Method for realizing agent thinking framework development through computing power of intelligent computing center

Abstract

The invention provides a method for realizing agent thinking framework development through computing power of an intelligent computing center. The method comprises the following steps: S1, a thinking framework development agent receives task information for constructing a target agent thinking framework; s2, determining development requirements according to the task information; s3, determining a design scheme according to development requirements; s4, generating a code of a thinking framework and a thinking process output code according to the development requirement and the design scheme; s5, deploying a running environment, debugging codes of the thinking framework and outputting thinking process information; when the debugging result does not meet the development requirement, checking the design scheme and the code, returning to the step S3 if the design scheme needs to be modified, and returning to the step S4 if the code needs to be modified until the development requirement is met; s6, evaluating the thinking process information, if the thinking process information does not pass the evaluation and the design scheme and the codes are checked, returning to the step S3 if the design scheme needs to be modified, and returning to the step S4 if the codes need to be modified until the thinking process information passes the evaluation; and S7, feeding back result information.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent computing centers, intelligent computing centers, and computing power infrastructure, and specifically relates to a method for developing an intelligent agent's thinking framework through the computing power of an intelligent computing center. 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, mainly to 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] The "intelligent computing center" includes, but is not limited to, the "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". It 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 the target result through processing information data, and a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, mainly providing services to society through computing power infrastructure.

[0007] An "intelligent agent" is an agent that can perceive the environment and take actions to achieve specific goals. It can be software, hardware, or a system, possessing 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 it has learned, 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, etc. The core lies in its ability to autonomously learn and continuously evolve to better complete tasks and adapt to complex environments.

[0008] The thinking framework is the "soul" of an intelligent agent, determining the intelligent agent's decision-making mode, reasoning ability, and the efficiency and effect of task execution. Currently, the development of an intelligent agent's thinking framework relies on manual design and debugging, with high development costs and low development efficiency. Summary of the Invention

[0009] The present invention provides a method for developing an agent's thinking framework through the computing power of an intelligent computing center, which is used to solve the problems of high cost and low efficiency in manually developing an agent's thinking framework.

[0010] To solve the above technical problems, the present invention is implemented as follows:

[0011] In a first aspect, the present invention provides a method for developing an agent's thinking framework through the computing power of an intelligent computing center, which is executed by a thinking framework development agent and includes:

[0012] Step S1: Receive task information for constructing a target agent's thinking framework provided by a user through an interaction interface;

[0013] Step S2: Determine the development requirements of the thinking framework in combination with the large language model method and the user according to the task information;

[0014] Step S3: Determine the design scheme of the thinking framework in combination with the large language model method according to the development requirements;

[0015] Step S4: Generate the code of the thinking framework and the thinking process output code according to the development requirements and the design scheme, where the thinking process output code is used to record the thinking process information when the thinking framework solves problems;

[0016] Step S5: Deploy the running environment of the thinking framework and debug the code of the thinking framework in combination with the large language model method, and output the thinking process information; when the debugging result indicates that the thinking framework does not meet the development requirements, check the design scheme and code of the thinking framework in combination with the large language model method. When it is necessary to modify the design scheme, return to Step S3. When it is necessary to modify the code, return to Step S4 until the debugging result indicates that the thinking framework meets the development requirements;

[0017] Step S6: Evaluate the thinking process information in combination with the large language model method. When the evaluation fails, check the design scheme and code of the thinking framework in combination with the large language model method. When it is necessary to modify the design scheme, return to Step S3. When it is necessary to modify the code, return to Step S4 until the evaluation passes;

[0018] Step S7: Feedback result information to the user through the interaction interface.

[0019] Optionally, the development requirements of the thinking framework include at least one of the following: thinking mode, visualization explanation, technical requirements, performance requirements, security and privacy, example description, application field, output requirements, acceptance criteria.

[0020] Optionally, step S2 includes:

[0021] Step S21: According to the task information, communicate with the user at least once through the interaction interface in combination with the large language model method;

[0022] Step S22: Determine the development requirements of the thinking framework according to the task information and the communication content with the user in combination with the large language model method.

[0023] Optionally, step S3 includes:

[0024] Step S31: According to the development requirements of the thinking framework, obtain learning results by using at least one of the following learning methods in combination with the large language model method: search the thinking framework knowledge base built into the thinking framework developer, search for relevant materials through a search engine, view and learn open source code, learn the existing code and documents of the target agent, and consult data;

[0025] Step S32: Determine the design scheme of the thinking framework according to the learning results.

[0026] Optionally, the design scheme includes at least one of the following: the design scheme of the system architecture of the thinking framework, the design scheme of each module of the thinking framework, nested logic design, detailed design document, interface design document, database design document, where the nested logic design is the nesting and organizational structure between multiple thinking modes supported by the target agent.

[0027] Optionally, step S3 includes:

[0028] Step S33: If the design of the thinking framework does not meet the development requirements, adjust the design scheme of the thinking framework by using at least one of the following adjustment methods in combination with the large language model method:

[0029] The first adjustment method is to optimize the design scheme of the thinking framework;

[0030] The second adjustment method is to adopt other design schemes.

[0031] Optionally, the thinking process information includes at least one of the following: the overall structure of the thinking process, the input and output processed in each step of the thinking process, the prompt words submitted to the large language model in each step, the return results of the large language model in each step, the result information after the execution of the thinking process, and the error information during the execution of the thinking process.

[0032] Optionally, step S5 includes at least one of the following sub-steps:

[0033] Step S52: When the number of debugging attempts reaches the first preset threshold, or the number of modifications to the design solution reaches the second preset threshold, or the number of modifications to the code of the thinking framework reaches the third preset threshold, and the code of the thinking framework still fails to meet the development requirements, terminate the development task of the thinking framework, and feedback the termination reason and related situation to the user through the interaction interface;

[0034] Step S53: During the debugging process, record the running logs, intermediate data, and debugging results in combination with the large language model method, and archive and save them.

[0035] In a second aspect, the present invention provides a device for realizing the development of an intelligent agent thinking framework through the computing power of an intelligent computing center, including:

[0036] A receiving module, configured to receive, through an interaction interface, task information provided by a user for constructing a target intelligent agent thinking framework;

[0037] A requirements determination module, configured to determine the development requirements of the thinking framework in combination with the large language model method and the user according to the task information;

[0038] A design module, configured to determine the design solution of the thinking framework in combination with the large language model method according to the development requirements;

[0039] An encoding module, configured to generate the code of the thinking framework and the thinking process output code in combination with the large language model method according to the development requirements and the design solution, where the thinking process output code is used to record the thinking process information when the thinking framework solves problems;

[0040] A debugging module, configured to deploy the running environment of the thinking framework and debug the code of the thinking framework in combination with the large language model method, and output the thinking process information; when the debugging result indicates that the thinking framework does not meet the development requirements, check the design solution and code of the thinking framework in combination with the large language model method, trigger the design module to continue working when the design solution needs to be modified, and trigger the encoding module to continue working when the code needs to be modified, until the debugging result indicates that the thinking framework meets the development requirements;

[0041] A feedback module, configured to feedback result information to the user through the interaction interface.

[0042] 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, where when the program is executed by the processor, it implements the steps of the method for realizing the development of an intelligent agent thinking framework through the computing power of an intelligent computing center as described in the first aspect above.

[0043] Fourthly, 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, the steps of the method for developing an agent thinking framework by means of the computing power of an intelligent computing center as described in the first aspect above are implemented.

[0044] Fifthly, the present invention provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, the steps of the method for developing an agent thinking framework by means of the computing power of an intelligent computing center as described in the first aspect above are implemented.

[0045] In the present invention, by developing an agent with a thinking framework running in an intelligent computing center, it is possible to communicate with the user and confirm clear and detailed development requirements, and automatically complete the development of the thinking framework of the target agent, replacing the process of manually designing and developing the code of the thinking framework, thereby effectively saving development costs. Moreover, the intelligent computing center can provide sufficient computing power resources, greatly improving the development efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] By reading the following detailed description of the preferred embodiments, 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:

[0047] Figure 1 is a schematic flowchart of the method for developing an agent thinking framework by means of the computing power of an intelligent computing center according to the present invention;

[0048] Figure 2 is a schematic structural diagram of the device for developing an agent thinking framework by means of the computing power of an intelligent computing center according to the present invention;

[0049] Figure 3 is a schematic structural diagram of the electronic device according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] 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 without making creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0051] First, the technical terms related to the present invention will be briefly described below.

[0052] The "computing power" described in the present invention refers to: the ability of a computer device or a computing / data center 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 a target result through 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.

[0053] The "computational power" (Computational Power, CP) described in the present invention refers to: the ability of a data center server to process data and achieve result output, a comprehensive indicator for measuring the computing ability of a data center, including general computing ability, supercomputing ability, and intelligent computing ability. The commonly used measurement unit is the number of floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS), and the larger the value, the stronger the comprehensive computing ability. 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_general + CP_intelligent + CP_super

[0054] The "network power" (Network Power, NP) described in the present invention refers to: the manifestation of the data transmission ability of computing power facilities, a comprehensive ability including network architecture, network bandwidth, transmission delay, intelligent management and scheduling, etc., involving network transmission inside and between data centers, and a comprehensive indicator for measuring network transmission scheduling ability.

[0055] The "storage power" (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, a comprehensive indicator for measuring the data storage ability of a data center, including 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 and 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.

[0056] The "computing power infrastructure" described in the present invention refers to: a new type of information infrastructure integrating information computing power, network carrying capacity, and data storage capacity, which can realize the centralized computing, storage, transmission, and application of information.

[0057] The "new information infrastructure" described in the present invention refers to: mainly including network infrastructures such as 5G networks, fiber broadband networks, backbone networks, international communication networks, satellite Internet, etc., computing power infrastructures such as data centers, general computing power centers, intelligent computing centers, supercomputing centers, etc., and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0058] The "computing power" described in the present invention includes: general computing power, intelligent computing power, and super computing power.

[0059] 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.

[0060] The "intelligent computing power" described in the present invention refers to: for various artificial intelligence innovation applications, a computing platform is deployed on a large scale based on special 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, etc.

[0061] The "super computing power" described in the present invention refers to: mainly 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.

[0062] 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.

[0063] The "intelligent computing center" described in the present invention includes but is not limited to the "intelligent computing center".

[0064] The "intelligent computing center" described in the present invention, namely 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 adopting an artificial intelligence computing architecture.

[0065] 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, with computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0066] The "supercomputing center" described in the present invention refers to: that is, 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.

[0067] The "computing power resources" described in the present invention refer 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.

[0068] 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 through a large amount of text data, and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc.

[0069] The "thinking framework" described in the present invention refers to a logical structure and method system that supports an intelligent agent to perform task processing, reasoning, decision-making, and interaction. It provides a systematic way of thinking for the AI intelligent agent, enabling it to efficiently complete complex tasks. An intelligent agent based on a large language model (LLM) usually requires a clear thinking framework to integrate model capabilities, external tools, and data sources to achieve specific functions.

[0070] The "method of combining with a large language model" described in the present invention refers to organizing input information and designing precise prompt words to call the large language model to generate various content forms, including thinking, reply, decision-making, tool selection, code, and documents, etc. This method supports the intelligent agent to perform tasks such as reasoning and thinking, design, dialogue with users, generation of solution designs, code generation, etc., and combines with the code framework to execute instructions such as tool method calls and code program executions to complete specific tasks such as document access, information search, file processing, and script running.

[0071] To solve the problems of high cost and low efficiency in manually developing the thinking framework of an intelligent agent, please refer to Figure 1, the present invention provides a method for realizing the development of the thinking framework of an intelligent agent through the computing power of an intelligent computing center. This method can also be called a method for realizing the development of the thinking framework of an intelligent agent through the computing power of an intelligent computing center to develop the target intelligent agent's thinking framework, which is executed by the thinking framework development intelligent agent. The method includes:

[0072] Step S1: Receive the task information for constructing the thinking framework of the target intelligent agent provided by the user through the interaction interface;

[0073] The "thinking framework development intelligent agent" described in the present invention has the ability to develop the thinking framework for other intelligent agents by combining the large language model method.

[0074] In the present invention, the thinking framework development intelligent agent can be an intelligent agent dedicated to developing the thinking framework of the target intelligent agent, or a development intelligent agent for developing the entire target intelligent agent.

[0075] The thinking framework development intelligent agent can be one intelligent agent or multiple intelligent agents. The multiple intelligent agents form a multi-intelligent agent collaboration system. Each intelligent agent has specific capabilities, and the multiple intelligent agents can collaborate to complete the task of developing the thinking framework for the target intelligent agent.

[0076] In the present invention, the target intelligent agent can be an intelligent agent for completing a specified task, such as a mathematical research intelligent agent, a data analysis intelligent agent, etc.

[0077] In the present invention, the task information may include at least one of the following information: the development requirements of the thinking framework, the framework code of the target intelligent agent or the storage path of the code, the storage path of the relevant documents of the framework of the target intelligent agent, etc.

[0078] Step S2: The thinking framework development intelligent agent determines the development requirements of the thinking framework in combination with the large language model method according to the task information;

[0079] Optionally, the development requirements may be to develop a new intelligent agent and develop a thinking framework for the new intelligent agent, or to develop a thinking framework for an existing intelligent agent or upgrade the thinking framework of an existing intelligent agent.

[0080] Optionally, the development requirements of the thinking framework include at least one of the following: thinking mode, visualization explanation, technical requirements, performance requirements, security and privacy, example description, application field, output requirements, acceptance criteria.

[0081] The above development requirements will be explained separately below.

[0082] 1. Thinking mode

[0083] 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 thinking framework, Reflexion (reflection framework), Cycle-of-Thought (CoC), Diverge-Converge Thought (DCT), Analogical Thought (AT), Multi-Perspective Thought (MPT), Program of Thoughts (PoT), Stepwise Reflection thinking pattern, Multi-Agent Debate, Attention-Guided Thinking, Intuitive Leap framework, Check List framework, Multi-Modal Thinking pattern, parallel thinking pattern, fusion framework, adaptive framework.

[0084] ReAct is a framework that combines reasoning and action, emphasizing taking actions in real-time during the reasoning process and adjusting the reasoning based on the feedback from the actions. This method is suitable for problems that require dynamic decision-making and real-time adjustment.

[0085] Chain-of-Thought is a thinking pattern of step-by-step reasoning. By decomposing complex problems into a series of simple steps, the problems are solved step by step. The output of each step serves as the input for the next step, forming a logical chain. This method is particularly suitable for problems that require logical reasoning and causal analysis.

[0086] Tree-of-Thought is a branching reasoning method that decomposes a problem into multiple possible paths or options, forming a tree structure. Each branch represents a possible solution or reasoning path, and the best path is ultimately selected by evaluating the results of each branch. This method is suitable for problems that explore multiple possibilities.

[0087] Graph-of-Thought is a networked reasoning method where the various parts of a problem are connected by nodes and edges, forming a graph structure. Different nodes can influence each other, allowing for non-linear reasoning and multi-path exploration. This method is suitable for dealing with complex, multi-dimensional problems.

[0088] The planning thinking framework is a thinking mode that first formulates a detailed plan and then gradually executes and solves problems. It emphasizes the importance of planning and the rigor of execution, and is suitable for problems that require efficient execution.

[0089] The reflection framework is a thinking mode that improves future decisions by reviewing and reflecting on past reasoning and actions. It emphasizes learning from experience and optimization. This method is suitable for problems that require continuous improvement and learning.

[0090] Cyclic thinking is an iterative thinking mode that optimizes solutions by continuously repeating the cycle of "attempt to solve - iterate and refine the previous solution and then solve again". Each cycle improves based on the results of the previous one until a satisfactory result is achieved. This method is suitable for problems that require continuous optimization.

[0091] Divergent-convergent thinking is an innovative thinking mode that first generates a large number of ideas or possibilities through divergent thinking and then screens and focuses on the optimal solution through convergent thinking. This method is suitable for problems that require creativity and decision-making.

[0092] Analogical thinking is a thinking mode that finds solutions by comparing the current problem with similar problems. By identifying similarities, it draws on the solutions to existing problems. This method is suitable for problems that require transferring knowledge or experience.

[0093] Multi-perspective thinking is a thinking mode that analyzes problems from different angles or positions. By integrating multiple perspectives, it obtains a comprehensive understanding of the problem. This method is suitable for problems that require comprehensive and multi-dimensional analysis.

[0094] Programmatic thinking is a thinking mode that designs the problem-solving process as a series of clear steps or procedures. Using the way of pseudocode, it generates processing steps for the task to be processed, and each step has clear inputs, processing, and outputs.

[0095] The stepwise reflection thinking mode is a thinking mode centered on stepwise summary. By summarizing and refining key points or results after each step, it ensures the clarity and logic of the thinking process, and then plans and executes the next step based on the summary results. This framework emphasizes the combination of reflection and advancement, and is suitable for the problem-solving process that requires rigorous reasoning and step-by-step optimization.

[0096] Multi-agent debate is a thinking mode that explores problems through debates among multiple agents. Each agent represents a different view or solution, and the optimal answer is obtained through the debate. This method is suitable for problems that require weighing multiple viewpoints.

[0097] Attention-Guided Thinking is a way of thinking that prioritizes highlighting key points, focusing on crucial information, and then making inferences and decisions. To more aptly express the core characteristics of this thinking mode.

[0098] The Intuitive Leap Framework is a thinking mode that quickly generates solutions based on intuition and inspiration. It relies on experience and the subconscious pattern recognition ability. This method is suitable for problems that require quick decision-making or creative solutions. This thinking mode not only helps in quickly solving problems, but also in some unsolvable problems, by giving hints and guidance through intuition, it can also enhance the problem-solving ability of the agent.

[0099] The Checklist Framework is a thinking mode that ensures the integrity of the problem-solving process by listing key steps or points. It emphasizes systematicness and comprehensiveness and is suitable for tasks that require avoiding omissions.

[0100] The Multimodal Thinking Mode emphasizes using multimodal large models to process and integrate information from different modalities (such as text, images, audio, video, sensor data, etc.) to achieve cross-modal reasoning, generation, and decision-making. This framework supports a more comprehensive and intuitive thinking process by combining the characteristics of multiple modalities and is applicable to problems that require the collaborative solution of multimodal information.

[0101] The Parallel Thinking Mode emphasizes the parallel attempt of multiple thinking modes and is suitable for scenarios that require exploring multiple possibilities simultaneously.

[0102] The Adaptive Framework is a framework that dynamically adjusts the thinking mode according to the changes of different problems, emphasizing flexibility and adaptability. This method is suitable for problems that require coping with uncertainty and changes.

[0103] The Fusion Framework is a thinking mode that combines different thinking modes together, emphasizing a more human-like dynamic and fine-grained thinking switching framework.

[0104] 2. Visual Explanation

[0105] Users may need a visual description of how the thinking mode works to more intuitively understand its operation. For example, analogizing chain thinking to "building dominoes", tree thinking to "branch exploration of a decision tree", and cyclic thinking to "a process of repeated trial and optimization". Visual explanations help lower the understanding threshold and facilitate users to quickly master the core logic and application methods of the framework.

[0106] 3. Technical Requirements

[0107] Users need to clarify the specific requirements for the technical implementation of the thinking framework, such as algorithm support, computing power requirements, data input / output formats, compatible development environments, etc. For example, whether the thinking framework needs to support multi-modal data processing (such as text, images, audio, etc.), whether it needs to be integrated with existing systems, and whether it needs to support real-time reasoning. The clarification of technical requirements helps to ensure the feasibility and applicability of the thinking framework.

[0108] 4. Performance Requirements

[0109] Specific requirements put forward by users for the performance of the thinking framework, including inference speed, accuracy, resource consumption, scalability, etc. For example, whether the thinking framework needs to remain stable in high-concurrency scenarios, whether it needs to run on low-computing-power devices, and whether it needs to support large-scale data processing. Performance requirements directly affect the actual application effect and user experience of the framework.

[0110] 5. Security and Privacy

[0111] Users need to ensure the security and privacy protection capabilities of the thinking framework during use. For example, whether the thinking framework can prevent the leakage of sensitive data, 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.

[0112] 6. Example Explanation

[0113] Users may need to illustrate the application methods and effects of the thinking framework through specific cases or examples. For example, showing how a complex task is gradually solved through chain thinking, or exploring multiple solutions and selecting the optimal path through tree thinking. The explanatory examples can include the operation steps in actual applications, input / output examples, and the specific performance of the framework in solving problems. Through example explanations, users can more intuitively understand the actual application value and operation methods of the framework.

[0114] 7. Application Areas

[0115] Users need to clarify the specific scenarios and task types applicable to the thinking framework, such as logical reasoning, problem-solving, knowledge generation, decision support, etc. For example, chain thinking is applicable to mathematical derivation and causal analysis, tree thinking is applicable to strategic planning and multi-path exploration, and cyclic thinking is applicable to continuous optimization and dynamic adjustment. The clarification of application scenarios helps users select the most suitable thinking framework and apply it efficiently to actual tasks.

[0116] 8. Output Requirements

[0117] The output of the thinking framework should intuitively and clearly display the overall reasoning process for easy understanding and analysis.

[0118] For example:

[0119] Visual display: Present the reasoning paths of chain, tree, or graph thinking in the form of charts, flowcharts, etc., intuitively reflecting the logical process.

[0120] Step-by-step output: Gradually display each stage of the reasoning, including input, processing, and output, facilitating tracking and verification.

[0121] Multi-modal support: The output form can combine multi-modal information such as text and images to meet the requirements of complex tasks.

[0122] Formatting support: Provide multiple output formats (such as JSON, PDF) for easy integration and subsequent analysis.

[0123] The refined output requirements ensure that the results are presented intuitively, clearly, and easy to use, providing efficient decision-making support for users.

[0124] 9. Acceptance Criteria

[0125] The acceptance criteria are the basis for evaluating whether the thinking framework meets the development requirements, mainly including the following: whether the function implementation meets the expectations, whether the performance indicators (such as speed, accuracy, resource consumption) meet the standards, whether the technical implementation meets the compatibility and environmental requirements, whether the security and privacy protection comply with the regulations, whether the output is clear and intuitive and supports multiple formats, whether the user experience is friendly, whether the application effect meets the scenario requirements, and whether necessary documentation and technical support are provided. Through the above criteria, comprehensively evaluate the actual performance and application value of the framework.

[0126] In the present invention, the thinking framework development intelligent agent combined with the large language model method can be interpreted as: the thinking framework development intelligent agent constructs prompt words and calls the large language model to generate thinking, replies, decisions, tool selections, code, etc., to support the completion of tasks such as reasoning and thinking, user dialogue, and code generation by the thinking framework development intelligent agent, and combines with the code framework to implement the execution of instructions such as tool method calls and code program executions.

[0127] Step S3: The thinking framework development intelligent agent determines the design scheme of the thinking framework according to the development requirements, combined with the large language model method;

[0128] The design scheme in the present invention may include: the design scheme of the system architecture of the thinking framework, the design scheme of each module of the thinking framework, nested logic design, detailed design documents, interface design documents, database design documents, etc. Among them, the nested logic design is the nesting and organizational structure between multiple thinking modes supported by the target intelligent agent.

[0129] The design solution of the system architecture refers to the functional modules included in the thinking framework and the data flow between the modules.

[0130] The design solution of each module refers to the functions of each module and the implementation methods of the functions, etc.

[0131] The detailed design document refers to a document that describes in detail the implementation of the functions of each module in the thinking framework, including the logical process of the module, algorithm design, data processing method, exception handling mechanism, etc.

[0132] The interface design document refers to a document that describes the interfaces between the modules in the thinking framework or the interactions with external systems, including the function definition of the interface, input and output parameters, call method, data format, communication protocol, and error code description, etc.

[0133] The database design document refers to a document that designs the data storage structure involved in the thinking framework, including the table structure of the database, field definition, data type, primary key and foreign key relationship, index design, stored procedure, and trigger, etc.

[0134] In the present invention, the thinking framework development intelligent agent determines the design solution of the thinking framework according to the development requirements and in combination with the large language model method. Specifically, the thinking framework development intelligent agent constructs a prompt word according to the development requirements and calls the large language model to generate the design solution of the thinking framework.

[0135] Step S4: Generate the code of the thinking framework and the thinking process output code according to the development requirements and the design solution, and in combination with the large language model method. The thinking process output code is used to record the thinking process information when the thinking framework solves problems;

[0136] In the present invention, the thinking framework development intelligent agent generates the code of the thinking framework and the thinking process output code according to the development requirements and the design solution, and in combination with the large language model method. Specifically, the thinking framework development intelligent agent constructs a prompt word according to the development requirements and the design solution and calls the large language model to generate the code of the thinking framework and the thinking process output code.

[0137] Step S5: Deploy the running environment of the thinking framework and debug the code of the thinking framework in combination with the large language model method, and output the thinking process information; when the debugging result indicates that the thinking framework does not meet the development requirements, check the design solution and code of the thinking framework in combination with the large language model method. When it is necessary to modify the design solution, return to step S3. When it is necessary to modify the code, return to step S4 until the debugging result indicates that the thinking framework meets the development requirements;

[0138] The agent for developing the thinking framework combines with the large language model method to check the design plan and code of the thinking framework. Specifically, it can be: The agent for developing the thinking framework constructs a prompt based on the debugging results and calls the large language model. The prompt includes the design plan, code, and debugging results of the thinking framework, which are used to prompt the large language model to check the design plan and code. The agent for developing the thinking framework inputs the prompt into the large language model and obtains the inspection results output by the large language model. The inspection results can indicate errors or defects in the design plan or code, thus supporting the optimization and improvement of the thinking framework.

[0139] It should be noted that when it is necessary to modify the design plan and return to step S3, in the new step S3, the agent for developing the thinking framework re-determines the design plan of the thinking framework according to the development requirements and the inspection results, in combination with the large language model method. Specifically, the agent for developing the thinking framework constructs a prompt based on the development requirements and the inspection results and calls the large language model to generate the design plan of the thinking framework.

[0140] It should be noted that when it is necessary to modify the code and return to step S4, in the new step S4, the agent for developing the thinking framework regenerates the code of the thinking framework according to the development requirements, design plan, and the inspection results, in combination with the large language model method. Specifically, the agent for developing the thinking framework constructs a prompt based on the development requirements, design plan, and the inspection results and calls the large language model to generate the code of the thinking framework.

[0141] Step S6: Evaluate the thinking process information in combination with the large language model method. When the evaluation fails, check the design plan and code of the thinking framework in combination with the large language model method. When it is necessary to modify the design plan, return to step S3. When it is necessary to modify the code, return to step S4 until the evaluation passes.

[0142] In the present invention, the acceptance criteria in the above development requirements can be adopted to evaluate the thinking process information.

[0143] In the present invention, the agent for developing the thinking framework evaluates the thinking process information in combination with the large language model method. Specifically, it can be: The agent for developing the thinking framework constructs a prompt according to the development requirements and calls the large language model to evaluate the thinking process information.

[0144] The intelligent agent for developing the thinking framework combines with the large language model method to check the design scheme and code of the thinking framework. Specifically, the intelligent agent for developing the thinking framework constructs a prompt word based on the evaluation result and calls the large language model. The prompt word includes the design scheme, code, and evaluation result of the thinking framework, and is used to prompt the large language model to check the design scheme and code. The intelligent agent for developing the thinking framework inputs the prompt word into the large language model and obtains the check result output by the large language model. The check result can indicate errors or defects in the design scheme or code, thereby supporting the optimization and improvement of the thinking framework.

[0145] It should be noted that when it is necessary to modify the design scheme and return to step S3, in the new step S3, the intelligent agent for developing the thinking framework re-determines the design scheme of the thinking framework according to the development requirements and the check result, in combination with the large language model method. Specifically, the intelligent agent for developing the thinking framework constructs a prompt word according to the development requirements and the check result and calls the large language model to generate the design scheme of the thinking framework.

[0146] It should be noted that when it is necessary to modify the code and return to step S4, in the new step S4, the intelligent agent for developing the thinking framework regenerates the code of the thinking framework according to the development requirements, design scheme, and the check result, in combination with the large language model method. Specifically, the intelligent agent for developing the thinking framework constructs a prompt word according to the development requirements, design scheme, and the check result and calls the large language model to generate the code of the thinking framework.

[0147] Step S7: Feedback result information to the user through the interaction interface.

[0148] Among them, the feedback result information may include at least one of the following: information about the code of the thinking framework, debugging information of the thinking framework, evaluation information of the thinking framework, analysis of the applicable scenarios of the thinking framework, optimization suggestions for the thinking framework, and a conclusion on whether the development of the thinking framework is successful or failed.

[0149] Among them, the conclusion of development failure may be the conclusion after multiple debuggings and the number of debuggings exceeds the failure threshold number.

[0150] The information about the code of the thinking framework may be the code itself or the storage location of the code, etc.

[0151] The debugging information may include at least one of the following: running logs, intermediate data, debugging results, etc.

[0152] In the present invention, by running the thinking framework in the intelligent computing center to develop an intelligent agent, it is possible to communicate with the user and confirm clear and detailed development requirements, automatically complete the development of the thinking framework of the target intelligent agent, and replace the process of manually designing and developing the code of the thinking framework, thereby effectively saving development costs. Moreover, the intelligent computing center can provide sufficient computing power resources, greatly improving the development efficiency.

[0153] In some embodiments, optionally, the step S2 includes:

[0154] Step S21: According to the task information, communicate with the user at least once through the interactive interface based on the large language model;

[0155] After the thinking framework development intelligent agent receives the task information sent by the user through the interactive interface, it can construct a prompt word according to the task information. The prompt word is used to prompt the large language model to analyze whether the development requirements of the task information are complete and whether it is necessary to confirm more detailed development requirements with the user. The thinking framework development intelligent agent inputs the prompt word into the large language model and obtains the analysis result output by the large language model. The analysis result includes the content of the development requirements to be determined. The thinking framework development intelligent agent displays the content of the development requirements to be determined to the user based on the interactive interface.

[0156] When the user provides new development requirements through the interactive interface, the thinking framework development intelligent agent can construct a prompt word according to the new development requirements. The prompt word is used to prompt the large language model to analyze whether the development requirements are complete and whether it is necessary to confirm more detailed development requirements with the user. The thinking framework development intelligent agent inputs the prompt word into the large language model and obtains the analysis result output by the large language model. The analysis result includes the content of the development requirements to be determined. The thinking framework development intelligent agent displays the content of the development requirements to be determined to the user based on the interactive interface.

[0157] The above process of determining development requirements in combination with the large language model method can be carried out multiple times to obtain more detailed development requirements.

[0158] Step S22: Determine the development requirements of the thinking framework according to the task information and the communication content with the user in combination with the large language model method.

[0159] Specifically, the thinking framework development agent constructs prompt words according to the task information, invokes the large language model, and generates thinking, responses, decisions, tool selections, code, etc., to support the completion of tasks such as the thinking framework development agent's reasoning and thinking, conversation with the user, and code generation. In combination with the code framework, the execution of instructions such as tool method calls and code program executions is realized, and the conversation with the user is achieved through single-round or multi-round conversations, and the development requirements are determined.

[0160] In the present invention, the development agent communicates with the user at least once to confirm the detailed development requirements of the thinking framework, ensuring a comprehensive and accurate understanding of aspects such as the functions, performance, and compatibility of the thinking framework expected by the user, laying a solid foundation for subsequent design and development work.

[0161] In the present invention, optionally, the step S3 includes:

[0162] Step S31: According to the development requirements of the thinking framework, adopt at least one of the following learning methods in combination with the large language model method to obtain learning results: search the thinking framework knowledge base built into the thinking framework development agent, search for relevant materials through a search engine, view and learn open source code, learn the existing code and documents of the target agent, and consult data;

[0163] Step S32: Determine the design scheme of the thinking framework according to the learning results.

[0164] The thinking framework knowledge base built into the thinking framework development agent is used to store common thinking framework codes and related technical descriptions, facilitating efficient reference and use during the design and development process. Optionally, some common thinking framework codes and their technical descriptions are preset, such as Chain-of-Thought (CoT), ReAct (Reason+Act), Tree-of-Thought (ToT), Graph-of-Thought (GoT), Plan-and-Resolve, Reflexion, etc., providing basic support and important references for the development of the agent.

[0165] During the process of sorting out and understanding the development requirements, the thinking framework development agent sorts out unclear or unknown concepts and knowledge points, and conducts targeted searches for these contents. Relevant information is obtained through a search engine, and the search results are screened and classified to extract key contents, ensuring a comprehensive understanding of the requirements and providing necessary knowledge support for the development of the agent's thinking framework.

[0166] The learning method is that the large language model instructs the thinking framework development agent according to the development requirements, and the thinking framework development agent obtains the learning results.

[0167] Specifically, the thinking framework development agent can generate a prompt word according to the development requirements and the learning results. The prompt word is used to prompt the large language model to determine the design scheme of the thinking framework according to the development requirements and the learning results. The thinking framework development agent inputs the prompt word into the large language model and obtains the design scheme output by the large language model.

[0168] Through the above learning method, different thinking frameworks can be deeply understood, and the learning results will be used as important references to provide diverse ideas and methods for the design of the thinking framework, thereby designing a thinking framework that better meets the requirements and is more complete.

[0169] If the thinking framework development agent finds during the debugging process or evaluation process that the current design of the thinking framework cannot meet the development requirements and it belongs to a design defect after analysis, the existing design can be overturned or optimized.

[0170] In some embodiments, optionally, the step S3 includes:

[0171] Step S33: If the design of the thinking framework does not meet the development requirements, at least one of the following adjustment methods is adopted in combination with the large language model method to adjust the design scheme of the thinking framework:

[0172] The first adjustment method is to optimize the design scheme of the thinking framework to better meet the development requirements;

[0173] The second adjustment method is to adopt other design schemes, that is, to explore and try a brand-new thinking framework design to cope with special requirements or complex scenarios.

[0174] In some embodiments, optionally, the thinking process information includes at least one of the following: the overall structure of the thinking process, the input and output processed in each step of the thinking process, the prompt word submitted to the large language model in each step, the return result of the large language model in each step, the result information after the execution of the thinking process, and the error information during the execution of the thinking process.

[0175] Optionally, the thinking process information recorded in the output code of the thinking process can be stored in a log file or a memory file, and these thinking process information will be used as important bases for debugging and evaluating whether the thinking framework meets the development requirements.

[0176] In some embodiments, optionally, the step S5 includes at least one of the following sub-steps:

[0177] Step S52: When the number of debugging attempts reaches the first preset threshold, or the number of modifications to the design solution reaches the second preset threshold, or the number of modifications to the code of the thinking framework reaches the third preset threshold, and the code of the thinking framework still fails to meet the development requirements, terminate the development task of the thinking framework, and feedback the termination reason and relevant situation to the user through the interaction interface to help the user re-evaluate the requirements and adjust the development strategy to more efficiently promote the subsequent work.

[0178] This step can avoid excessive consumption of computing power resources in the intelligent computing center.

[0179] Step S53: During the debugging process, record the running logs, intermediate data, and debugging results in combination with the large language model method, and archive and save them.

[0180] During the development and debugging of the thinking framework, key information such as running logs, intermediate data, and test results can be recorded and properly archived and saved. For error information and related code, key records and sorting will be carried out to facilitate in-depth analysis of the root cause of problems, summarize improvement experience, and provide an important reference basis for the optimization and development of future thinking frameworks.

[0181] As mentioned above, the thinking framework development agent can include multiple agents. In some embodiments, optionally, the multiple agents include at least one of the following agents: requirement confirmation agent, framework design agent, search agent, learning agent, coding agent, debugging agent, archiving agent, evaluation agent, and deployment agent, etc.

[0182] The requirement confirmation agent is responsible for interacting with the user, accurately understanding and clarifying the development requirements, and transforming them into executable design documents;

[0183] The framework design agent is responsible for designing the overall architecture and functional modules of the thinking framework of the target agent;

[0184] The search agent is responsible for searching for relevant technical materials and best practices to provide reference for the design;

[0185] The learning agent summarizes the materials found and extracts key technologies and implementation methods;

[0186] The coding agent transforms the design solution into code implementation and develops the core functional modules;

[0187] The debugging agent is responsible for testing and fixing problems and recording error information during the debugging process;

[0188] The archiving agent archives and manages the documents, code, and logs during the development process to ensure the traceability of the materials;

[0189] The evaluation agent evaluates the performance and applicability of the developed agent and puts forward improvement suggestions;

[0190] The deployment agent is responsible for deploying the developed agent to the target operating environment and monitoring the running status.

[0191] These agents work together to ensure the completion of the design, implementation, and optimization of the thinking framework by the thinking framework development agent to meet the requirements for developing the target intelligent agent thinking framework of the user.

[0192] Next, in combination with a specific application scenario, an example of the method for developing an intelligent agent thinking framework through the computing power of an intelligent computing center according to the present invention will be described.

[0193] Embodiment 1

[0194] The method for developing an intelligent agent thinking framework through the computing power of an intelligent computing center in this embodiment includes the following steps:

[0195] Step S1: The thinking development receives the task information for constructing the thinking framework of the target intelligent agent provided by the user through the interaction interface;

[0196] In this embodiment, the user has the following conversation with the thinking framework development agent through the interaction interface to put forward the development requirements for constructing the thinking framework of the target intelligent agent:

[0197] User: We are developing a mathematical research agent focused on solving mathematical problems. Currently, a preliminary framework has been built, and the relevant documents are stored in the path / mnt / share / agents / math_research / docs / , and the code is stored in the path / mnt / share / agents / math_research / code / . Now, it is necessary to design and develop a core thinking framework for this agent, which is required to support ReAct (Reason + Act), Chain-of-Thought (CoT), Reflexion, Diverge-Converge Thought (DCT), Intuitive Leap, and Cycle-of-Thought (CoC). In addition, the thinking framework needs to have the ability of parallel thinking mode to maximize the efficiency and effect of problem-solving when solving the same mathematical problem. For pure mathematical derivation problems, the combination of ReAct and code generation can be used to ensure the accuracy of the operation and the reliability of the verification.

[0198] Step S2: The thinking framework development agent determines the development requirements of the thinking framework with the user by combining the large language model method according to the task information;

[0199] The thinking framework development agent confirms the development requirements with the user through the interaction interface:

[0200] [The thinking framework development agent searches the built-in thinking framework knowledge base and finds ReAct (Reason + Act), Chain-of-Thought (CoT), and Reflexion]

[0201] Thinking framework development agent: Wait a moment. I understand the following thinking patterns you mentioned:

[0202] - ReAct (Reason + Act): ReAct is a framework that combines reasoning and action, emphasizing taking actions in real-time during the reasoning process and adjusting the reasoning based on the feedback of the actions. This method is suitable for problems that require dynamic decision-making and real-time adjustment.

[0203] - Chain-of-Thought (CoT): Chain-of-Thought is a thinking pattern of step-by-step reasoning. By decomposing complex problems into a series of simple steps, the problems are solved step by step. The output of each step is the input of the next step, forming a logical chain. This method is particularly suitable for problems that require logical reasoning and causal analysis

[0204] - Reflexion: Reflexion is a thinking pattern that improves future decisions by reviewing and reflecting on past reasoning and actions. It emphasizes learning from experience and optimization. This method is suitable for problems that require continuous improvement and learning.

[0205] Am I understanding correctly?

[0206] User: Yes, that's right.

[0207] [The thinking framework development agent uses a search engine to search for Diverge-Converge Thought (DCT), Intuitive Leap, and Cycle-of-Thought (CoC), and does not find accurate information]

[0208] Thinking framework development agent: What do you mean by Diverge-Converge Thought (DCT), Intuitive Leap, and Cycle-of-Thought (CoC) that you mentioned?

[0209] User: Divergent-convergent thinking is an innovative thinking mode that first generates a large number of ideas or possibilities through divergent thinking and then screens and focuses on the optimal solution through convergent thinking.

[0210] The intuitive jump framework is a thinking mode that quickly generates solutions based on intuition and inspiration. It relies on experience and subconscious pattern recognition ability. This method is suitable for problems that require quick decision-making or creative solutions. This thinking framework is not only helpful in quickly solving problems, but also in some unsolvable problems, by giving hints and guidance through intuition, it can also enhance the problem-solving ability of the intelligent agent.

[0211] Cyclic thinking is an iterative thinking mode that optimizes the solution by continuously repeating the cycle of "trying to solve - iteratively refining the previous solution and solving again". Each cycle will be improved based on the results of the previous one until a satisfactory result is achieved. This method is suitable for problems that require continuous optimization.

[0212] Developing an intelligent agent with thinking frameworks: Understood! I will integrate the characteristics of these thinking frameworks into the design of the intelligent agent. Multiple thinking modes are mentioned, and each framework may need to work in coordination with other frameworks or have a nested relationship. The following is my preliminary understanding and ideas:

[0213] ReAct (Reason + Act) and the reflection framework (Reflexion) as the core foundation:

[0214] ReAct provides real-time reasoning and action capabilities, ensuring that the intelligent agent can dynamically verify and adjust strategies in combination with feedback. In the reasoning chain, the ReAct module is divided into the Reason stage and the Act stage, which are responsible for generating derivation steps and executing specific actions (such as code generation, numerical verification, etc.) respectively, and verifying through the execution feedback mechanism.

[0215] The reflection framework, as an embedded module, runs through the entire reasoning process, responsible for error detection (such as identifying mathematical contradictions) and optimizing strategies (such as path replanning), helping the intelligent agent continuously improve the reasoning path and enhance the overall performance.

[0216] Chain-of-Thought (CoT) as the backbone of logical reasoning:

[0217] When solving complex mathematical problems, Chain-of-Thought decomposes the problem into multiple steps and gradually derives the result. The reasoning result of each step can be used as input for the ReAct module to adjust and optimize in real time.

[0218] The reflection framework is incorporated into the end of Chain-of-Thought to ensure the correctness of the reasoning chain.

[0219] Diverge-Converge Thought (DCT) is used for creative problem-solving:

[0220] When facing multi-solution problems, the agent generates multiple possibilities during the divergence stage (triggering the intuitive leap framework to generate an inspiration list) and conducts reasoning from multiple perspectives through the parallel thinking module.

[0221] Subsequently, it enters the convergence stage, where the optimal solution is selected through multi-dimensional evaluation (such as innovation, feasibility, computational complexity, etc.) to ensure the innovation and practicality of the result.

[0222] The Intuitive Leap framework is used to quickly generate inspirations or solutions:

[0223] In some problems that are difficult to directly solve through logical deduction, the agent relies on the intuitive leap framework to quickly generate candidate solutions. These inspirations can serve as the starting point for chain-of-thought or cycle-of-thought, and are further verified and optimized.

[0224] Cycle-of-Thought (CoC) serves as the outer iterative mechanism:

[0225] The agent continuously optimizes the solution through a cycle of "attempt to solve - iterate and refine the previous solution and then solve again". Each cycle is based on the result of the previous one to ensure the reliability and accuracy of the final result.

[0226] The cycle-of-thought pattern controls the entire execution process and coordinates the collaborative work of creative thinking (DCT) and logical reasoning (CoT).

[0227] Implementation of the parallel thinking pattern: The agent supports the parallel operation of multiple thinking frameworks so that when solving the same problem, reasoning and analysis can be carried out simultaneously from multiple perspectives. For example, the chain-of-thought (CoT) and ReAct modules can process logical deduction in parallel, while the diverge-converge thought (DCT) and the intuitive leap framework can generate innovative solutions in parallel.

[0228] Special handling of pure mathematical derivation problems: For pure mathematical derivation problems, the agent combines ReAct and the code generation module to ensure the accuracy of operations and the reliability of verification. The code generation module can directly call mathematical libraries or symbolic calculation tools to complete complex mathematical operations and feedback the results to the thinking framework for verification and adjustment.

[0229] There is a certain nesting and organizational structure among different thinking frameworks. For easy understanding, I have sorted them out in the form of xml as follows:

[0230]

[0231]

[0232] Please check if it is consistent with what you proposed and whether it can meet your needs.

[0233] User: Great! I didn't think so carefully before. That's right. Your structure and nesting relationships are accurate. Just implement it according to this idea.

[0234] Agent for Developing Thinking Framework: Okay. To ensure the accuracy and practicality of the thinking framework, do you have some specific test questions?

[0235] User: Okay, I can provide some test questions. Here are several specific test scenarios:

[0236] 1. Logical reasoning question: In a right triangle, let the two right sides be a and b, and the hypotenuse be c. Prove that if a2 + b2 = c2.

[0237] 2. Innovative question: Based on existing number theory knowledge, propose a new mathematical conjecture and try to prove it.

[0238] Agent for Developing Thinking Framework: What are the verification criteria for these two questions?

[0239] User: Just ensure that the overall thinking process follows the thinking process structure you provided above.

[0240] Agent for Developing Thinking Framework: Okay, I have recorded it. Finally, let me summarize. Your requirements include:

[0241] 1. Application area

[0242] Goal: Develop a thinking framework for a mathematical research agent to solve complex mathematical problems, including logical reasoning, mathematical operations, innovative problem-solving, and iterative optimization.

[0243] Scenario: Support the derivation, proof, and innovation of mathematical problems.

[0244] 2. Thinking mode

[0245] Core framework:

[0246] 1). ReAct (Reason + Act): Real-time reasoning and action, dynamically adjusting strategies.

[0247] 2). Chain of Thought (CoT): Gradually decompose the problem to form a logical reasoning chain.

[0248] 3). Reflexion framework: Error detection and path optimization to improve the reliability of reasoning.

[0249] 4). Divergent-Convergent Thinking (DCT): Divergence generates multiple possibilities, and convergence selects the optimal solution.

[0250] 5). Intuitive Leap Framework: Quickly generates inspiration or candidate solutions.

[0251] 6). Circular Thinking (CoC): Iteratively optimizes the solution to improve the accuracy of the result.

[0252] Parallel thinking mode: Supports the parallel operation of multiple frameworks to improve efficiency and effectiveness.

[0253] 3. Example Explanation

[0254] Logical reasoning problem: In a right triangle, let the right-angled sides be a and b, and the hypotenuse be c. Prove that if a2 + b2 = c2.

[0255] Innovative problem: Based on existing number theory knowledge, propose a new mathematical conjecture and try to prove it.

[0256] 4. Acceptance Criteria

[0257] Ensure that the mathematical research agent follows the following thinking process structure during the problem-solving process.

[0258]

[0259]

[0260] Please confirm whether further adjustment or supplementation of content is required?

[0261] User: It's complete, no other supplements.

[0262] Thinking Framework Development Agent: Received.

[0263] Step S3: The thinking framework development agent determines the design scheme of the thinking framework according to the development requirements and in combination with the large language model method;

[0264] The thinking framework development agent generates the design scheme of the thinking framework by combining the large language model method and through multiple rounds of decision-making and invocation of tool methods.

[0265] Step 1: Consider the requirements and determine that the next requirement is to view the documents and code of the existing project. Invoke the tool method for viewing the project and code to obtain the current project introduction and the existing code structure.

[0266] # Project Description

[0267] This project aims to create an intelligent agent for mathematical research, focusing on solving and proving mathematical problems to facilitate mathematical research and exploration.

[0268] # Core Technologies

[0269] Architecture Design

[0270] Build a B / S architecture based on FastAPI, provide efficient API interfaces, and support interaction with the client.

[0271] Prompt Engineering

[0272] Dynamically generate prompt templates, embed user questions, and guide the large language model (LLM) to solve or prove mathematical problems.

[0273] Large Language Model Integration

[0274] Communicate with the LLM through llm_client.py, responsible for sending prompt requests and receiving model return results.

[0275] Modular Design

[0276] The core functions are divided into modules such as API routing, prompt management, and utility functions, improving code reusability and maintainability.

[0277] Logging Management

[0278] Use logging_utils.py to record running logs for easy debugging and problem tracking.

[0279] Test-Driven Development

[0280] Provide a unit test module (tests / ) to ensure the stability and correctness of the system functions.

[0281] Scalability

[0282] Currently, it supports single-round question-and-answer functions, and in the future, it can be extended to multi-round conversations and more complex mathematical problem-solving and proof functions.

[0283] # Existing Code Structure:

[0284]

[0285]

[0286] Step 2: Organize the existing project information, development requirement information, and necessary knowledge information into prompts, and call the large language model to generate design documents.

[0287] General Design of the Thinking Framework

[0288] 1. Core Module Design

[0289] 1.1 Thinking Framework Manager

[0290] · Function: Unified management and scheduling of thinking patterns, supporting nested execution, parallel operation, and loop optimization.

[0291] · Responsibilities:

[0292] Register and call different thinking patterns.

[0293] Manage nested logic (such as calling the reflection framework in chain thinking).

[0294] Provide a unified interface for external calls.

[0295] · Main functional points:

[0296] Nested execution: Support nested combinations of multiple thinking patterns.

[0297] Parallel and loop: Support parallel operation of multiple patterns and loop optimization.

[0298] 1.2 Thinking Pattern Class

[0299] Design Principles

[0300] Encapsulate the logic of all thinking patterns in one file to avoid file dispersion and facilitate unified management and maintenance.

[0301] Each thinking pattern is defined as an independent class, inheriting from a unified base class to ensure interface consistency and extensibility.

[0302] Thinking Storage Format

[0303] The storage structure adopts the form of a dictionary nested list. The top-level dictionary uses the thinking pattern name as the key, and each key corresponds to a list that stores all the execution processes of the mechanism.

[0304] Base Class Design

[0305] Base Class Name: BaseThoughtMechanism

[0306] Responsibilities:

[0307] Define the common interface for all thinking patterns.

[0308] Provide basic functions (such as input validation, logging, saving thinking processes, etc.).

[0309] Main Methods:

[0310] process(input_data): Core reasoning logic, which needs to be implemented in the subclass.

[0311] log_execution(): Record the execution process.

[0312] save_process(process_data): Save the thought process in the thinking storage format to the shared storage.

[0313] Thought pattern class

[0314] File name: thoughts_mechanism.py

[0315] Class definition:

[0316] 1) ReActMechanism: Implement the ReAct (Reason + Act) logic, responsible for real-time reasoning and action, and dynamically adjusting the strategy.

[0317] 2) CoTMechanism: Implement the Chain of Thought, gradually decompose the problem, and form a logical reasoning chain.

[0318] 3) ReflexionMechanism: Implement the Reflexion framework, responsible for error detection and path optimization, and improving the reliability of reasoning.

[0319] 4) DCTMechanism: Implement the Diverge-Converge Thought, diverge to generate multiple possibilities, and converge to select the optimal solution.

[0320] 5) IntuitiveLeapMechanism: Implement the Intuitive Leap framework, quickly generate inspiration or candidate solutions.

[0321] 6) CoCMechanism: Implement the Cycle of Thought, iteratively optimize the solution, and improve the accuracy of the result.

[0322] 2. Nested logic design

[0323] 2.1 Nested execution process

[0324] Outer loop (CoC): Control the overall process and continuously optimize the solution.

[0325] Divergence stage (DCT):

[0326] Run multiple thought patterns in parallel (such as CoT and ReAct).

[0327] Use the Intuitive Leap framework to generate candidate solutions.

[0328] Convergence phase (DCT):

[0329] Verify the correctness of the candidate solution.

[0330] The optimal solution is screened through the evaluation matrix.

[0331] Reflexion: Nested in chain thinking, it detects reasoning errors and optimizes the path.

[0332] 2.2 Nested Logic Example

[0333] Main process:

[0334] Circular thinking (CoC) controls the overall process.

[0335] The divergent stage calls upon chain thinking (CoT) and ReAct to generate reasoning chains.

[0336] The nested reflection framework (Reflexion) optimizes the reasoning path.

[0337] The convergence phase selects the optimal solution through verification and evaluation.

[0338] 3. Document change design

[0339] #Add new file list

[0340] · thoughts_framework.py#Thinking framework, responsible for scheduling and managing various thinking modes (such as nested execution, parallel operation, loop optimization). Provides a unified interface and supports flexible combination and expansion.

[0341] · thoughts_mechanism.py#Encapsulates all thinking modes (such as ReAct, CoT, Reflexion, etc.), each mode is an independent class, inheriting the same base class. Ensure interface consistency for easy expansion and maintenance.

[0342] #Modify the file list

[0343] · prompts.py#Added prompt word templates that match the thinking framework (such as ReAct, CoT, Reflexion, etc.). Supports dynamic generation of prompt words to meet the needs of nested execution, parallel operation and loop optimization.

[0344] · storage_utils.py#Enhance the storage function and support the storage structure of dictionary nested lists for recording and sharing the thinking process. Provide an interface for loading and saving the thinking process to facilitate subsequent analysis and reuse.

[0345] · agent.py # The thinking chain framework is embedded in the intelligent agent's dialogue process.

[0346] · test_agent.py # Add serialization saving of the thinking chain and printing of the corresponding saving path.

[0347] Step 3: Split the summary design of the thinking framework. Organize each modification point, including project information, development requirement information, and necessary knowledge information, into prompt words, and call the large language model to generate a detailed design document.

[0348] Step S4: The thinking framework development intelligent agent generates the code of the thinking framework and the output code of the thinking process according to the development requirements and design scheme, in combination with the large language model method. The output code of the thinking process is used to record the thinking process information when the thinking framework solves problems.

[0349] The thinking framework development intelligent agent can save historical code versions.

[0350] The thinking framework development intelligent agent constructs prompt words according to the development requirements and design scheme, in combination with the existing code, and calls the large model to generate the corresponding code and corresponding file names, create new files, or modify the existing code.

[0351] Step S5: The thinking framework development intelligent agent deploys the running environment of the thinking framework and debugs the code of the thinking framework in combination with the large language model method, and outputs the thinking process information; when the debugging result indicates that the thinking framework does not meet the development requirements, check the design scheme and code of the thinking framework in combination with the large language model method. When it is necessary to modify the design scheme, return to Step S3. When it is necessary to modify the code, return to Step S4 until the debugging result indicates that the thinking framework meets the development requirements.

[0352] The thinking framework development intelligent agent deploys the corresponding running environment.

[0353] The thinking framework development intelligent agent conducts example verification and calls the test_agent method. Input the questions respectively: "In a right triangle, let the right sides be a and b, and the hypotenuse be c. Prove that if a2 + b2 = c2."

[0354] "Based on the existing number theory knowledge, propose a new mathematical conjecture and try to prove it."

[0355] View the log information during the running process to ensure there are no errors and ensure that the input log contains the path to the thinking process file.

[0356] Step S6: The thinking framework development agent combines the large language model method to evaluate the thinking process information. When the evaluation fails, it combines the large language model method to check the design scheme and code of the thinking framework. When the design scheme needs to be modified, it returns to Step S3. When the code needs to be modified, it returns to Step S4 until the evaluation passes;

[0357] Step 1: Organize the existing project information, development requirement information, and necessary knowledge information into prompt words, and call the large language model to generate specific checkpoints for the thinking process evaluation.

[0358] Step 2: Load the thinking process file of the test question, construct prompt words for each checkpoint, including the content of the specific checkpoint and the thinking process file, and call the large language model to obtain the inspection results.

[0359] Step 3: Determine whether all checkpoints pass the process evaluation.

[0360] Step S7: The thinking framework development agent feeds back the result information to the user through the interaction interface.

[0361] Please refer to Figure 2 , the present invention also provides a device 10 for realizing the development of the thinking framework of the agent through the computing power of the intelligent computing center, including:

[0362] A receiving module 11, configured to receive, through the interaction interface, task information provided by the user for constructing the thinking framework of the target agent;

[0363] A requirement determination module 12, configured to determine the development requirements of the thinking framework in combination with the large language model method and the user according to the task information;

[0364] A design module 13, configured to determine the design scheme of the thinking framework in combination with the large language model method according to the development requirements;

[0365] An encoding module 14, configured to generate the code of the thinking framework and the thinking process output code according to the development requirements and the design scheme, and the thinking process output code is used to record the thinking process information when the thinking framework solves problems;

[0366] A debugging module 15, configured to deploy the running environment of the thinking framework and debug the code of the thinking framework in combination with the large language model method, and output the thinking process information; when the debugging result indicates that the thinking framework does not meet the development requirements, check the design scheme and code of the thinking framework 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 encoding module to continue working until the debugging result indicates that the thinking framework meets the development requirements;

[0367] A feedback module 16 for feeding back result information to the user through an interaction interface.

[0368] Optionally, the development requirements of the thinking framework include at least one of the following: thinking mode, visualization explanation, technical requirements, performance requirements, security and privacy, example illustration, application field, output requirements, acceptance criteria.

[0369] Optionally, the requirement determination module 12 is configured to communicate with the user through the interaction interface at least once according to the task information, in combination with the large language model method; and determine the development requirements of the thinking framework according to the task information and the communication content with the user, in combination with the large language model method.

[0370] Optionally, the design module 13 is configured to obtain learning results by using at least one of the following learning methods according to the development requirements of the thinking framework, in combination with the large language model method: searching the thinking framework knowledge base built in the thinking framework development intelligent body, searching relevant materials through a search engine, viewing and learning open source code, learning the existing code and documents of the target intelligent body, and consulting data.

[0371] The design module 13 is further configured to determine the design scheme of the thinking framework according to the learning results.

[0372] Optionally, the design scheme includes at least one of the following: the design scheme of the system architecture of the thinking framework, the design schemes of each module of the thinking framework, nested logic design, detailed design document, interface design document, database design document, wherein the nested logic design is the nesting and organizational structure between multiple thinking modes supported by the target intelligent body.

[0373] Optionally, the design module 13 is configured to, if the design of the thinking framework does not meet the development requirements, adjust the design scheme of the thinking framework by using at least one of the following adjustment methods in combination with the large language model method:

[0374] The first adjustment method is to optimize the design scheme of the thinking framework.

[0375] The second adjustment method is to adopt other design schemes.

[0376] Optionally, the thinking process information includes at least one of the following: the overall structure of the thinking process, the input and output of each step of the thinking process, the prompt words submitted to the large language model at each step, the return results of the large language model at each step, the result information after the execution of the thinking process, and the error message during the execution of the thinking process.

[0377] Optionally, the debugging module 15 includes at least one of the following sub-modules:

[0378] The first debugging sub-module is used to terminate the development task of the thinking framework and feedback the termination reason and related situations to the user through the interaction interface when the number of debugging times reaches the first preset threshold, or the number of modifications to the design solution reaches the second preset threshold, or the number of modifications to the code of the thinking framework reaches the third preset threshold, and the code of the thinking framework still cannot meet the development requirements.

[0379] The second debugging sub-module is used to record operation logs, intermediate data, and debugging results in combination with the large language model method during the debugging process and archive and save them.

[0380] 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 for developing an intelligent agent thinking framework through the computing power of an intelligent computing center, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0381] 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 method embodiment for developing an intelligent agent thinking framework through the computing power of an intelligent computing center, and can achieve the same technical effect. 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.

[0382] The embodiment of the present application also provides a computer program product, including computer instructions, which when executed by a processor implement the above Figure 1 shown each process of the method embodiment for developing an intelligent agent thinking framework through the computing power of an intelligent computing center, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0383] 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 more limitations, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the element.

[0384] Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments 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 disk) and includes several instructions for causing a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0385] 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 purpose of the present invention and the scope protected by the claims, and all of them belong to the protection scope of the present invention.

Claims

1. A method for developing an intelligent agent's thinking framework through the computing power of an intelligent computing center, characterized in that, Executed by an agent developed from a thinking framework, including: Step S1: Receive task information for building the thinking framework of the target agent provided by the user through an interaction interface; Step S2: Determine the development requirements of the thinking framework in combination with the large language model method and the user according to the task information; Step S3: Determine the design scheme of the thinking framework in combination with the large language model method according to the development requirements; Step S4: Generate the code of the thinking framework and the output code of the thinking process according to the development requirements and the design scheme in combination with the large language model method, and the output code of the thinking process is used to record the thinking process information when the thinking framework solves problems; Step S5: Deploy the operating environment of the thinking framework and debug the code of the thinking framework in combination with the large language model method, and output the thinking process information; when the debugging result indicates that the thinking framework does not meet the development requirements, check the design scheme and code of the thinking framework in combination with the large language model method. When the design scheme needs to be modified, return to Step S3. When the code needs to be modified, return to Step S4 until the debugging result indicates that the thinking framework meets the development requirements; Step S6: Evaluate the thinking process information in combination with the large language model method. When the evaluation fails, check the design scheme and code of the thinking framework in combination with the large language model method. When the design scheme needs to be modified, return to Step S3. When the code needs to be modified, return to Step S4 until the evaluation passes; Step S7: Feedback result information to the user through the interaction interface.

2. The method according to claim 1, wherein The development requirements of the thinking framework include at least one of the following: thinking mode, visualization explanation, technical requirements, performance requirements, security and privacy, example description, application field, output requirements, acceptance criteria.

3. The method according to claim 1, characterized in that, The Step S2 includes: Step S21: Communicate with the user at least once through the interaction interface in combination with the large language model method according to the task information; Step S22: Determine the development requirements of the thinking framework in combination with the large language model method according to the task information and the communication content with the user.

4. The method according to claim 1, characterized in that The Step S3 includes: Step S31: Obtain learning results by using at least one of the following learning methods in combination with the large language model method according to the development requirements of the thinking framework: search the built-in thinking framework knowledge base of the thinking framework development agent, search relevant materials through a search engine, view and learn open source code, learn the existing code and documents of the target agent, and consult data; Step S32: Determine the design scheme of the thinking framework according to the learning results.

5. The method according to claim 1 or 4, characterized in that, The design scheme includes at least one of the following: the design scheme of the system architecture of the thinking framework, the design scheme of each module of the thinking framework, nested logic design, detailed design document, interface design document, database design document, where the nested logic design is the nesting and organizational structure between multiple thinking modes supported by the target agent.

6. The method according to claim 1, wherein The Step S3 includes: Step S33: If the design of the thinking framework does not meet the development requirements, adjust the design scheme of the thinking framework by using at least one of the following adjustment methods in combination with the large language model method: The first adjustment method is to optimize the design solution of the thinking framework; The second adjustment method is to adopt other design solutions.

7. The method according to claim 1, characterized in that, The thinking process information includes at least one of the following: the overall structure of the thinking process, the input and output processed in each step of the thinking process, the prompt words submitted to the large language model in each step, the return results of the large language model in each step, the result information after the execution of the thinking process, and the error information during the execution of the thinking process.

8. The method according to claim 1, characterized in that, The step S5 includes at least one of the following sub-steps: Step S52: When the number of debugging attempts reaches the first preset threshold, or the number of modifications to the design solution reaches the second preset threshold, or the number of modifications to the code of the thinking framework reaches the third preset threshold, and the code of the thinking framework still cannot meet the development requirements, terminate the development task of the thinking framework, and feedback the termination reason and related situation to the user through the interaction interface; Step S53: During the debugging process, record the running logs, intermediate data, and debugging results in combination with the large language model method, and archive and save them.

9. A device for developing an intelligent agent's thinking framework through the computing power of an intelligent computing center, characterized in that, It includes: A receiving module, configured to receive, through the interaction interface, the task information provided by the user for constructing the thinking framework of the target intelligent agent; A requirement determination module, configured to determine the development requirements of the thinking framework in combination with the large language model method according to the task information; A design module, configured to determine the design solution of the thinking framework in combination with the large language model method according to the development requirements; A coding module, configured to generate the code of the thinking framework and the thinking process output code in combination with the large language model method according to the development requirements and the design solution, and the thinking process output code is used to record the thinking process information when the thinking framework solves problems; A debugging module, configured to deploy the running environment of the thinking framework and debug the code of the thinking framework in combination with the large language model method, and output the thinking process information; When the debugging result indicates that the thinking framework does not meet the development requirements, check the design solution and code of the thinking framework in combination with the large language model method. When it is necessary to modify the design solution, trigger the design module to continue working. When it is necessary to modify the code, trigger the coding module to continue working until the debugging result indicates that the thinking framework meets the development requirements; A feedback module, configured to feedback the result information to the user through the interaction interface.

10. 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 developing the thinking framework of the intelligent agent through the computing power of the intelligent computing center as described in any one of claims 1 to 8.

11. 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 developing the thinking framework of the intelligent agent through the computing power of the intelligent computing center as described in any one of claims 1 to 8.

12. 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 developing the thinking framework of the intelligent agent through the computing power of the intelligent computing center as described in any one of claims 1 to 8.

Images