CN-121996283-A - Prompt word version management system, method, medium, terminal and program product

CN121996283ACN 121996283 ACN121996283 ACN 121996283ACN-121996283-A

Abstract

The application provides a prompt word version management system, a method, a medium, a terminal and a program product, wherein the system comprises a prompt word creation module, a prompt word generation module and a prompt word generation module, wherein the prompt word creation module creates a prompt word and distributes an initial version number for the created prompt word; the interactive test module carries out interactive test on the prompt words based on the LLM Playground platform and carries out iterative optimization on the tested prompt words, the version update module updates the version number of the prompt words subjected to iterative optimization according to the change category and marks the prompt words with specific version numbers, the evaluation module evaluates the optimized prompt words according to the preset evaluation indexes to obtain the prompt words meeting the evaluation indexes, and the issuing and calling module issues the prompt words meeting the evaluation indexes and calls the corresponding prompt words according to the version numbers or the marks based on the integrated software development kit. The application can realize the full-flow management automation of generating, optimizing, evaluating, publishing and monitoring the prompt words, and greatly improves the management efficiency of the prompt words.

Inventors

CAI MEIJIE
Deng Chengjing

Assignees

北京点富科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251215

Claims (10)

1. A prompt word versioning management system, comprising: the prompting word creating module is used for acquiring a preset prompting word template, assigning a value to variables in the acquired prompting word template to create a prompting word, and distributing an initial version number to the created prompting word; the interactive test module is used for carrying out interactive test on the prompt words of the initial version number based on the LLM Playground platform and carrying out iterative optimization on the prompt words after the test; the version updating module is used for updating the version number of the optimized prompting word according to the change category and labeling the prompting word with the specific version number; The evaluation module is used for evaluating the optimized prompt words according to preset evaluation indexes so as to obtain the prompt words meeting the evaluation indexes; and the issuing and calling module is used for issuing the prompt words meeting the evaluation indexes and calling the corresponding prompt words according to the version number or the label based on the integrated software development kit.
2. The system of claim 1, wherein the version numbers include a major version number, a minor version number, and a patch version number, the major version number being used for updating when the semantics of the hint word change significantly, the minor version number being used for updating when the semantics of the hint word change slightly, and the patch version number being used for updating when the text of the hint word changes.
3. The alert word versioning system of claim 1, wherein the tag is an environmental tag, the environmental tag comprising production, testing, development.
4. The prompt word version management system of claim 1, further comprising an intelligent tuning module, disposed after the interactive test module, for optimizing the prompt word of the current version based on the large language model according to the preset meta prompt word to output an optimization result.
5. The prompt word versioning management system of claim 4, further comprising a monitoring module configured to monitor usage and effects of the prompt word, and to feed back monitoring data to the intelligent tuning module.
6. The system of claim 1, further comprising a hint word library module for storing hint words for each version.
7. The prompt word versioning management method is characterized by comprising the following steps of: Acquiring a preset prompting word template, assigning a value to a variable in the acquired prompting word template to create a prompting word, and assigning an initial version number to the created prompting word; Performing interactive test on the prompt words of the initial version number based on LLM Playground platform, and performing iterative optimization on the tested prompt words; Updating the version number of the optimized prompting word for each iteration according to the change category, and labeling the prompting word with a specific version number; Evaluating the optimized prompting words according to preset evaluation indexes to obtain prompting words meeting the evaluation indexes; And issuing the prompt words meeting the evaluation indexes, and calling the corresponding prompt words according to the version number or the label based on the integrated software development kit.
8. A computer readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the method as claimed in claim 7.
9. A computer program product comprising computer program code means for causing a computer to carry out the method as claimed in claim 7 when said computer program code means are run on the computer.
10. An electronic terminal comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to implement the method as claimed in claim 7.

Description

Prompt word version management system, method, medium, terminal and program product Technical Field The application relates to the technical field of artificial intelligence, in particular to a prompt word version management system, a method, a medium, a terminal and a program product. Background In recent years, with the comprehensive popularization of large language model technology and rapid enhancement of multi-modal capability, word engineering has been suggested to develop from initial experience skills to core engineering subjects in the artificial intelligence field. The prompting words are not simple input text, but core assets for defining AI behaviors, controlling output quality and realizing safety compliance, and the development mode of the prompting words is changed from ' tuning single prompting words ' to constructing a large-scale, standardized and automatic prompting platform '. However, the existing prompt word engineering technology still faces a plurality of bottlenecks in practical application. Firstly, with the increase of the number of prompt words and frequent iteration, the problems of history high-quality version loss, no record of modification, inconsistency among versions and the like generally occur, version backtracking capability is lacked, version management is seriously disordered, unstable model output quality is caused, and uncertainty is filled in an optimization process. Secondly, the prior art relies on manual testing of single scene, lacks coverage of edge condition and multi-module linkage, and lacks testing and verification mechanism. Finally, the same prompting words can generate different results under different conditions, the version quality is difficult to objectively compare, and complex dependency relations exist among a plurality of prompting words in a complex system, and the influence of modification is difficult to separate and evaluate, so that the evaluation effect of the prompting words is greatly uncertain. Therefore, there is a need to provide a prompt word versioning management system, method, medium, terminal and program product, so as to solve the above-mentioned problems in the prior art. Disclosure of Invention In view of the above drawbacks of the prior art, an object of the present application is to provide a prompt word versioning management system, method, medium, terminal and program product, which are used for solving the technical problems of serious confusion, missing test and verification mechanisms and great uncertainty of evaluation effect of the prompt word in the prior art version management. In order to achieve the above object and other related objects, a first aspect of the present application provides a prompt word versioning management system, which includes a prompt word creation module configured to obtain a preset prompt word template, assign a value to a variable in the obtained prompt word template to create a prompt word, and assign an initial version number to the created prompt word, an interactive test module configured to perform an interactive test on the prompt word with the initial version number based on a LLM Playground platform, and perform iterative optimization on the tested prompt word, a version update module configured to update a version number of the optimized prompt word for each iteration according to a change category, and tag the prompt word with a specific version number, and an evaluation module configured to evaluate the optimized prompt word according to a preset evaluation index to obtain a prompt word meeting the evaluation index, and issue and call the prompt word meeting the evaluation index, and call the corresponding prompt word according to an integrated software development kit and a version number or a tag. In some embodiments of the first aspect of the present application, the version number includes a major version number, a minor version number, and a patch version number, where the major version number is used for updating when the semantics of the hint word change significantly, the minor version number is used for updating when the semantics of the hint word change slightly, and the patch version number is used for updating when the text of the hint word changes. In some embodiments of the first aspect of the present application, the tag is an environmental tag, the environmental tag including production, testing, development. In some embodiments of the first aspect of the present application, the system further includes an intelligent tuning module, disposed after the interactive test module, configured to optimize the current version of the alert word according to the preset meta alert word based on the large language model, so as to output an optimization result. In some embodiments of the first aspect of the present application, the system further includes a monitoring module, configured to monitor usage and effects of the alert words, and feed back monitoring data to the intell