Method and system for intelligent computing center cloud platform to realize agent self-evolution through computing power

Abstract

The invention provides a method and system for an intelligent computing center cloud platform to realize agent self-evolution through computing power, and relates to the technical field of intelligent computing centers, intelligent computing centers and computing power infrastructures, the method is executed by an embedded agent in a self-evolution agent, and the method comprises the following steps: S1, determining a function module needing to be developed and determining a development demand, the self-evolution intelligent agent is a target intelligent agent embedded with the development intelligent agent; s2, determining a design scheme of a functional module according to development requirements; s3, generating a code of the function module according to the development requirement and the design scheme; s4, debugging codes of the function modules; when the development requirement is not met, checking the design scheme and the code of the function module, returning to the step S2 if the design scheme needs to be modified, and returning to the step S3 if the code needs to be modified until the development requirement is met or the debugging frequency exceeds a preset threshold value; s5, feeding back development result information to the user; and S6, an intelligent agent online function module is embedded.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent computing centers, intelligent computing centers, and computing power infrastructure, and particularly relates to a method and system for an intelligent computing center cloud platform to achieve the self-evolution of intelligent agents through computing power. Background Art

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "intelligent computing centers" have emerged as the times require.

[0003] An "intelligent computing center" refers to a facility that uses large-scale heterogeneous computing power resources, including general computing power and intelligent computing power, to mainly provide the required computing power, data, and algorithms for artificial intelligence applications (such as scenarios like artificial intelligence deep learning model development, model training, and model inference). An intelligent computing center encompasses facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0004] An "intelligent computing center" includes but is not limited to an "intelligent computing center".

[0005] An "intelligent computing center", that is, an artificial intelligence computing center, is a type of computing power infrastructure based on artificial intelligence theory, adopting an artificial intelligence computing architecture, and providing computing power services, data services, and algorithm services required for artificial intelligence applications.

[0006] "Computing power" is the core of "intelligent computing centers" and "intelligent computing centers", which is the ability of computer devices or computing / data centers to process information, the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement, the computing ability to achieve the output of target results by processing information data, a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, and mainly provides services to society through computing power infrastructure.

[0007] An "intelligent agent" (Agent) refers to an agent that can perceive the environment and take actions to achieve specific goals. It can be software, hardware, or a system, with autonomy, adaptability, and interaction capabilities. An intelligent agent perceives changes in the environment (such as through sensors or data input), makes judgments and decisions based on the knowledge and algorithms learned by itself, and then executes actions to affect the environment or achieve a predetermined goal. Intelligent agents are widely used in the field of artificial intelligence, commonly found in automation systems, robots, virtual assistants, and game characters, and their core lies in the ability to learn independently and evolve continuously to better complete tasks and adapt to complex environments.

[0008] In the prior art, intelligent agents often can only perform specific tasks. If they need to upgrade, expand, or optimize their own functions (or capabilities), usually the required function modules need to be developed through manual development. However, the development cost of manually developing the function modules of intelligent agents is high and the efficiency is low. Summary of the Invention

[0009] The present invention provides a method and system for realizing the self - evolution of an intelligent agent through computing power in an intelligent computing center cloud platform, which is used to solve the problems of high cost and low efficiency in the existing artificial development of functional modules of intelligent agents.

[0010] In order to solve the above - mentioned technical problems, the present invention is implemented as follows: In a first aspect, the present invention provides a method for realizing the self - evolution of an intelligent agent through computing power in an intelligent computing center cloud platform, including: Step S1: The embedded intelligent agent in the self - evolving intelligent agent determines that a target functional module needs to be developed for the self - evolving intelligent agent and determines the development requirements of the target functional module. The self - evolving intelligent agent is a target intelligent agent embedded with a development intelligent agent, and the embedded development intelligent agent is the embedded intelligent agent; Step S2: The embedded intelligent agent determines the design scheme of the target functional module according to the development requirements and in combination with the large - language model method; Step S3: The embedded intelligent agent generates the code of the target functional module according to the development requirements and the design scheme and in combination with the large - language model method; Step S4: The embedded intelligent agent deploys the running environment of the target functional module in combination with the large - language model method and debugs the code of the target functional module based on the running environment; when the debugging result indicates that the target functional module does not meet the development requirements, the embedded intelligent agent checks the design scheme and code of the target functional module in combination with the large - language model method. When the design scheme needs to be modified, return to Step S2, and when the code needs to be modified, return to Step S3 until the debugging result indicates that the target functional module meets the development requirements or the number of debugging times exceeds a preset threshold; Step S5: The embedded intelligent agent feeds back the development result information of the target functional module to the user through the interaction interface of the self - evolving intelligent agent; Step S6: The embedded intelligent agent puts the target functional module on line.

[0011] Optionally, before Step S1, it further includes: Step S01: The development intelligent agent receives the task information given by the user and embedded into the target intelligent agent through the interaction interface of the development intelligent agent; Step S02: The development intelligent agent automatically identifies the relevant information of the target intelligent agent and embeds itself into the target intelligent agent according to the relevant information of the target intelligent agent to form the self - evolving intelligent agent.

[0012] Optionally, Step S02 includes: Step S021: The development agent combines the large language model method to read the documents and code of the target agent, and determines the relevant information of the target agent according to the reading results. The relevant information includes at least one of the following: architecture, interface, dependent model, extensible functional module, and functional module already possessed by the target agent.

[0013] Optionally, step S1 includes: Step S11: The embedding agent determines the target functional module to be developed for the self-evolving agent according to the user's requirements for the function upgrade, expansion or optimization of the self-evolving agent, and determines the development requirements of the target functional module; or, the embedding agent reflects on the current task or historical task of the self-evolving agent, and determines the target functional module to be developed for the self-evolving agent according to the reflection results, and determines the development requirements of the target functional module.

[0014] Optionally, step S1 includes: Step S12: The embedding agent compares the development requirements with the historical development requirements in the database, and determines whether to continue developing the target functional module according to the comparison results. The database records the historical development requirements and the status of the historical development requirements. The status includes at least one of the following: collected, confirmed, developed, launched, abandoned.

[0015] Optionally, the target functional module includes at least one of the following: reflection system, memory system, tool function, thinking framework, requirement module, and learning system.

[0016] In a second aspect, the present invention provides a system for realizing the self-evolution of an agent through computing power in an intelligent computing center cloud platform, including a self-evolving agent, where the self-evolving agent is a target agent embedded with a development agent, and the embedded development agent is an embedding agent; The embedding agent includes: A requirement confirmation module, configured to determine the target functional module to be developed for the self-evolving agent, and determine the development requirements of the target functional module; A design module, configured to determine the design scheme of the target functional module according to the development requirements, in combination with the large language model method; An encoding module, configured to generate the code of the target functional module according to the development requirements and the design scheme, in combination with the large language model method; A debugging module, which is used to deploy the operating environment of the target functional module in combination with the large language model method, and debug the code of the target functional module based on the operating environment; when the debugging result indicates that the target functional module does not meet the development requirements, check the design scheme and code of the target functional module in combination with the large language model method. When the design scheme needs to be modified, trigger the design module to continue working. When the code needs to be modified, trigger the coding module to continue working until the debugging result indicates that the target functional module meets the development requirements or the number of debugging times exceeds the preset threshold; A feedback module, which is used to feedback the development result information of the target functional module to the user through the interaction interface of the self-evolving agent; A go-live module, which is used to go live the target functional module.

[0017] Optionally, the system further includes: A development agent, which is used to receive the task information embedded in the target agent given by the user through the interaction interface of the development agent; automatically identify the relevant information of the target agent, and embed itself into the target agent according to the relevant information of the target agent to form the self-evolving agent.

[0018] In a third aspect, the present invention provides an electronic device, including: a processor, a memory, and a program stored on the memory and executable on the processor. When the program is executed by the processor, it implements the steps of the method for realizing the self-evolution of the agent by the computing power of the intelligent computing center cloud platform as described in the first aspect above.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the method for realizing the self-evolution of the agent by the computing power of the intelligent computing center cloud platform as described in the first aspect above.

[0020] In a fifth aspect, the present invention provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the steps of the method for realizing the self-evolution of the agent by the computing power of the intelligent computing center cloud platform as described in the first aspect above.

[0021] In the present invention, by embedding a development agent in a target agent, a self-evolving agent with self-evolution ability is formed. The development agent embedded in the self-evolving agent can confirm whether to develop a target function module for the self-evolving agent, determine the development requirements of the target function module, automatically complete the development, debugging, and online deployment of the target function module, replace the process of manually developing the target function module, realize the self-evolution of the agent, thereby effectively saving development costs. And the intelligent computing center cloud platform can provide sufficient computing power resources for the self-evolution of the agent, greatly improving the efficiency of the self-evolution of the development agent. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings: Figure 1 FIG. is a schematic flow chart of a method for realizing the self-evolution of an agent by the intelligent computing center cloud platform of the present invention through computing power; Figure 2 FIG. is a schematic structural diagram of a system for realizing the self-evolution of an agent by the intelligent computing center cloud platform of the present invention through computing power; Figure 3 FIG. is a schematic structural diagram of an electronic device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] The technical solutions in the present invention will be clearly and completely described below with reference to the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0024] First, the technical terms related to the present invention will be briefly described below.

[0025] The "computing power" referred to in the present invention means: the ability of a computer device or a computing / data center to process information, the ability of computer hardware and software to cooperate to execute a certain computing requirement, the computing ability to realize the output of a target result by processing information data, a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, and mainly providing services to society through computing power infrastructure.

[0026] The "Computational Power (CP)" described in the present invention refers to: the ability of a data center server to process data and output results, which is a comprehensive indicator for measuring the computing power of a data center and includes general computing power, supercomputing power, and intelligent computing power. The commonly used measurement unit is the number of floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS). The larger the value, the stronger the comprehensive computing power. It is estimated that 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs or 2 million mainstream laptops. The calculation formula is: CP = CP 通用 + CP 智能 + CP 超级 。

[0027] The "Network Power (NP)" described in the present invention refers to: the performance of the data transmission ability of computing power facilities, which is a comprehensive ability including network architecture, network bandwidth, transmission delay, intelligent management and scheduling, etc., and involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling ability.

[0028] The "Storage Power (SP)" described in the present invention refers to: the comprehensive ability of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon, which is a comprehensive indicator for measuring the data storage ability of a data center and includes external storage devices such as storage arrays and server internal storage devices. The commonly used measurement unit for storage capacity is exabyte (EB, 1 EB = 2^60 bytes), the commonly used measurement unit for performance is the number of read / write operations per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB), and the disaster recovery ratio is an important manifestation of security and reliability.

[0029] The "computing power infrastructure" described in the present invention refers to: a new type of information infrastructure that integrates information computing power, network carrying power, and data storage power, and can realize the centralized computing, storage, transmission, and application of information.

[0030] The "new type of information infrastructure" described in the present invention mainly includes network infrastructures such as 5G networks, fiber broadband networks, backbone networks, international communication networks, and satellite Internet, computing power infrastructures such as data centers, general computing power centers, intelligent computing centers, and supercomputing centers, and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0031] The "computing power" described in the present invention includes: general computing power, intelligent computing power, and supercomputing power.

[0032] The "general computing power" described in the present invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0033] The "intelligent computing power" described in the present invention refers to the computing platform that is scaled and deployed for various artificial intelligence innovation applications based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit), such as natural language processing, machine vision, and so on.

[0034] The "super computing power" described in the present invention mainly refers to the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and processes extremely complex or data-intensive problems through a dedicated operating system. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, gene analysis, etc.

[0035] The "intelligent computing center" described in the present invention refers to a facility that mainly provides the required computing power, data, and algorithms for artificial intelligence applications (such as scenarios like artificial intelligence deep learning model development, model training, and model inference) by using large-scale heterogeneous computing power resources, including general computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enabling.

[0036] The "intelligent computing center cloud platform" described in the present invention refers to a cloud computing platform that comprehensively serves based on the hardware resources and software resources of the intelligent computing center.

[0037] The "intelligent computing center" described in the present invention includes, but is not limited to, the "intelligent computing center".

[0038] The "intelligent computing center" described in the present invention, that is, the artificial intelligence computing center, is a type of computing power infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications based on artificial intelligence theory and using an artificial intelligence computing architecture.

[0039] The "computing power center" described in the present invention refers to a facility mainly composed of infrastructure such as wind, fire, water, and electricity and IT software and hardware devices, and having computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0040] The "supercomputing center" described in the present invention refers to: a supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters, capable of providing functions such as large-scale computing, storage, and network services, and is widely used in application scenarios such as aerospace, national defense, oil exploration, climate modeling, and genome sequencing.

[0041] The "computing power resources" described in the present invention refers to: technologies and facilities with information computing, transmission, storage, and application capabilities required for the development of the digital society, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and support and guarantee resources such as wind, fire, water, and electricity.

[0042] The "large language model" described in the present invention refers to a large language model (LLM), which is a language model with a relatively large number of parameters, aiming to understand and generate human language, trained with a large amount of text data, and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc.

[0043] The "agent" described in the present invention refers to an entity that can perceive the environment and take actions to achieve specific goals, and is an important part of the cloud platform of the intelligent computing center. It can be software, hardware, or a system, and has autonomy, adaptability, and interaction capabilities. The agent perceives changes in the environment (such as through sensors or data input), makes judgments and decisions based on the knowledge and algorithms it has learned, and then executes actions to affect the environment or achieve a predetermined goal. Agents are widely used in the field of artificial intelligence, commonly found in automation systems, robots, virtual assistants, and game characters, and their core lies in the ability to learn independently and evolve continuously to better complete tasks and adapt to complex environments.

[0044] The "self-evolution of the agent" described in the present invention refers to the ability of the agent to develop the required functional modules for itself. Self-evolution can also be called self-upgrade or self-expansion or self-optimization.

[0045] The "reflection system" described in the present invention refers to the ability of the agent to review, evaluate, correct, and summarize its own behavior feedback, thinking, or decision-making, aiming to discover and correct errors in a timely manner, summarize experience for future applications, and thus continuously improve the ability level of task processing.

[0046] The "self-evolving agent" described in the present invention refers to an agent embedded with a developing agent, which can use the embedded developing agent to develop the required functional modules for itself and has the ability of self-evolution (or self-upgrade or self-expansion or self-optimization).

[0047] The "development agent" described in the present invention refers to an agent formed by integrating multiple manufacturing agents with different capabilities, modularizing and standardizing these capabilities, which can be reused and embedded. The development agent can be embedded into any target agent, enabling the target agent to have the ability of self-evolution.

[0048] The "target agent" described in the present invention refers to an agent that has not embedded the development agent and is used to complete a specified task.

[0049] The "embedded agent" described in the present invention refers to the development agent embedded into the target agent. The embedded agent and the target agent form a self-evolving agent.

[0050] The "memory system" described in the present invention refers to a functional module in the agent, which is mainly used to perform the following functions: realizing the memory function of the agent, including recording the following information: historical dialogue information, decision-making process information, execution feedback information, generated design document or code information, as well as existing materials and learned and summarized knowledge; extracting relevant memories or knowledge when the agent executes tasks to assist in decision-making and complete tasks.

[0051] The "tool function" described in the present invention can also be called "tool method", which is an executable function method called by the agent to complete a specific task; it is an executable function method selected and called by the agent according to the task context through intelligent decision-making; it aims to use the large language model to intelligently construct parameters and dynamically call to complete the subtask goal.

[0052] The "thinking framework" described in the present invention refers to the logical structure and method system that supports the agent to perform task processing, reasoning, decision-making and interaction. It provides a systematic thinking mode for the AI agent, enabling it to efficiently complete complex tasks. An agent based on the large language model (LLM) usually requires a clear thinking framework to integrate model capabilities, external tools and data sources to achieve specific functions.

[0053] The "requirement module" described in the present invention can also be called the requirement confirmation module or requirement confirmation function. The agent needs to interact with the user through the requirement module to determine the user's task requirements, and then execute the task based on the determined task requirements.

[0054] The "combined large language model method" described in the present invention refers to organizing input information and designing precise prompt words, calling the large language model to generate various content forms, including thinking, replying, decision-making, tool selection, code and documents, etc. This method supports the intelligent agent to perform tasks such as reasoning, design, dialogue with users, generate solution design, and code generation, and combines with the code framework to realize the execution of instructions such as tool method calling and code program execution, so as to complete specific tasks such as consulting documents, searching for information, file processing, and script running.

[0055] To solve the problem of high cost and low efficiency of the existing artificially developed intelligent agent function modules, please refer to Figure 1 The present invention provides a method for realizing self-evolution of intelligent agents through computing power on an intelligent computing center cloud platform, the method comprising: Step S1: an embedded agent embedded in a self-evolving agent determines that a target functional module needs to be developed for the self-evolving agent, and determines the development requirements of the target functional module, wherein the self-evolving agent is a target agent embedded with a development agent, and the embedded development agent is the embedded agent; The self-evolving agent in the present invention can be an agent used to complete a specified task, such as a mathematical processing agent, a data analysis agent, etc.

[0056] The embedded agent in the present invention has the ability to develop functional modules for the self-evolving agent by combining the large language model method.

[0057] The embedded agent in the present invention may be one agent or may include multiple agents, wherein the multiple agents form a multi-agent collaborative system, each of the agents has specific capabilities, and the multiple agents may collaborate to complete the task of developing functional modules for the self-evolving agent.

[0058] In the present invention, the development demand may be to develop a new functional module for the self-evolving intelligent body, or to upgrade the existing functional modules of the self-evolving intelligent body.

[0059] In the present invention, optionally, the target function module includes at least one of the following: a reflection system, a memory system, a tool function, a thinking framework, a demand module and a learning system.

[0060] In the present invention, optionally, different functional modules have different corresponding development requirements, and the common development requirements in different functional modules may include at least one of the following: test cases, acceptance criteria, output requirements, technical requirements, performance requirements, security and privacy, etc.

[0061] The following is an explanation of each of the above development requirements.

[0062] A test case refers to an example used to test whether a target functional module meets the acceptance criteria.

[0063] The acceptance criteria are the basis for evaluating whether a target functional module meets the development requirements.

[0064] The output requirements refer to the requirements for the content or format of the output of the target functional module, which are convenient for understanding and analysis.

[0065] The technical requirements refer to the specific requirements for the technical implementation of the target functional module, such as algorithm support, computing power requirements, compatible development environments, etc. For example, whether the target functional module needs to support multi-modal data processing (such as text, images, audio, etc.), whether it needs to be integrated with existing systems, etc. The clarification of technical requirements helps to ensure the feasibility and applicability of the target functional module.

[0066] The performance requirements refer to the specific requirements put forward for the performance of the target functional module, including inference speed, accuracy, resource consumption, or scalability, etc.

[0067] Security and privacy refer to the ability of the target functional module to protect data security and privacy during use. For example, whether the target functional module can prevent sensitive data leakage, whether it supports data encryption, and whether it complies with relevant laws and regulations (such as GDPR, CCPA, etc.). Security and privacy protection are important requirements that cannot be ignored in the actual application of the framework, especially in scenarios involving user data or sensitive information.

[0068] When the target functional module is a reflection system, the development requirements may also include the reflection ability requirements of the reflection system. The reflection ability requirements of the reflection system refer to whether the reflection system has the ability of online reflection and / or offline reflection.

[0069] The "online reflection" as described in the present invention means that during the process of an agent executing a user task, it conducts real-time reflection, correction, and experience summary to improve the accuracy and quality of completing this task.

[0070] The "offline reflection" as described in the present invention means that it regularly conducts reflection and experience summary on the execution process of historical tasks through historical logs, and saves the summarized experience to the knowledge base or model parameters. The experience obtained from offline reflection will be applied during the execution of future tasks to improve the accuracy and quality of subsequent tasks.

[0071] When the target functional module is a reflection system, the acceptance criteria may refer to whether the execution process includes a reflection process during the process of a self-evolving agent running test cases. If the execution process includes a reflection process, it indicates that the development of the reflection system is successful; if the execution process does not include a reflection process, it indicates that the development of the reflection system fails.

[0072] When the target function module is a memory system, the development requirements may further include at least one of the following: memory mechanism, memory content, memory format, extraction timing.

[0073] The memory mechanism includes at least one of the following: memory access, multi-round memory, long-term and short-term memory, working memory, memory compression, retrieval-augmented generation (RAG), cross-modal memory, knowledge graph memory, associative memory, parallel / serial memory extraction, distributed memory access, memory cleaning and archiving, hybrid-mode memory, inferential retrieval.

[0074] Memory content may include at least one of the following: the user's historical interaction records, context information, relevant information in the external knowledge base, dynamically generated intermediate results, data provided by the user (such as files, pictures, audio, etc.), the inference chains generated by the system, etc. These memory contents can be customized according to specific application scenarios to meet different functional requirements.

[0075] Memory format may include storage in at least one of the forms of structured data, text fragments, image features, knowledge graph nodes, embedding vectors, etc., for efficient retrieval and use.

[0076] Extraction timing, that is, the timing of extracting memory, may include at least one of the following: extracting during intent recognition, extracting during the execution of a specific task (such as when writing code), extracting during the thought chain reasoning process (such as when making an action decision), extracting during context switching (such as when the user switches sessions), extracting during initialization or state update (such as when the user starts a session).

[0077] When the target function module is a thinking framework, the development requirements may further include at least one of the following: thinking mode, visualization explanation.

[0078] The described thinking patterns may include at least one of the following: ReAct (Reason + Act), Chain-of-Thought (CoT), Tree-of-Thought (ToT), Graph-of-Thought (GoT), Plan-and-Resolve, Reflexion (reflection framework), Cycle-of-Thought (CoC), Diverge-Converge Thought (DCT), Analogical Thought (AT), Multi-Perspective Thought (MPT), Program of Thoughts (PoT), Stepwise Reflection, Multi-Agent Debate, Attention-Guided Thinking, Intuitive Leap, Check List, Multi-Modal Thinking, parallel thinking pattern, fusion framework, adaptive framework.

[0079] The described visualization explanation means that users may need a visualized description of how the thinking pattern works to more intuitively understand its operation method. For example, analogizing Chain-of-Thought to "building dominoes", Tree-of-Thought to "branch exploration of a decision tree", and Cycle-of-Thought to "a process of repeated trial and optimization". The visualization explanation helps to lower the understanding threshold and enables users to quickly master the core logic and application methods of the framework.

[0080] When the target function module is a requirements module, the development requirements may also include at least one of the following: interactive function requirements, location requirements in the workflow of the self-evolving agent, requirements confirmation point requirements, etc.

[0081] Interactive function requirements refer to whether multi-round dialogue functions are supported when the self-evolving agent receives a user question for specific requirements confirmation of the problem.

[0082] Location requirements in the workflow of the self-evolving agent refer to the position where requirements confirmation is located in the entire workflow of the self-evolving agent. For example, it is located between "receiving data analysis instructions" and "generating data analysis code".

[0083] The requirements confirmation point requirements refer to which requirements confirmation points the requirements module of the self-evolving intelligent agent needs to confirm with the user. For example, the requirements confirmation points may include: core goals, data confirmation, analysis dimensions, chart types, delivery forms, etc.

[0084] When the target function module is a learning system, the development requirements may further include at least one of the following: the input method of learning materials, whether to prompt the user to provide learning key points during learning, whether to require the user to provide knowledge tags during learning, the learning capabilities required by the learning system, the learning depth, whether to provide the ability to visually access the learned knowledge, and the application link of the learning system in the self-evolving intelligent agent.

[0085] The input method of learning materials refers to the way of providing the materials to be learned to the learning system. For example, it can be provided in the form of files or in a dialogue way on the interaction interface.

[0086] Whether to prompt the user to provide learning key points during learning. For example, when providing the code to be learned to the learning system, the learning system can be prompted to mainly learn which functions in the code.

[0087] The learning capabilities may include at least one of the following: knowledge learning, knowledge archiving, knowledge extraction, knowledge matching, knowledge application, knowledge practice, knowledge display, knowledge encoding, knowledge maintenance, knowledge version maintenance, multi-document knowledge learning, large-document knowledge learning, parallel learning.

[0088] Learning depth: It refers to the learning capabilities at different depths and may include at least one of the following: simple archiving, summary, concept summary, methodology summary, drawing inferences from one instance.

[0089] The application link of the learning system in the intelligent agent to be developed refers to the position of the learning system in the entire workflow of the self-evolving intelligent agent.

[0090] Step S2: The embedded intelligent agent determines the design scheme of the target function module according to the development requirements and in combination with the large language model method; In the present invention, the embedded intelligent agent determines the design scheme of the target function module according to the development requirements and in combination with the large language model method. Specifically, the embedded intelligent agent constructs prompt words according to the development requirements and calls the large language model to complete the task of generating the design scheme of the target function module.

[0091] In the present invention, according to the development requirements, the embedded agent determines the design scheme of the target functional module by combining the large language model method, which may specifically include: the embedded agent analyzes the development requirements to determine the design considerations and implementation strategies, and determines the functional architecture of the target functional module and the functional modification of the self-evolving agent according to the implementation strategy.

[0092] When the target functional module is a reflection system, the design scheme may include at least one of the following: Adding a reflection experience storage and extraction module, which is used to store the experiences obtained from reflection; and extracting the stored experiences during online reflection; Adding an online reflection link in the thought chain of the self-evolving agent, which is used to reflect on the answers given by the self-evolving agent during the conversation with the user; Adding an offline reflection module, which is used to reflect on the execution process of the historical tasks of the self-evolving agent to obtain experiences; Adding a storage module for saving historical conversations; Adding an experience confirmation module, which is used to manually confirm the experiences obtained from reflection.

[0093] Among them, the experiences stored in the reflection experience storage and extraction module may include the experiences obtained during online reflection (in this case, the self-evolving agent needs to have the ability of online reflection), or may include the experiences obtained during offline reflection (in this case, the self-evolving agent needs to have the ability of offline reflection).

[0094] Among them, adding an offline reflection module means that it can reflect on the execution process of the historical tasks of the self-evolving agent offline.

[0095] Among them, the purpose of adding a storage module for saving historical conversations is that it can be used for offline reflection and for querying historical conversations. In the present invention, it is also possible to support the recording of learning history and save it to the database to avoid repeated reflection on historical conversations.

[0096] Among them, the purpose of adding an experience confirmation module is to manually confirm the experiences obtained from reflection. Only after manual confirmation can they be put into storage to avoid storing inaccurate experiences. If the experiences are not verified, reflection and experience summary can be carried out again until the experience summary is aligned or reaches the upper limit of the failure threshold.

[0097] In the present invention, optionally, the reflection experience storage and extraction module includes at least one of the following: The storage and extraction module of the RAG (Retrieval-augmented Generation) method is used to store the reflected questions, answers, and experiences obtained from reflection in the vector database in vector form; and, when extracting experiences during online reflection, vector similarity matching and large language model discrimination methods are used for experience extraction; The storage and extraction module based on the experience model, where the experience model is obtained by fine-tuning the large language model of the self-evolving intelligent agent with the experiences obtained from reflection; when extracting experiences during online reflection, the self-evolving intelligent agent uses the experience model to extract experiences and discriminates the extracted experiences through the large language model.

[0098] Among them, the experiences stored in the storage and extraction module of the RAG method can be experiences obtained from online reflection or offline reflection. When extracting experiences during online reflection, vector similarity matching can be directly performed using the questions of the current task, or vector similarity matching can be performed using the answers to the questions of the current task.

[0099] In the present invention, large language model discrimination means that the large language model of the target intelligent agent is used to discriminate the extracted experiences, and it is determined whether to use the experiences after discrimination.

[0100] In the present invention, optionally, the reflection includes at least one of the following: Reflection based on a search engine, where the reflection based on a search engine means that when reflecting, the answer to the question is searched through the search engine, and reflection is carried out based on the searched answer to obtain experience; if standard questions and answers, or similar questions and answers can be searched, then experience learning and summarization can be carried out based on the searched questions and answers.

[0101] Reflection based on a large language model, where the reflection based on a large language model means that when reflecting, the large language model conducts reflection to obtain experience.

[0102] In the present invention, the above reflection can be online reflection or offline reflection, that is, online reflection can use the reflection methods of reflection based on a search engine and / or reflection based on a large language model for reflection, and offline reflection can also use the reflection methods of reflection based on a search engine and / or reflection based on a large language model for reflection.

[0103] Since the reflection based on the large language model has no reference to real answers, it is necessary to try to generate answers from different problem-solving methods and compare different problem-solving methods. In the present invention, optionally, the reflection based on the large language model includes: using the large language model to answer the problem from different perspectives to obtain multiple answers, comparing the multiple answers, determining the correct answer according to the comparison result, and reflecting based on the correct answer to obtain experience. Among them, "comparing the multiple answers and determining the correct answer according to the comparison result" can be: obtaining the answer with more occurrences as the preliminary reference answer, comparing the differences between other answers and the reference answer, including the problem-solving process and the answer, giving a difference analysis and a possible error analysis, and checking the error analysis points on each answer to verify whether the answer is correct. The finally obtained experience points can also be verified by manual verification.

[0104] In some embodiments, the correct answer can also be obtained from the conversation context. For example, the user has multiple rounds of questions and answers with the target intelligent agent, finds the information indicating the correct answer from the multiple rounds of questions and answers, and uses this correct answer as the standard answer to reflect on other rounds of questions and answers to obtain experience.

[0105] When the target functional module is the memory system, the design solution may include at least one of the following: the design solution of the system architecture of the memory system, the design solutions of the various modules of the memory system, etc. The design solution of the system architecture refers to the functional modules included in the memory system and the data flow between the various modules. The design solutions of the various modules refer to the functions of the various modules and the implementation methods of the functions, etc.

[0106] When the target functional module is the thinking framework, the design solution may include at least one of the following: the design solution of the system architecture of the thinking framework, the design solutions of the various modules of the thinking framework, nested logic design. Among them, the nested logic design is the nesting and organizational structure between various thinking modes supported by the target intelligent agent. The design solution of the system architecture refers to the functional modules included in the thinking framework and the data flow between the various modules. The design solutions of the various modules refer to the functions of the various modules and the implementation methods of the functions, etc.

[0107] When the target functional module is the requirements module, the design solution may include at least one of the following: the design of the interaction function of the requirements module, the design of the position of the requirements module in the workflow of the to-be-developed intelligent agent, the design of the requirements confirmation point, the design of adding the requirements confirmation point to the prompt words of the work steps after requirements confirmation, etc.

[0108] When the target functional module is the learning system, the design solution may include at least one of the following: Modify the task processing module of the self-evolving agent so that the task processing module can identify whether the user task is a learning task; Modify the upload module of the self-evolving agent to support the upload and archiving of learning materials; Add a learning system; Modify the task processing module in the self-evolving agent so that the task processing module can call the knowledge extraction module of the learning system and integrate the knowledge learned by the learning system into the prompt words of the task processing module; Add a practice module.

[0109] Step S3: The embedding agent generates the code of the target functional module according to the development requirements and design scheme, in combination with the large language model method; In the present invention, the embedding agent generates the code of the target functional module according to the development requirements and design scheme, in combination with the large language model method. Specifically, the embedding agent constructs prompt words according to the development requirements and design scheme and calls the large language model to complete the task of generating the code of the target functional module.

[0110] Step S4: The embedding agent deploys the running environment of the target functional module in combination with the large language model method and debugs the code of the target functional module based on the running environment; when the debugging result indicates that the target functional module does not meet the development requirements, check the design scheme and code of the target functional module in combination with the large language model method. When it is necessary to modify the design scheme, return to step S2. When it is necessary to modify the code, return to step S3 until the debugging result indicates that the target functional module meets the development requirements or the number of debugging times exceeds the preset threshold; In the present invention, the embedding agent deploys the running environment of the target functional module in combination with the large language model method and debugs the code of the target functional module based on the running environment. Specifically, the embedding agent constructs prompt words according to the development requirements, design scheme and the code, and calls the large language model to complete the tasks of running environment deployment and code debugging.

[0111] The embedding agent checks the design scheme and code of the target functional module in combination with the large language model method. Specifically, the embedding agent constructs prompt words according to the design scheme, code and debugging result of the target functional module and calls the large language model to complete the task of checking the design scheme and code.

[0112] It should be noted that when it is necessary to modify the design solution and return to step S2, in the new step S2, the embedded agent needs to re-determine the design solution of the target functional module according to the development requirements, the design solution, and the debugging results, in combination with the large language model method.

[0113] It should be noted that when it is necessary to modify the code and return to step S3, in the new step S3, the embedded agent regenerates the code of the target functional module according to the development requirements, the code, and the debugging results, in combination with the large language model method.

[0114] Step S5: The embedded agent feeds back the development result information of the target functional module to the user through the interaction interface of the self-evolving agent; In the present invention, the fed-back development result information may include at least one of the following: information about the code of the target functional module, debugging information of the target functional module, analysis of the applicable scenarios of the target functional module, optimization suggestions for the target functional module, and a conclusion on whether the development of the target functional module is successful or failed.

[0115] Among them, the conclusion of development failure may be a conclusion after multiple debugging sessions and the number of debugging sessions exceeds a preset threshold number.

[0116] The information about the code of the target functional module may be the code itself or the storage location of the code, etc.

[0117] The debugging information may include at least one of the following: running logs, intermediate data, debugging results, etc.

[0118] Step S6: The embedded agent puts the target functional module online.

[0119] In the present invention, by embedding a development agent in a target agent, a self-evolving agent with self-evolving ability is formed. The development agent embedded in the self-evolving agent can confirm whether it is developing a target functional module for the self-evolving agent, determine the development requirements of the target functional module, automatically complete the development, debugging, and online deployment of the target functional module, replace the process of manually developing the target functional module, realize the self-evolution of the agent, thus effectively saving development costs, and the intelligent computing center cloud platform can provide sufficient computing power resources for the self-evolution of the agent, greatly improving the efficiency of the self-evolution of the development agent.

[0120] In the present invention, optionally, before step S1, the following steps are further included: Step S01: The development agent receives the task information given by the user and embedded into the target agent through the interaction interface of the development agent; In the present invention, the task information may include at least one of the following: the name of the target agent, function description, current version, operating environment, interface document, existing functions (or capabilities), core code structure, etc.

[0121] In the present invention, the development agent can be a single agent or include multiple agents. The multiple agents form a multi-agent cooperation system. Each agent has specific capabilities, and the multiple agents can cooperate to complete the task of embedding the target agent and, subsequently, developing function modules for the target agent.

[0122] In the present invention, optionally, the multiple agents in the development agent include at least one of the following manufacturing agents: integration agent, tool function development agent, memory system development agent, thinking framework development agent, requirement confirmation development agent, learning system development agent, reflection system development agent, etc.

[0123] Among them, each of the above manufacturing agents can include at least one of the following basic agents: requirement confirmation agent, framework design agent, search agent, learning agent, coding agent, debugging agent, archiving agent, and deployment agent, etc. Or, some or all of the above multiple agents can also share one or more of the above basic agents.

[0124] Among them, the requirement confirmation agent is responsible for accurately understanding and clarifying the development requirements and converting them into executable design documents; The framework design agent is responsible for designing the overall architecture and sub-function modules of the target function module; The search agent is responsible for finding relevant technical materials and best practices to provide reference for the design; The learning agent summarizes the materials found and extracts key technologies and implementation methods; The coding agent converts the design scheme into code implementation; The debugging agent is responsible for testing and fixing problems and recording error information during the debugging process; The archiving agent archives and manages the documents, codes, and logs during the development process to ensure the traceability of the materials; The deployment agent is responsible for deploying the developed agent to the target operating environment and monitoring the running state.

[0125] The embedding agent is a copy of the above development agent, with the same structure as the above development agent, and will not be elaborated here.

[0126] Step S02: The development agent automatically identifies the relevant information of the target agent, and embeds itself into the target agent according to the relevant information of the target agent to form the self-evolving agent.

[0127] In the present invention, the development agent embedding into the target agent may include: the development agent copies its own code and copies the copied code into the target agent.

[0128] In the present invention, the development agent embedding into the target agent may further include at least one of the following: Modify the dialogue interface of the target agent, add an online dialogue monitoring module to the target agent, and the online dialogue monitoring module is used to send the execution process of the current task of the target agent to the embedded agent at the same time, and the embedded agent reflects on the execution process of the current task to determine the development requirements.

[0129] Add an offline dialogue monitoring module to the target agent, and the offline dialogue monitoring module is used to reflect on the execution process of the historical tasks of the target agent to determine the development requirements.

[0130] "Online" as described in the present invention means during the process of the agent executing the user task.

[0131] "Offline" as described in the present invention means when the agent is not executing the task.

[0132] "Execution process" as described in the present invention means the whole process that the agent receives the user instruction, parses the instruction, continuously makes decisions on the next action, executes the corresponding operation, generates a reply, reflects and corrects, and finally replies to the user. That is, the execution process may include the dialogue between the user and the target agent, as well as the thinking process of the target agent, etc.

[0133] In the present invention, the development agent automatically embeds itself into the target agent to form a self-evolving agent, without the need for artificial development of the development agent for the target agent, thus effectively saving the development cost.

[0134] In the present invention, optionally, the step S02 includes: Step S021: The development agent combines the large language model method to read the documents and codes of the target agent, and determines the relevant information of the target agent according to the reading results. The relevant information includes at least one of the following: architecture, interface, dependent model, extensible function module, function module already possessed by the target agent.

[0135] In the present invention, optionally, the step S1 includes: Step S11: The embedded agent determines that a target function module needs to be developed for the self-evolving agent according to the user's requirements for upgrading, expanding, or optimizing the functions of the self-evolving agent, and determines the development requirements of the target function module; or, the embedded agent reflects on the current task or historical tasks of the self-evolving agent, and determines that a target function module needs to be developed for the self-evolving agent according to the reflection result, and determines the development requirements of the target function module.

[0136] In the present invention, optionally, the embedded agent determines that a target function module needs to be developed for the self-evolving agent according to the user's requirements for upgrading, expanding, or optimizing the functions of the self-evolving agent, and determines the development requirements of the target function module. Specifically, the embedded agent constructs a prompt and invokes a large language model to generate responses or questions, etc., to complete the task of communicating with the user, and obtains the development requirements according to the conversation content.

[0137] In the present invention, the embedded agent can communicate with the user based on the large language model for at least one round. According to the communication content with the user, combined with the large language model method, it determines that a target function module needs to be developed for the self-evolving agent, and determines the development requirements of the target function module.

[0138] In the present invention, the embedded agent communicates with the user for at least one round to confirm the detailed development requirements of the target function module, ensuring a comprehensive and accurate understanding of the functions, performance, compatibility, etc. of the function module expected by the user, and laying a solid foundation for subsequent design and development work.

[0139] In the present invention, the embedded agent reflects on the current task or historical tasks of the self-evolving agent, and determines that a target function module needs to be developed for the self-evolving agent according to the reflection result, and determines the development requirements of the target function module. Specifically, it may include: the embedded agent identifies and sorts out the development requirements from the conversations of the current task or historical tasks of the self-evolving agent. Among them, sorting out the development requirements may include converting the identified development requirements into requirement descriptions in the required format.

[0140] In the present invention, optionally, step S1 includes: Step S12: The embedded agent compares the development requirements with the historical development requirements in the database, and determines whether to continue developing the target function module according to the comparison result. Among them, the database records the historical development requirements and the status of the historical development requirements, and the status includes at least one of the following: collected, confirmed, developed, launched, abandoned.

[0141] In the present invention, optionally, determining whether to continue developing the target functional module according to the comparison result may be: If the comparison result indicates that there is a historical development requirement in the database that is the same as or similar to the development requirement, and the status of the historical development requirement is developed or on-line or abandoned, then there is no need to continue developing the target functional module.

[0142] If the comparison result indicates that there is no historical development requirement in the database that is the same as or similar to the development requirement, then the target functional module can be continued to be developed.

[0143] By comparing the development requirement with the historical development requirements in the database and removing duplicates, requirement explosion or duplicate development can be avoided.

[0144] In the present invention, optionally, the step S1 may further include: Step S13: The embedded agent provides a development requirement display interface for displaying the historical development requirements in the database and the status of the historical development requirements for the user to view.

[0145] In the present invention, optionally, the step S1 may further include: Step S14: The embedded agent provides a confirmation module for the administrator to confirm whether to develop the target functional module and the development requirements determined by the embedded agent, so as to avoid uncontrollable transformation.

[0146] In the present invention, optionally, the step S6 may include: Step S61: The embedded agent uses the gray release method to go live with the target functional module. The gray release method means that the self-evolving agent with the target functional module added is opened to some users, and when there are no large-scale error reports, it is then officially launched.

[0147] In the present invention, optionally, the step S6 may include: Step S61: The embedded agent provides an administrator review module, and after the administrator reviews the target functional module that has been developed, it is then launched.

[0148] The following is an example to illustrate how the embedded agent determines whether to develop a target functional module for the self-evolving agent, how to determine the development requirements of the target functional module, and how to determine the design solution according to the development requirements.

[0149] First Embodiment: In this embodiment, the embedded agent of the question-and-answer robot (self-evolving agent) upgrades the thinking framework (target functional module) of the question-and-answer robot.

[0150] The following content is the historical conversation of the user with the Q&A robot, as well as the process by which the embedded agent of the Q&A robot identifies the development requirements in the historical conversation.

[0151] 1. Requirement Identification In this step, the embedded agent reflects on the historical tasks of the Q&A robot, determines whether a target function module needs to be developed for the Q&A robot, and determines the development requirements of the target function module.

[0152] <promp> # The conversation history is as follows **User**: The data analysis code you wrote often misses the null value check for each field, resulting in very low code availability. Can you perform a null value analysis for each field before data analysis in the future? **Q&A Bot**: The question you raised is very important. Indeed, ignoring null value checks in data analysis is a common problem that can lead to reduced code reliability.

[0153] In the future, when performing data analysis, I will ensure a null value analysis for each field, specifically including:[[]] Checking the number and proportion of missing values in each column; Visualizing the distribution of missing values; Providing handling suggestions based on the analysis results.

[0154] The following is a simple null value analysis template that you can refer to in future data analysis: ```python # Null value analysis # 1. Check the overview of missing values missing_values = df.isnull().sum() missing_percentage = (missing_values / len(df)) * 100 missing_df = pd.DataFrame({'Number of missing values': missing_values, 'Percentage of missing values (%)': missing_percentage}) missing_df = missing_df[missing_df['Number of missing values']>0].sort_values('Number of missing values', ascending=False) # 2. Visualize the missing values # Visualize the distribution of missing values for important fields # 3. Handling suggestions # Provide filling or deletion suggestions based on the missing situation ``` **User**: You're answering off-topic. I don't want you to write code for me. I hope you can have such a handling mechanism in the future.

[0155] **Question Answering Bot**: You're right. I misunderstood your intention. I understand that you want me to proactively incorporate a null value check mechanism when providing data analysis for you in the future, rather than asking for code examples.

[0156] I will keep this in mind and ensure that in future data analysis: Automatically perform null value analysis on each field; Report the situation of missing values; Provide reasonable handling suggestions; Thank you for your feedback. This will help me provide more reliable data analysis services.

[0157] # Your task You are a question answering bot. The above is the conversation between the user and you. Please indicate in XML format if there are any requirements in a certain aspect or implicit requirements for improvement in a certain aspect.

[0158] # Format parameters <thought> ... < / thought> <result> yes or no < / result> <content> ... < / content> Among them: - thought: is your thinking - result: can only be yes or no - content: is your sorting out of the requirements. Only when result is yes does content need to be given, otherwise content does not need to be provided Please start.

[0159] <response> <thought> The user clearly pointed out that the Q&A robot often omits the check for null values in data analysis, resulting in a decrease in the usability of the code, and hopes that the robot can automatically add a null value check mechanism in future data processing. This indicates that the user has put forward higher requirements for the reliability of the robot in data analysis. At the same time, the user emphasizes that it is not necessary to provide the code separately, but hopes that such processing logic is built into the robot. This is an implicit requirement, that is, to improve the quality and practicality of data analysis services and ensure that every step of analysis is based on a reliable data foundation.

[0160] < / thought> <result>yes< / result> <content> The user's requirement is that when the Q&A robot provides data analysis-related content, it should automatically include a processing mechanism for checking field null values. Specifically, it includes: 1. Automatically check the number and proportion of missing values for each field.

[0161] 2. Visualize the null value distribution (if applicable).

[0162] 3. Based on the analysis results, give reasonable suggestions for handling missing data.

[0163] This requirement indicates that the user expects the robot to have more rigorous and practical professional capabilities when outputting data analysis, in order to improve the usability of the code and the reliability of the analysis conclusions.

[0164] < / content> < / response> The following content is the process of organizing the development requirements identified by the embedded agent of the question answering bot.

[0165] 2. Requirement organization <prompt> # The following information has been collected: The user's requirement is that when the Q&A robot provides data analysis-related content, it should automatically include a processing mechanism for checking field null values. Specifically, it includes: 1. Automatically check the number and proportion of missing values for each field.

[0166] 2. Visualize the null value distribution (if applicable).

[0167] 3. Based on the analysis results, give reasonable suggestions for handling missing data.

[0168] This requirement indicates that the user expects the robot to have more rigorous and practical professional capabilities when outputting data analysis, in order to improve the usability of the code and the reliability of the analysis conclusions.

[0169] # Your task Please convert it into a specific requirement description, and the format requirement is to give it in xml format.

[0170] # Format reference <thought> ...< / thought> <content> ...< / content> Among them: - thought: is your thinking - content: is your sorting out of the requirements, and it is the specific requirements to be handed over to the design and development personnel for design. Please start.

[0171] < / prompt> <response> <thought>The user clearly requests that a field null value check and processing mechanism be included in the data analysis related content, which specifically includes three aspects: checking the quantity and proportion of missing values, visualizing the distribution of null values, and giving processing suggestions based on the analysis results. These requirements need to be clearly and accurately transformed according to the format to ensure that designers and developers can clearly understand the functions to be implemented. < / thought> <content>When processing data analysis-related content, the following requirements for the field null value check and processing mechanism need to be implemented: Automatically perform checks on the number and proportion of missing values for each field.

[0172] If applicable, visually present the null value distribution.

[0173] Based on the analysis results of the missing values, give reasonable suggestions for handling the missing data.< / content> < / response> The following content is the process of the embedded agent of the question answering bot (self-evolving agent) determining the design solution based on the organized development requirements.

[0174] 3. Design <prompt>Sample prompt words are omitted here< / prompt> <response> <thought> Requirement Analysis: The agent needs to automatically check for null values in the data analysis scenario; The existing CoT / ReAct inference process needs to be modified; Code execution ability needs to be integrated; Null value checks need to be triggered at the start of the analysis.

[0175] Design Considerations: How to identify the start of a data analysis task; How to seamlessly integrate null value checks without interrupting the analysis process; How to provide valuable null value handling suggestions; How to embed this functionality in the CoT / ReAct framework.

[0176] Implementation Strategy: Design a trigger mechanism to identify the data analysis scenario; Create a standardized null value analysis template code; Modify the inference process to insert null value checks at appropriate positions; Ensure that the analysis results are visualized and targeted suggestions are provided.

[0177] < / thought> <content> 1. Null value check function architecture Generate data analysis code request recognition → Whether data has been uploaded [Yes → Automatic null value check → Null value check result generation] → Continue to generate code.

[0178] 2. Function modification Add the running of Python scripts to support the running of null value analysis code and the observation of results.

[0179] < / content> <detail> 1. Modify the CoT / ReAct and ReAct process framework code to support the expansion of tool method functions, and add Python methods to execute tool methods, generate data analysis code request recognition tool methods, automatic null value check tool methods, and null value check result generation tool methods.

[0180] 2. Modify the Prompt of CoT / ReAct to support the selection of the above tool methods, and add relevant prompts for "generate data analysis code request recognition".

[0181] < / detail> < / response> Second Embodiment: In this embodiment, the embedded agent of the question answering bot (self-evolving agent) adds a memory system (target functional module) to the question answering bot.

[0182] The following content is the historical conversation of the user with the Q&A robot and the process by which the embedded agent of the Q&A robot identifies the development requirements in the historical conversation.

[0183] 1. Requirement Identification In this step, the embedded agent reflects on the historical tasks of the Q&A robot to determine whether a target functional module needs to be developed for the Q&A robot and to determine the development requirements of the target functional module.

[0184] <prompt> # The conversation history is as follows **User**: Do you remember the math problem I mentioned 10 days ago? **Q&A Bot**: I can't automatically remember the specific math problem you asked 10 days ago. If you need to review previous math problems, you can describe relevant information to me so that I can answer or recall relevant content based on the clues you provide.

[0185] # Your task You are the Q&A bot. The above is the conversation between the user and you. Please indicate in xml format whether it poses a certain need or implies a certain area for improvement.

[0186] # Format reference <thought> ... < / thought> <result> yes or no < / result> <content> ... < / content> Among them: - thought: is your thinking; - result: can only be yes or no; - content: is your summary of the need. Only provide content when result is yes, otherwise do not provide content.

[0187] Please start.

[0188] < / prompt> <response> <thought>The user asks if the mathematical problem mentioned 10 days ago is remembered. Since it cannot automatically remember the specific mathematical problem asked 10 days ago, the user's requirement exceeds its own capabilities.

[0189] < / thought> <result>yes< / result> <content>The requirement put forward by the user is to let the robot remember the math problem mentioned 10 days ago. However, the robot cannot automatically remember the specific math problem asked 10 days ago, so this requirement exceeds the capabilities of the robot.

[0190] < / content> < / response> The following content is the process of organizing the development requirements identified by the embedded agent of the Q&A robot.

[0191] 2. Requirement Organization <prompt> # The following information has been collected: The user's requirement is to let the robot remember the math problems asked 10 days ago, but the robot cannot automatically remember the specific math problems asked 10 days ago, so this requirement is beyond the robot's capabilities.

[0192] # Your Task Please convert it into a specific requirement description, format requirements, please provide it in XML format.

[0193] # Format reference <thought> ...< / thought> <content> ...< / content> in: - thought: your thoughts - content: This is your summary of the requirements, which will be handed over to designers and developers for design. Please start.

[0194] < / prompt> <response> <thought> The core requirement of the user is that the robot should have the ability to retrieve and match memories across 10 days, with the emphasis on memory storage and active retrieval in subsequent interactions. The original requirement states that "the robot cannot automatically remember questions from 10 days ago", indicating that the current system lacks this basic mechanism. Therefore, the expression should be simplified and the focus should be on building a functional module that supports a memory cycle of at least 10 days, enabling the system to match and reference historical memories when answering questions.

[0195] It should be emphasized that: - The preservation of memories should be automatically or semi-automatically completed by the system (e.g., based on context); - When answering new questions, the system should be able to retrieve and utilize relevant memories from the past 10 days at most; - This is an enhanced memory mechanism used to improve the relevance and personalization of answers.

[0196] The above can be translated into a requirement description that is understandable for design and development as follows: < / thought> <content> Requirement Name: Memory Storage and Retrieval Function Function Objective: Provide users with a memory cycle of at least 10 days. During this period, the system can extract and use the user's historical questions or conversation content when answering questions and can recall the questions asked in the past.

[0197] < / content> < / response> The following content is the process by which the embedded agent of the Q&A robot determines the design solution based on the organized development requirements.

[0198] 3. Design <prompt>Sample prompt words are omitted here< / prompt> <response> <thought> Requirement Analysis: The agent needs to provide user conversation memory for at least 10 days; It needs to be able to extract and use historical conversation content when answering questions; It needs to be able to recall the questions the user has asked in the past; It needs to maintain memory continuity between different conversation sessions.

[0199] Design Considerations: How to efficiently store a large amount of conversation history; How to quickly retrieve relevant historical information; How to manage long-term memory in a limited context window; How to distinguish important information from unimportant information; How to handle memory conflicts and updates; How to ensure privacy security while maintaining memory.

[0200] Implementation Strategies: Design a hierarchical memory architecture to distinguish between short-term and long-term memory; Implement vectorized storage to support semantic retrieval; Create a memory importance scoring mechanism; Design memory compression and summary generation algorithms; Establish a memory retrieval and activation mechanism; Implement a memory regular update and elimination strategy.

[0201] < / thought> <content> 1. Memory Storage and Retrieval Architecture User conversation → Memory processing → Memory storage → Memory retrieval → Memory integration → Response generation.

[0202] 2. Memory System Component Design Memory storage layer: Short-term memory: Complete conversation history of the current session; Medium-term memory: Conversation summary and key information in the last 7 days; Long-term memory: Storage of important information vectors for 10 - 30 days.

[0203] Memory processing mechanism: Conversation summary generation: Automatically extract key information points; Entity recognition: Identify key entities such as names, locations, times, etc.; Importance scoring: Based on factors such as user sentiment, repeated questions, etc.; Memory retrieval system: Semantic similarity search: Retrieve relevant history based on the current question; Time decay mechanism: Recent memories get higher weights; Context correlation analysis: Identify the topic continuity across conversations.

[0204] 3. Memory Integration Process User question → Analyze the current question → Retrieve relevant historical memories → Evaluate memory relevance → Integrate into the current context → Generate a response.

[0205] 4. Memory Management Strategies Memory compression: Regularly compress detailed conversations into key information summaries; Memory update: Update when new information conflicts with existing memories; Memory elimination: Automatically eliminate unimportant memories that exceed the retention period; Memory activation: Automatically activate historical memories when related topics appear.

[0206] < / content> <detail> Memory storage: Vector database: Uses vector storage to store the user's historical memories and recalls memories in the RAG manner; Graph database storage: Adopts the associated graph database method. The graph database includes timestamps, conversation content, summaries, entity tags, and importance scores; Hierarchical storage structure: Active memory layer, archived memory layer.

[0207] System integration: Modify the dialogue management module to perform memory retrieval before each user question; Integrate the retrieved historical memories during the context construction stage; Update the memory storage after reply generation; Implement regular memory maintenance tasks (compression, update, elimination); Performance optimization: Implement a memory caching mechanism to reduce retrieval latency; Design batch memory updates to reduce computational resource consumption; Create a memory index to optimize retrieval efficiency; < / detail> < / response> Third Embodiment: In this embodiment, the embedded agent of the Q&A robot (self-evolving agent) adds an interaction function (target functional module) to the Q&A robot.

[0208] The following content is the historical conversation of the user with the Q&A robot and the process by which the embedded agent of the Q&A robot identifies the development requirements in the historical conversation.

[0209] 1. Requirement Identification In this step, the embedded agent reflects on the historical tasks of the Q&A robot to determine whether a target functional module needs to be developed for the Q&A robot and to determine the development requirements of the target functional module.

[0210] <prompt> # The conversation history is as follows **User**: May I get a download link? I hope to download the document of each conversation in markdown format.

[0211] **Q&A Bot**: There is no official button or link to directly "download each conversation as a document" for now. However, there are several common workarounds for you to save the conversation content as a document.

[0212] # Your task You are the Q&A bot. The above is the conversation between the user and you. Please indicate in xml format if it poses a certain need or implies a certain area for improvement.

[0213] # Format reference <thought> ... < / thought> <result> yes or no < / result> <content> ... < / content> Among them: - thought: Your thinking; - result: Can only be yes or no; - content: Your summary of the need. Only provide content when result is yes, otherwise do not provide content.

[0214] Please start.

[0215] < / prompt> <response> <thought> The user's requirement is to obtain the download link for downloading each conversation as a markdown-formatted document. Currently, the Q&A robot does not have this function, and the user's requirement cannot be directly met, exceeding the current capabilities.

[0216] < / thought> <result>yes< / result> <content> The user requests to obtain the download link for each conversation as a markdown-formatted document. However, the Q&A bot currently does not have an official button or link to directly provide this function and cannot meet the user's needs.< / content> < / response> The following content is the process of organizing the development requirements identified by the embedded agent of the Q&A robot.

[0217] 2. Requirement Organization <prompt> # The following information has been collected: The user asked for a download link for each conversation in markdown format, but the Q&A robot does not yet provide an official button or link for this function, which cannot meet the user's needs. The above is the demand summarized based on user communication.

[0218] # Your Task Please convert it into a specific requirement description, format requirements, please provide it in XML format.

[0219] # Format reference <thought> ...< / thought> <content> ...< / content> in: - thought: your thoughts; - content: This is your summary of the requirements, which will be handed over to designers and developers for design. Please start.

[0220] < / prompt> <response> <thought> The user has put forward the requirement to obtain the download link of the conversation document presented in markdown format. Since the Q&A robot does not have the corresponding official button or link to meet this requirement, it is necessary to sort out this requirement to provide it to the design and development personnel for function design. When sorting out, it should be clear that the function the user wants is to obtain the download link, and the current situation where this function cannot be realized due to the lack of buttons or links.

[0221] < / thought> <content> The current user has a need to obtain the document download link for each conversation saved in Markdown format. However, there is no official button or link in the existing functions of the Q&A robot to directly provide this download link. Now, it is necessary to design and develop a function in the Q&A robot to allow users to obtain the document download link for each conversation in Markdown format, so that users can download the documents of each conversation presented in Markdown format through the corresponding button or link.

[0222] < / content> < / response> The following is the process by which the embedded agent of the Q&A robot determines the design solution based on the sorted development requirements.

[0223] 3. Design <prompt>Sample prompt words are omitted here< / prompt> <response> <thought> Requirement Analysis: Users need to obtain conversation documents saved in Markdown format; The current system does not have an official function to directly provide download links; A solution for adding a download function needs to be designed and developed; It is necessary to ensure that the exported document format complies with the Markdown specification.

[0224] Design Considerations: How to naturally integrate the download function into the existing UI; How to correctly convert conversation content into Markdown format; How to handle special elements in the conversation (code blocks, images, tables, etc.); How to generate and manage download links; How to ensure the security and user privacy of the download function; How to handle the export requirements for long conversations.

[0225] Implementation Strategy: Design UI components to add an obvious export button; Create a converter from conversation content to Markdown; Implement a mechanism for generating temporary files and link management; Design the file naming and organizational structure; Ensure the performance and reliability of the export function; Provide export options and customization functions.

[0226] < / thought> <content> 1. Markdown Export Function Architecture User requests export → Conversation content collection → Markdown conversion → File generation → Download link provision → User download.

[0227] 2. Functional Component Design UI Interaction Component; Add an "Export to Markdown" button at the top / bottom of the conversation interface; Export options pop-up window (selectable export range, format options); Download progress indicator; Markdown Conversion Engine; Conversation Structure Parser: Distinguish between user and AI messages; Content Format Converter: Process rich text, code blocks, images, etc.; Metadata Adder: Add information such as timestamp, session ID, etc.

[0228] File Management System: Temporary File Generator: Create.md files; Download Link Generator: Generate secure temporary download URLs; File Clean-up Mechanism: Regularly clean up expired files.

[0229] 3. Markdown Conversion Rules User message format: ## User (timestamp)\n\nUser message content; AI reply format: ## Assistant (timestamp)\n\nAssistant reply content; Code block handling: Wrap code content with the language name and three backticks (```); Image handling: Convert to the format; Table handling: Convert to Markdown table syntax; Mathematical formulas: Preserve the LaTeX format, wrap with $ and $$.

[0230] 4. File Organization Structure File naming format: conversation_YYYYMMDD_HHMMSS.md.

[0231] The file header contains: conversation summary, date, participant information.

[0232] Directory structure: Organize by conversation topic and date.

[0233] < / content> <detail> Front-end implementation: Add an export button component to the conversation interface; Implement the export option pop-up window and configuration interface; Create a download progress indicator and status feedback; Implement the client-side Markdown preview function.

[0234] Back-end implementation: Develop an API endpoint for collecting conversation content; Implement the Markdown conversion service; Create a temporary file storage and management system; Develop a secure file download link generator; Implement a file expiration and cleaning mechanism.

[0235] Security and privacy: Implement access control and verification for download links; Ensure the secure storage of temporary files; Add an option for user privacy data desensitization; Implement logging for download operations.

[0236] Performance optimization: Implement background asynchronous export processing; Add export task queue management; Optimize the file generation and download speed.

[0237] < / detail> Please refer to Figure 2 Figure 2 , the present invention also provides a system 10 for the self-evolution of an intelligent agent through computing power in an intelligent computing center cloud platform, including a self-evolving intelligent agent 11. The self-evolving intelligent agent 11 is a target intelligent agent embedded with a developed intelligent agent, and the embedded developed intelligent agent is an embedded intelligent agent 111; The embedded intelligent agent 111 includes: A requirement confirmation module 1111, configured to determine that a target function module needs to be developed for the self-evolving intelligent agent and determine the development requirements of the target function module; A design module 1112, configured to determine the design scheme of the target function module according to the development requirements in combination with the large language model method; A coding module 1113, configured to generate the code of the target function module according to the development requirements and design scheme in combination with the large language model method; A debugging module 1114, configured to deploy the operating environment of the target function module in combination with the large language model method and debug the code of the target function module based on the operating environment; when the debugging result indicates that the target function module does not meet the development requirements, check the design scheme and code of the target function module in combination with the large language model method. When the design scheme needs to be modified, trigger the design module 1112 to continue working. When the code needs to be modified, trigger the coding module 1113 to continue working until the debugging result indicates that the target function module meets the development requirements or the number of debugging times exceeds a preset threshold; A feedback module 1115, configured to feedback the development result information of the target function module to the user through the interaction interface of the self-evolving intelligent agent; A go-live module 1116, configured to go live the target function module.

[0238] In the present invention, by embedding a developed intelligent agent in a target intelligent agent, a self-evolving intelligent agent with self-evolution ability is formed. The developed intelligent agent embedded in the self-evolving intelligent agent can confirm whether to develop a target function module for the self-evolving intelligent agent and determine the development requirements of the target function module, automatically complete the development, debugging and go-live of the target function module, replace the process of manually developing the target function module, realize the self-evolution of the intelligent agent, thereby effectively saving the development cost, and the intelligent computing center cloud platform can provide sufficient computing power resources for the self-evolution of the intelligent agent, greatly improving the efficiency of the self-evolution of the developed intelligent agent.

[0239] Optionally, the intelligent computing center cloud platform's system 10 for realizing the self-evolution of intelligent agents through computing power further includes: A development intelligent agent, configured to receive task information given by a user and embedded in the target intelligent agent through the interaction interface of the development intelligent agent; automatically identify relevant information of the target intelligent agent, and embed itself into the target intelligent agent according to the relevant information of the target intelligent agent to form the self-evolving intelligent agent.

[0240] Optionally, the development intelligent agent is configured to combine the large language model method to read the documents and code of the target intelligent agent, and determine the relevant information of the target intelligent agent according to the reading results. The relevant information includes at least one of the following: architecture, interface, dependent model, extensible function module, and function module already possessed by the target intelligent agent.

[0241] Optionally, the requirement confirmation module 1111 is configured to determine the target function module to be developed for the self-evolving intelligent agent and determine the development requirements of the target function module according to the function upgrade, expansion, or optimization requirements of the self-evolving intelligent agent proposed by the user; or, the embedded intelligent agent reflects on the current task or historical task of the self-evolving intelligent agent, and determines the target function module to be developed for the self-evolving intelligent agent according to the reflection results, and determines the development requirements of the target function module.

[0242] Optionally, the requirement confirmation module 1111 is configured to compare the development requirements with the historical development requirements in the database, and determine whether to continue developing the target function module according to the comparison results. The database records historical development requirements and the status of the historical development requirements. The status includes at least one of the following: collected, confirmed, developed, launched, abandoned.

[0243] Optionally, the target function module includes at least one of the following: a reflection system, a memory system, tool functions, a thinking framework, a requirement module, and a learning system.

[0244] Please refer to Figure 3 , the present invention also provides an electronic device 20, including a processor 21, a memory 22, and a computer program stored on the memory 22 and executable on the processor 21. When the computer program is executed by the processor 21, it implements each process of the method embodiment of the intelligent computing center cloud platform for realizing the self-evolution of intelligent agents through computing power, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0245] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements each process of the above-mentioned method embodiment of the intelligent computing center cloud platform realizing the self-evolution of the intelligent agent through computing power, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0246] The present invention also provides a computer program product, including computer instructions, which when executed by a processor, implement the above-mentioned Figure 1 each process of the method embodiment of the intelligent computing center cloud platform realizing the self-evolution of the intelligent agent through computing power shown, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0247] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or device including that element.

[0248] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions for causing a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in various embodiments of the present invention.

[0249] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the present invention and the claims, and all of them belong to the protection scope of the present invention.< / response> < / promp>

Claims

1. A method for realizing the self - evolution of an intelligent agent through computing power in an intelligent computing center cloud platform, characterized in that, Including: Step S1: The embedded agent embedded in the self-evolving agent determines that a target function module needs to be developed for the self-evolving agent and determines the development requirements of the target function module. The self-evolving agent is a target agent embedded with a development agent, and the embedded development agent is the embedded agent; Step S2: The embedded agent determines the design scheme of the target function module according to the development requirements in combination with the large language model method; Step S3: The embedded agent generates the code of the target function module according to the development requirements and design scheme in combination with the large language model method; Step S4: The embedded agent deploys the running environment of the target function module in combination with the large language model method and debugs the code of the target function module based on the running environment; When the debugging result indicates that the target function module does not meet the development requirements, check the design scheme and code of the target function module in combination with the large language model method. When the design scheme needs to be modified, return to step S2. When the code needs to be modified, return to step S3 until the debugging result indicates that the target function module meets the development requirements or the number of debugging times exceeds the preset threshold; Step S6: The embedded agent feeds back the development result information of the target function module to the user through the interaction interface of the self-evolving agent; Step S7: The embedded agent goes live with the target function module.

2. The method according to claim 1, wherein Before the said step S1, it also includes: Step S01: The development agent receives the task information embedded in the target agent given by the user through the interaction interface of the development agent; Step S02: The development agent automatically identifies the relevant information of the target agent and embeds itself into the target agent according to the relevant information of the target agent to form the self-evolving agent.

3. The method according to claim 2, characterized in that, The said step S02 includes: Step S021: The development agent reads the document and code of the target agent in combination with the large language model method and determines the relevant information of the target agent according to the reading result. The relevant information includes at least one of the following: architecture, interface, dependent model, extensible function module, and function module already possessed by the target agent.

4. The method according to claim 1, characterized in that The said step S1 includes: Step S11: The embedded agent determines that a target function module needs to be developed for the self-evolving agent according to the user's requirements for function upgrade, expansion or optimization of the self-evolving agent, and determines the development requirements of the target function module; or, the embedded agent reflects on the current task or historical task of the self-evolving agent and determines that a target function module needs to be developed for the self-evolving agent according to the reflection result, and determines the development requirements of the target function module.

5. The method according to claim 1, wherein The said step S1 includes: Step S12: The embedded agent compares the development requirements with the historical development requirements in the database, and determines whether to continue developing the target function module according to the comparison result. The database records the historical development requirements and the status of the historical development requirements, and the status includes at least one of the following: collected, confirmed, developed, launched, abandoned.

6. The method according to claim 1, characterized in that, The target function module includes at least one of the following: reflection system, memory system, tool function, thinking framework, requirement module, and learning system.

7. A system for the self - evolution of intelligent agents through computing power in an intelligent computing center cloud platform, characterized in that, It includes a self-evolving agent, which is a target agent embedded with a development agent, and the embedded development agent is an embedded agent; The embedded agent includes: A requirement confirmation module, which is used to determine the target function module to be developed for the self-evolving agent and determine the development requirements of the target function module; A design module, which is used to determine the design scheme of the target function module according to the development requirements and in combination with the large language model method; A coding module, which is used to generate the code of the target function module according to the development requirements and design scheme and in combination with the large language model method; A debugging module, which is used to deploy the operating environment of the target function module in combination with the large language model method and debug the code of the target function module based on the operating environment; when the debugging result indicates that the target function module does not meet the development requirements, check the design scheme and code of the target function module in combination with the large language model method. When the design scheme needs to be modified, trigger the design module to continue working. When the code needs to be modified, trigger the coding module to continue working until the debugging result indicates that the target function module meets the development requirements or the number of debugging times exceeds the preset threshold; A feedback module, which is used to feedback the development result information of the target function module to the user through the interaction interface of the self-evolving agent; A launch module, which is used to launch the target function module.

8. The system according to claim 7, wherein It also includes: A development agent, which is used to receive the task information embedded in the target agent given by the user through the interaction interface of the development agent; automatically identify the relevant information of the target agent, and embed itself into the target agent according to the relevant information of the target agent to form the self-evolving agent.

9. An electronic device, characterized in that, It includes: A processor, a memory, and a program stored on the memory and executable on the processor. When the program is executed by the processor, it implements the steps of the method for realizing the self-evolution of the agent by the intelligent computing center cloud platform through computing power as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is executed by the processor, it implements the steps of the method for realizing the self-evolution of the agent by the intelligent computing center cloud platform through computing power as described in any one of claims 1 to 6.

11. A computer program product, characterized in that, It includes computer instructions. When the computer instructions are executed by the processor, it implements the steps of the method for realizing the self-evolution of the agent by the intelligent computing center cloud platform through computing power as described in any one of claims 1 to 6.

Images