized context data based on to-be-responded input information of a large language model; multiple times of iteration processing are executed, and each time of iteration processing comprises the steps that state feature coding is conducted on the context data to be processed, and state features are obtained; performing feature mapping on the state features to obtain evaluation values of multiple types of candidate actions in the action space, and determining a target action from the multiple types of candidate actions based on the evaluation values; and when the target action is to complete the generation of the response data, stopping the iterative processing, and extracting the response data of the input information generated by the large language model from the context data to be processed. According to the method and the device, the target action can be adaptively selected by iteratively evaluating the current state, and the flexibility and the response accuracy of data processing are improved.
Need to check novelty before this filing date? Find Prior Art