Self-adaptive large-model AI application construction system based on multi-model logic orchestration

Abstract

The present invention discloses a self-adaptive large-model AI application construction system based on multi-model logic arrangement, which relates to the field of large-model AI application construction technology, including an operation module, which supports users to select required logic components from a component library by dragging and dropping, place them in a function configuration area, and define the interaction logic between various logic components through connecting lines. The self-adaptive large-model AI application construction system based on multi-model logic arrangement, through the design of an intuitive and easy-to-use operation interface, allows users to arrange AI applications through simple drag and drop operations, greatly reducing the threshold for technical use; using a set of developed automatic adaptation algorithms, it can intelligently recommend the optimal model combination according to user needs and application scenarios, reducing the workload of manual configuration; it integrates a variety of large-scale pre-trained models, covering multiple fields, and meeting the needs of different users.

Description

Technical Field

[0001] The present invention relates to the technical field of large-model AI application construction, and specifically to a self-adaptive large-model AI application construction system based on multi-model logic orchestration. Background Art

[0002] With the rapid development of artificial intelligence technology, big models have attracted widespread attention due to their powerful data processing and generalization capabilities. However, there are many types of big models on the market with different technical characteristics, which brings huge challenges to enterprises and individual users. On the one hand, users need to choose from many models to adapt to different application scenarios, and this selection process is often time-consuming and full of uncertainty; on the other hand, most of the existing big models lack good integration and compatibility, and are difficult to embed directly into users' business processes; in addition, for non-professional developers, building personalized AI applications also has a high technical threshold, because this usually requires certain programming knowledge and in-depth understanding of AI technology; therefore, the market urgently needs a solution that can simplify the AI ​​application construction process while supporting multi-model integration and automatic adaptation. Summary of the invention

[0003] In order to solve the above technical problems, the present invention is implemented through the following technical solutions: a self-adaptive large model AI application construction system based on multi-model logic arrangement, comprising:

[0004] The operation module supports users to select the required logic components from the component library by dragging and dropping, place them in the function configuration area, and define the interaction logic between various logic components through connecting lines;

[0005] The algorithm module is connected with the operation module and is used to recommend a large model combination according to the user's request information;

[0006] Integration module, which is used to integrate large AI models and provide a unified standard interface for recommendation models to ensure that they can be seamlessly connected to the system platform. At the same time, the model is adapted according to the application's interaction logic and operating environment.

[0007] A management module that supports users to upload private data sets and clean and pre-process the data in order to train or fine-tune private AI models;

[0008] The deployment module allows users to test AI applications after they are built, and supports deploying AI applications to cloud servers or local environments.

[0009] Preferably, the operation module includes:

[0010] A component library is used to record and store logic components from different sources; wherein the logic components include data input components, data processing components, model calling components, logic control components, data output components and logic supplement components;

[0011] The area unit has a model selection area, a data upload area, a function configuration area, and a preview area. Each area is designed with instruction documents and help documents to facilitate user operation;

[0012] The instruction document is used to provide guidance to help users understand and operate the various modules and functions in the self-adaptive large-model artificial intelligence application construction system based on multi-model logic arrangement; the instruction document contains an introduction to the functions and usage methods of different areas of the system, as well as an explanation of the user operation steps and parameter meanings, so that users can use the system to build artificial intelligence applications more efficiently and accurately;

[0013] The help documentation includes system functions, operation procedures, module usage and parameter settings;

[0014] The editing unit includes a script editor, which supports users to write script rules to supplement functional logic based on application requirements through the script editor, obtain logic supplement components, store the logic supplement components in the component library, and participate in the interactive logic definition between logic components.

[0015] Preferably, the process of defining the interaction logic between various logical components through connecting lines:

[0016] Based on the component library, users read the instruction documents and help documents to understand the functions and applicable scenarios of each component. Then, according to application requirements, users select the required logic components from the component library by dragging and dropping. Then, users place the selected logic components one by one in the function configuration area.

[0017] Determine the logical components for which interaction logic needs to be established. In the function configuration area, find the tool for connecting components. Click the starting logical component, hold down the left mouse button, drag the mouse pointer to the target logical component, and release the left mouse button. At this time, a connection line is established between the starting logical component and the target logical component, indicating the direction of the data flow or control flow between the two components. Repeat this step according to application requirements to establish connection relationships for different logical components and build an initial interaction framework.

[0018] Then use the preview area to make a preliminary preview of the interaction framework, and check whether the data flow or control flow is correct based on the application business process, and check whether the connection between each logical component conforms to the expected interaction logic. If not, return to the previous step and make modifications according to the operating instructions in the instruction document, including deletion, reconnection, or adjustment of the connection direction, until the interaction logic is correct and the complete interaction logic is obtained; at the same time, you can refer to the principles and FAQs about the interaction of logical components in the help document to ensure that the adjusted interaction logic can work normally when the system is running; then save the current interaction logic based on the operation process.

[0019] Preferably, the algorithm module includes:

[0020] A demand analysis unit is used to obtain user request information and extract user requirements based on the application scenarios and functional requirements in the request information, wherein the user requirements include task type, data scale, accuracy requirements, and response time;

[0021] The model library management unit is provided with a model library for maintaining the model library. The model library stores artificial intelligence large models in the fields of natural language processing, image processing, video processing, and mathematical reasoning. Each artificial intelligence large model has performance indicators.

[0022] The performance evaluation unit selects candidate models from the model library based on user needs and interaction logic, evaluates each candidate model, calculates the matching degree between it and user needs, and obtains the matching function value of each candidate model;

[0023] The decision unit is used to select the model with the highest matching degree as the recommended model according to the matching function value.

[0024] The integration module is used to unify the standard interface and seamlessly integrate the recommended artificial intelligence models based on the standard interface, covering natural language processing, image processing, video processing and mathematical reasoning.

[0025] Preferably, the management module includes:

[0026] Data upload unit, used to provide a secure data upload mechanism, support users to upload private data sets, and clean and pre-process the data;

[0027] The data interface unit is provided with a data interface for connecting with an external data source and for importing and exporting data to enhance the logical behavior of artificial intelligence applications.

[0028] Preferably, the deployment module includes:

[0029] A simulation environment unit to test the functionality of AI applications to ensure they meet expected requirements;

[0030] A deployment option unit is used to support users to choose to deploy AI applications to a cloud server or a local environment to meet different usage requirements;

[0031] The continuous optimization unit collects user feedback and application performance data after the deployment of AI applications, and based on user feedback and application performance data, trains and adjusts the model, and updates the model and optimization algorithm to improve the quality of AI applications and user experience.

[0032] Preferably, the method for using the self-adaptive large model artificial intelligence application construction system comprises the following steps:

[0033] S1: Component selection: Users select required logical components from the component library according to their own needs through the operation module to form a component set;

[0034] S2: Logic arrangement: The user places the selected component set in the function configuration area and uses connecting lines to define the interaction logic between the components;

[0035] S3: Script execution: Users write script rules in the script editor to customize the behavior of the AI ​​application;

[0036] S4: Data connection: Users connect with external data sources through the data interface unit in the management module to import and export data;

[0037] S5: Application construction and testing: After completing the above steps, a complete AI application logic architecture is constructed; then, by deploying modules, the functions of the AI ​​application are tested in the simulation environment unit to ensure that it meets the expected requirements;

[0038] S6: Application deployment: Deploy AI applications to cloud servers or local environments to run them and provide services to users, completing the entire process from application construction to online operation.

[0039] Preferably, component selection: the user selects the required logical components from the component library according to his / her own needs through the operation module to form a component set:

[0040] ;

[0041] in, is the set of component indexes selected by the user;

[0042] This is the basis for building applications. Users select appropriate components based on the functional requirements of the application, including data input components, data processing components, and model calling components, to prepare for subsequent logical orchestration.

[0043] Preferably, step S2 includes step S21: the user drags the component to the function configuration area and uses connecting lines to define the interaction logic between the components; each connecting line represents a data flow or control flow, which is represented by a directed graph G=(V,E); wherein V is a node set, i.e., a component; E is an edge set, i.e., a connecting line, and e∈E represents a data flow or a control flow;

[0044] Among them, the node set V: ;

[0045] is the set of component indexes selected by the user;

[0046] In this way, users can design the data processing flow and control logic of the application, determine the data transmission path and processing order between various components, including data flowing from input components to processing components, and then flowing to output components after being processed by the model.

[0047] Preferably, step S3 includes step S31: the user writes script rules in the script editor to customize the behavior of the artificial intelligence application; wherein the script rules include simple conditional statements, loop statements, and calls to external API codes; wherein the script is expressed in a formal language as:

[0048] ;

[0049] ;in, is the i-th script instruction;

[0050] It is an index set of script instructions written by the user;

[0051] Scripts can provide more sophisticated control over component behavior, implement some complex logical judgments, loop operations, or interact with external systems, enhance the flexibility and functionality of applications, and enable them to better adapt to specific business needs;

[0052] Data connection: Users can connect to external data sources through the data interface unit in the management module to import and export data to enhance the logical behavior of the application;

[0053] The data interface is represented by the function D(I,O), where I is the input data and O is the output data;

[0054] Among them, the calculation formula of function D(I,O) is:

[0055] ;

[0056] in, It is a set of data indexes imported or exported by the user;

[0057] This enables applications to obtain external data for processing, or output processing results to external systems, expanding the data sources and application scope of the application and improving the application's ability to interact with the external environment.

[0058] The present invention provides a self-adaptive large-model AI application construction system based on multi-model logic arrangement, which has the following beneficial effects:

[0059] This self-adaptive large-model AI application construction system based on multi-model logic orchestration has designed an intuitive and easy-to-use operation interface, allowing users to orchestrate AI applications through simple drag and drop operations, greatly reducing the threshold for technology use; using a set of automatic adaptation algorithms developed, it can intelligently recommend the optimal model combination based on user needs and application scenarios, reducing the workload of manual configuration; it integrates a variety of large-scale pre-trained models, covering natural language processing, image processing, video processing, mathematical reasoning and other fields to meet the needs of different users. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flowchart of the self-adaptive large model AI application construction system based on multi-model logic arrangement of the present invention. DETAILED DESCRIPTION

[0061] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0062] like Figure 1 As shown, the present invention provides a technical solution: a self-adaptive large model AI application construction system based on multi-model logic arrangement, including:

[0063] The operation module supports users to select the required logic components from the component library by dragging and dropping, place them in the function configuration area, and define the interaction logic between various logic components through connecting lines;

[0064] The algorithm module is connected with the operation module and is used to recommend a large model combination according to the user's request information;

[0065] Integration module, which is used to integrate large AI models and provide a unified standard interface for recommendation models to ensure that they can be seamlessly connected to the system platform. At the same time, the model is adapted according to the application's interaction logic and operating environment.

[0066] A management module that supports users to upload private data sets and clean and pre-process the data in order to train or fine-tune private AI models;

[0067] The deployment module allows users to test AI applications after they are built, and supports deploying AI applications to cloud servers or local environments.

[0068] The operation modules include:

[0069] A component library is used to record and store logic components from different sources; wherein the logic components include data input components, data processing components, model calling components, logic control components, data output components and logic supplement components;

[0070] The area unit has a model selection area, a data upload area, a function configuration area, and a preview area. Each area is designed with instruction documents and help documents to facilitate user operation;

[0071] The instruction document is used to provide guidance to help users understand and operate the various modules and functions in the self-adaptive large-model artificial intelligence application construction system based on multi-model logic arrangement; the instruction document contains an introduction to the functions and usage methods of different areas of the system, as well as an explanation of the user operation steps and parameter meanings, so that users can use the system to build artificial intelligence applications more efficiently and accurately;

[0072] The help document includes system functions, operation procedures, module uses, and parameter settings; the help document is used to provide users with detailed information and guidance resources to answer questions encountered by users in the process of using the system, help users become familiar with the specific functions and uses of each module of the system, provide detailed step-by-step instructions for the system's operation procedures, and assist users in understanding the meaning of various parameters involved in the system. In the management module part, the help document is used to explain the data format supported by the data upload unit, the method and purpose of data cleaning and preprocessing, so that users can clearly understand how to upload private data sets and make them suitable for training or fine-tuning models; provide guidance for users in the operation of deploying modules; inform users of the test purpose and operation procedures of the simulation environment unit, and help users use the unit to effectively test the various functions of artificial intelligence applications to ensure that they meet expected requirements; help users understand the system's openness and scalability; when users encounter problems in the process of using the system, it serves as a reference for users to self-check and solve problems, find information related to the problem, try to solve the problem by themselves, reduce dependence on technical support personnel, and improve usage efficiency, so that users can make full use of the system and build artificial intelligence applications that meet their own needs;

[0073] The editing unit includes a script editor, which supports users to write script rules to supplement functional logic through the script editor based on application requirements, obtain logic supplement components, store the logic supplement components in the component library, and participate in the interactive logic definition between logic components.

[0074] The process of defining the interaction logic between various logical components through connecting lines:

[0075] Based on the component library, users read the instruction documents and help documents to understand the functions and applicable scenarios of each component. Then, according to application requirements, users select the required logic components from the component library by dragging and dropping. Then, users place the selected logic components one by one in the function configuration area.

[0076] Determine the logical components for which interaction logic needs to be established. In the function configuration area, find the tool for connecting components. Click the starting logical component, hold down the left mouse button, drag the mouse pointer to the target logical component, and release the left mouse button. At this time, a connection line is established between the starting logical component and the target logical component, indicating the direction of the data flow or control flow between the two components. Repeat this step according to application requirements to establish connection relationships for different logical components and build an initial interaction framework.

[0077] Then use the preview area to make a preliminary preview of the interaction framework, and check whether the data flow or control flow is correct based on the application business process, and check whether the connection between each logical component conforms to the expected interaction logic. If not, return to the previous step and make modifications according to the operating instructions in the instruction document, including deletion, reconnection, or adjustment of the connection direction, until the interaction logic is correct and the complete interaction logic is obtained; at the same time, you can refer to the principles and FAQs about the interaction of logical components in the help document to ensure that the adjusted interaction logic can work normally when the system is running; then save the current interaction logic based on the operation process.

[0078] The algorithm modules include:

[0079] A demand analysis unit is used to obtain user request information and extract user requirements based on the application scenarios and functional requirements in the request information, wherein the user requirements include task type, data scale, accuracy requirements, and response time;

[0080] The model library management unit is provided with a model library for maintaining the model library. The model library stores artificial intelligence large models in the fields of natural language processing, image processing, video processing, and mathematical reasoning. Each artificial intelligence large model has performance indicators.

[0081] The performance evaluation unit selects candidate models from the model library based on user needs and interaction logic, evaluates each candidate model, calculates the matching degree between it and user needs, and obtains the matching function value of each candidate model;

[0082] A decision unit is used to select the model with the highest matching degree as the recommended model according to the matching degree function value;

[0083] The model library management unit maintains the model library process:

[0084] By paying attention to the research progress in the field of artificial intelligence, industry trends, and new models released by major model manufacturers, we can obtain relevant information in a timely manner, understand the characteristics, advantages, and applicable scenarios of different models, and form a potential model resource library;

[0085] Based on the target user groups, common application scenarios and business needs served by the system, potential models are preliminarily screened to evaluate whether the models meet the system integration requirements in terms of functions, performance and scalability, and to determine the candidate models that can be included in the model library; functions include natural language processing capabilities and image processing accuracy; performance includes computational complexity, memory usage and processing speed;

[0086] Negotiate with model suppliers and developers to obtain relevant resources of the selected model, including model files, algorithm codes, technical documents, and licenses, on the premise that the acquisition process is legal and compliant;

[0087] Arrange and classify the acquired models, archive them according to the model type, functional characteristics, and applicable scenarios, and establish a clear model directory structure to facilitate subsequent search and management;

[0088] According to the unified standard interface specifications of the system, configure the parameters of the stored model to ensure that the model can be seamlessly connected with other components and modules of the system; set the initial operating parameters of the model, including the default input and output formats and precision settings, and record the original default parameters of the model for subsequent comparison and recovery;

[0089] During the model operation, based on the model's accuracy, processing speed, resource consumption, combined with user feedback and changes in business needs, analyze whether the model meets the current application requirements;

[0090] When it is found that the model performance has deteriorated, new requirements have emerged, or a new version has been released, timely obtain the updated resources of the model, including the updated model files and improved algorithm codes;

[0091] Perform functional tests, performance tests, and compatibility tests on the updated model in the test environment to ensure that the updated model can work properly and does not affect the overall stability of the system and the coordinated operation of other models. Compare the model performance indicators before and after the update to evaluate the update effect.

[0092] If the test passes, the updated model will be officially deployed to the model library in the production environment to replace the old version of the model, and the model-related documents and configuration information will be updated in a timely manner to ensure that the system can correctly call the new model;

[0093] Regularly evaluate the models in the model library, and determine whether any models need to be eliminated based on the model's frequency of use, performance, compatibility with new business requirements, and maintenance costs; list models that have not been used for a long time, have outdated performance and cannot be improved through updates, are no longer applicable to current business scenarios, or have excessively high maintenance costs as models to be eliminated;

[0094] Before eliminating a model, back up the model and its related data, and record the reasons for elimination and historical usage of the model for subsequent review and auditing;

[0095] Write detailed documentation for each model in the library, including model overview, function description, input and output formats, parameter description, usage examples, performance indicators, and update records to ensure the accuracy, completeness, and timeliness of the documentation, so as to facilitate users and developers to understand and use the model;

[0096] Based on the update, upgrade or elimination of the model, timely update the corresponding document version to keep the document consistent with the actual situation of the model and ensure that the document can accurately reflect the latest status and usage of the model;

[0097] Through the setting of the model library management unit and the model library, a framework for centralized model management is provided, which facilitates the classification, storage and update of models. Model version control can be performed to ensure that a stable and reliable model version is used during the application construction process. At the same time, it is also convenient for model authorization management to ensure the legal use of model resources. Compared with the decentralized management of multiple models, the model management cost is reduced. The unified management method of the model library reduces the complexity and cost of management, and can centrally perform model maintenance work, including repairing model vulnerabilities, updating model parameters, etc., to improve management efficiency and ensure the quality and availability of the model.

[0098] Assume that the user request information is represented by a vector R, R=(r1,r2,...,r i ), where r i represents the i-th demand parameter;

[0099] Each model uses a vector M j =(m j1 ,m j2 ,...,m ji ) indicates that m ji is the i-th performance indicator of model j. The performance indicators include computational complexity, memory usage, processing speed, and accuracy.

[0100] The performance evaluation unit is used to evaluate each candidate model, calculate the matching degree between it and the user requirements, and obtain the matching function value, wherein the matching function is defined as:

[0101] ;

[0102] Among them, w i is the weight of the i-th demand parameter; g(r i ,m ji ) is the demand parameter r i And the model performance index m ji A function of the difference between

[0103] A decision unit, used to select the model with the highest matching degree as the recommended model according to the value of the matching degree function;

[0104] Assume that there are K models in the model library, then the recommended model M * The calculation formula is:

[0105] ;

[0106] Among them, the model M was finally selected j , j=1,2,...,k;

[0107] After calculating the matching degree between each model and the user's needs, the model with the highest matching degree is selected as the recommended model according to the value of the matching function. Here, M is the model finally selected, which is specifically represented as M in the model library. j , j=one of 1,2,...,k;

[0108] The integration module is used to unify the standard interface and seamlessly integrate the recommended artificial intelligence models based on the standard interface, covering natural language processing, image processing, video processing and mathematical reasoning.

[0109] The management modules include:

[0110] Data upload unit, used to provide a secure data upload mechanism, support users to upload private data sets, and clean and pre-process the data;

[0111] The data interface unit is provided with a data interface for connecting with an external data source and for importing and exporting data to enhance the logical behavior of artificial intelligence applications.

[0112] The deployment modules include:

[0113] A simulation environment unit to test the functionality of AI applications to ensure they meet expected requirements;

[0114] A deployment option unit is used to support users to choose to deploy AI applications to a cloud server or a local environment to meet different usage requirements;

[0115] The continuous optimization unit collects user feedback and application performance data after the deployment of AI applications, and adjusts the model training based on user feedback and application performance data, and updates the model and optimization algorithm to improve the quality of AI applications and user experience;

[0116] During the operation of artificial intelligence applications, user feedback and application performance data are collected and fed back to the continuous optimization unit. By training or fine-tuning the model, the performance of the model is continuously improved and the accuracy and efficiency of the model are improved. The model library provides the basic resources for this continuous optimization, allowing the model to evolve continuously with the accumulation of usage experience.

[0117] The method for using the self-adaptive large model artificial intelligence application construction system includes the following steps:

[0118] S1: Component selection: Users select required logical components from the component library according to their own needs through the operation module to form a component set;

[0119] S2: Logic arrangement: The user places the selected component set in the function configuration area and uses connecting lines to define the interaction logic between the components;

[0120] S3: Script execution: Users write script rules in the script editor to customize the behavior of the AI ​​application;

[0121] S4: Data connection: Users connect with external data sources through the data interface unit in the management module to import and export data;

[0122] S5: Application construction and testing: After completing the above steps, a complete AI application logic architecture is constructed; then, by deploying modules, the functions of the AI ​​application are tested in the simulation environment unit to ensure that it meets the expected requirements;

[0123] S6: Application deployment: Deploy AI applications to cloud servers or local environments to run them and provide services to users, completing the entire process from application construction to online operation.

[0124] Component selection: Users can select the required logical components from the component library according to their own needs through the operation module to form a component set:

[0125] ;

[0126] in, is the set of component indexes selected by the user;

[0127] This is the basis for building applications. Users select appropriate components based on the functional requirements of the application, including data input components, data processing components, and model calling components, to prepare for subsequent logical orchestration.

[0128] Step S2 includes step S21: the user drags the component to the function configuration area and uses connecting lines to define the interaction logic between the components; each connecting line represents a data flow or control flow, which is represented by a directed graph G=(V,E); wherein V is a node set, i.e., a component; E is an edge set, i.e., a connecting line, and e∈E represents a data flow or a control flow;

[0129] Among them, the node set V: ;

[0130] is the set of component indexes selected by the user;

[0131] In this way, users can design the data processing flow and control logic of the application, determine the data transmission path and processing order between various components, including data flowing from input components to processing components, and then flowing to output components after being processed by the model.

[0132] Step S3 includes step S31: the user writes script rules in the script editor to customize the behavior of the artificial intelligence application; wherein the script rules include simple conditional statements, loop statements, and calls to external API codes; wherein the script is expressed in a formal language as:

[0133] ;

[0134] ;in, is the i-th script instruction;

[0135] It is an index set of script instructions written by the user;

[0136] Scripts can provide more sophisticated control over component behavior, implement some complex logical judgments, loop operations, or interact with external systems, enhance the flexibility and functionality of applications, and enable them to better adapt to specific business needs;

[0137] Data connection: Users can connect to external data sources through the data interface unit in the management module to import and export data to enhance the logical behavior of the application;

[0138] The data interface is represented by the function D(I,O), where I is the input data and O is the output data;

[0139] Among them, the calculation formula of function D(I,O) is:

[0140] ;

[0141] in, It is a set of data indexes imported or exported by the user;

[0142] This enables applications to obtain external data for processing, or output processing results to external systems, expanding the data sources and application scope of the application and improving the application's ability to interact with the external environment.

[0143] It should be further explained that in the specific implementation process, the construction process of artificial intelligence application is as follows:

[0144] Business demand research: Work closely with relevant business departments or users to fully understand the application's usage scenarios, business processes, and the problems that are expected to be solved;

[0145] Determine functional and performance goals: Based on business needs, determine the specific functions that artificial intelligence applications should have, such as data collection and preprocessing, model training and prediction, and result visualization; at the same time, clarify performance indicators, such as the model accuracy must reach more than 95% and the processing time for a single medical image must not exceed 10 seconds, to provide a clear direction for subsequent design and development;

[0146] Data collection: Collect relevant data from various data sources according to application requirements;

[0147] Data cleaning and preprocessing: Clean the collected data and process missing values, outliers and noise data;

[0148] Logic arrangement design stage: Component selection and combination: Use the operation module to select the required logic components from the component library, including data input components, data processing components, model call components, logic control components, and data output components; Place these components reasonably in the design area according to the application logic, and build the interaction logic between components through connecting lines to form the basic framework of the application; Script writing and customization: Use the script editor to write simple scripts or rules to further customize the behavior of the artificial intelligence application;

[0149] Model selection and adaptation stage: Algorithm module requirement analysis: The algorithm module analyzes the requirement parameters according to the application scenarios and functional requirements specified by the user to form a requirement vector, including task type, data scale, accuracy requirements, and response time;

[0150] Model library evaluation and selection: The model library maintained by the model library management unit contains a variety of large artificial intelligence models;

[0151] The performance evaluation unit calculates the matching degree between each candidate model and the user's requirements, and the decision unit selects the model with the highest matching degree as the recommended model based on the matching degree function;

[0152] Model integration and adaptation optimization: The integration module provides a unified standard interface for the selected model to ensure that it can be seamlessly connected to the system platform; at the same time, the model is adapted and optimized according to the application's logical arrangement and operating environment;

[0153] Testing and verification phase: Simulation environment testing: The simulation environment unit in the deployment module is used to test the various functions of the AI ​​application; input the test data set, simulate the actual operation scenario, check whether the data is processed correctly between various components and models, and whether the output results meet expectations;

[0154] Performance evaluation and tuning: Collect performance data during the test, such as model prediction accuracy, processing time, and resource usage, and compare and analyze them with the predetermined performance targets. If performance is found to be substandard or functionally abnormal, perform tuning by adjusting logic arrangement, optimizing model parameters, or replacing components.

[0155] Deployment and launch phase: Deployment decision and configuration: According to user needs and the actual operating environment, choose to deploy the AI ​​application to the cloud server or local environment; during the deployment process, perform corresponding configurations, such as server resource allocation, network settings, and security configuration; Online operation and monitoring: The AI ​​application is officially launched online and its operating status is continuously monitored, including system performance, data quality, and model effects; a monitoring indicator system and alarm mechanism are established to promptly discover and handle possible problems;

[0156] Continuous optimization stage: User feedback collection: Collect user feedback after using AI applications through various channels, such as user satisfaction surveys, in-app feedback functions, and customer service feedback; Performance data monitoring and analysis: Continuously collect performance data during the application operation, including model prediction results, user operation behaviors, and system resource usage information; Use data analysis technology to deeply explore potential problems and optimization space in the data;

[0157] Model update and algorithm optimization: Update the model regularly based on user feedback and performance data analysis results, such as retraining the model to adapt to new data distribution or changes in business needs; at the same time, optimize the matching calculation function and model selection strategy algorithm in the algorithm module to improve the accuracy of model recommendations and the overall performance of the system;

[0158] Iteration and expansion of application functions: Based on user needs and technology development trends, the functions of artificial intelligence applications are iteratively upgraded or new functional modules are expanded.

[0159] It should be further explained that, in the specific implementation process, by providing a graphical user interface and drag-and-drop component selection, even users without programming experience can quickly get started and build their own AI applications; it reduces dependence on professional development teams, allowing small and medium-sized enterprises and individual developers to enjoy advanced AI technology; the use of automatic adaptation algorithms can intelligently recommend the most suitable model combination according to the application scenario, greatly saving the time of manual selection and configuration, allowing users to quickly build a fully functional AI application prototype.

[0160] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. An adaptive large-model AI application construction system based on multi-model logic arrangement, characterized by: include: The operation module supports users to select the required logic components from the component library by dragging and dropping, place them in the function configuration area, and define the interaction logic between various logic components through connecting lines; The algorithm module is connected with the operation module and is used to recommend a large model combination according to the user's request information; Integration module, which is used to integrate large AI models and provide a unified standard interface for recommendation models to ensure that they can be seamlessly connected to the system platform. At the same time, the model is adapted according to the application's interaction logic and operating environment. Management module, used to support users to upload private data sets and clean and pre-process the data; The deployment module allows users to test AI applications after they are built and supports deploying AI applications to cloud servers or local environments; The algorithm module includes: A demand analysis unit is used to obtain user request information and extract user requirements based on the application scenarios and functional requirements in the request information, wherein the user requirements include task type, data scale, accuracy requirements, and response time; The model library management unit is provided with a model library for maintaining the model library. The model library stores artificial intelligence large models in the fields of natural language processing, image processing, video processing, and mathematical reasoning. Each artificial intelligence large model has performance indicators. The performance evaluation unit selects candidate models from the model library based on user needs and interaction logic, evaluates each candidate model, calculates the matching degree between it and user needs, and obtains the matching function value of each candidate model; The decision unit is used to select the model with the highest matching degree as the recommended model according to the matching function value.

2. According to claim 1, the self-adaptive large model AI application construction system based on multi-model logic arrangement is characterized by: The operation module includes: A component library is used to record and store logic components from different sources; wherein the logic components include data input components, data processing components, model calling components, logic control components, data output components and logic supplement components; The regional unit has a model selection area, a data upload area, a function configuration area, and a preview area. Each area is designed with instruction documents and help documents. The instruction document contains the introduction of functions and usage of different areas of the system, as well as explanations of user operation steps and parameter meanings; The help documentation includes system functions, operation procedures, module usage and parameter settings; The editing unit includes a script editor, which supports users to write script rules to supplement functional logic based on application requirements through the script editor, obtain logic supplement components, store the logic supplement components in the component library, and participate in the interactive logic definition between logic components.

3. The self-adaptive large model AI application construction system based on multi-model logic arrangement according to claim 2 is characterized by: The process of defining the interaction logic between various logical components through connecting lines: Based on the component library, users read the instruction documents and help documents to understand the functions and applicable scenarios of each component. Then, according to application requirements, users select the required logic components from the component library by dragging and dropping. Then, users place the selected logic components one by one in the function configuration area. Determine the logical components for which interaction logic needs to be established. In the function configuration area, find the tool for connecting components. Click the starting logical component, hold down the left mouse button, drag the mouse pointer to the target logical component, and release the left mouse button. At this time, a connection line is established between the starting logical component and the target logical component, indicating the direction of the data flow or control flow between the two components. Repeat this step according to application requirements to establish connection relationships for different logical components and build an initial interaction framework. Then use the preview area to make a preliminary preview of the interaction framework, and check whether the data flow or control flow is correct based on the application business process, and check whether the connection between each logical component conforms to the expected interaction logic. If not, return to the previous step and make modifications according to the operating instructions in the instruction document until the interaction logic is correct and the complete interaction logic is obtained.

4. The self-adaptive large model AI application construction system based on multi-model logic arrangement according to claim 3 is characterized by: The management modules include: The data upload unit is used to provide a data upload mechanism, support users to upload private data sets, and clean and pre-process the data; The data interface unit is provided with a data interface for connecting with an external data source and for importing and exporting data.

5. The self-adaptive large model AI application construction system based on multi-model logic arrangement according to claim 4 is characterized in that: The deployment module includes: A simulation environment unit for testing the functionality of AI applications; A deployment option unit is used to support users to choose whether to deploy the AI ​​application to a cloud server or a local environment; The continuous optimization unit collects user feedback and application performance data after the deployment of artificial intelligence applications, and based on user feedback and application performance data, it trains and adjusts the model, and updates the model and optimization algorithm.

6. The self-adaptive large model AI application construction system based on multi-model logic arrangement according to claim 5 is characterized by: The method for using the self-adaptive large model artificial intelligence application construction system includes the following steps: S1: Component selection: Users select required logical components from the component library according to their own needs through the operation module to form a component set; S2: Logic arrangement: The user places the selected component set in the function configuration area and uses connecting lines to define the interaction logic between the components; S3: Script execution: Users write script rules in the script editor to customize the behavior of the AI ​​application; S4: Data connection: Users connect with external data sources through the data interface unit in the management module to import and export data; S5: Application construction and testing: After completing steps S1 to S4 above, a complete AI application logic architecture is constructed; Then, by deploying the module, the functionality of the AI ​​application is tested in a simulated environment unit to ensure that it meets the expected requirements; S6: Application deployment: Deploy AI applications to cloud servers or local environments to run them and provide services to users.

7. The self-adaptive large model AI application construction system based on multi-model logic arrangement according to claim 6 is characterized by: Component selection: Users can select the required logical components from the component library according to their own needs through the operation module to form a component set: ;in, is the set of component indexes selected by the user.

8. The self-adaptive large model AI application construction system based on multi-model logic arrangement according to claim 7 is characterized by: Step S2 includes step S21: the user drags the component to the function configuration area and uses connecting lines to define the interaction logic between the components; each connecting line represents a data flow or control flow, which is represented by a directed graph G=(V,E); wherein V is a node set, i.e., a component; E is an edge set, i.e., a connecting line, and e∈E represents a data flow or a control flow; Among them, the node set V: ; is the component index set selected by the user.

9. The self-adaptive large model AI application construction system based on multi-model logic arrangement according to claim 8 is characterized by: Step S3 includes step S31: the user writes script rules in the script editor to customize the behavior of the artificial intelligence application; wherein the script rules include simple conditional statements, loop statements, and calls to external API codes; wherein the script is expressed in a formal language as: ； ; Among them, is the i-th script instruction; It is an index set of script instructions written by users.

Images