Method and device for constructing decentralized agent network by computing power of intelligent computing center cloud platform

Abstract

The invention provides a method and a device for constructing a decentralized agent network by computing power of an intelligent computing center cloud platform, and relates to the technical field of intelligent computing centers, intelligent computing centers and computing power infrastructure, the method is executed by a first agent, and comprises the following steps: S1, sending broadcast information, receiving broadcast information sent by other agents, and sending the broadcast information to the first agent; the broadcast information comprises capability information of an intelligent agent for sending the broadcast information, a routing table maintained by the intelligent agent for sending the broadcast information and an intelligent agent capability table; s2, updating a maintained routing table and an intelligent agent capability table according to the received broadcast information; s3, receiving a user task sent by the edge agent, decomposing the user task into a plurality of sub-tasks, selecting a second agent for executing the sub-tasks according to the maintained agent capability table and the routing table, and distributing the sub-tasks to the second agent; and S4, receiving a subtask execution result of the second agent, generating a task execution result, and sending the task execution result to the edge agent.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent computing centers, intelligent computing centers, and computing power infrastructure, and particularly relates to a method and device for constructing a decentralized intelligent agent network for the computing power of an intelligent computing center cloud platform. 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, and mainly provides the required computing power, data, and algorithms for artificial intelligence applications (such as scenarios for artificial intelligence deep learning model development, model training, and model inference). An intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

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

[0005] An "intelligent computing center", that is, an artificial intelligence computing center, is a type of computing power infrastructure that is based on artificial intelligence theory, adopts an artificial intelligence computing architecture, and provides 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 process information data and output a target result, and a new type of productive force that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to society through computing power infrastructure.

[0007] Intelligent agents based on large language models or multi-modal large models have become the core components of the service ecosystem of intelligent computing centers. Intelligent computing centers provide powerful computing power support and resource scheduling capabilities for intelligent agents. The current intelligent agent technology is evolving from single intelligent agents to multi-agent systems to cope with complex task scenarios and distributed task processing requirements. In existing multi-agent systems, although a multi-agent cooperation mechanism is introduced, a centralized cooperation mechanism is generally adopted to manage the behavior and resource allocation of all intelligent agents. Such a multi-agent system requires the topology graph of intelligent agents to be determined at the system design stage and to be known to all participating intelligent agents. When the number of intelligent agents increases significantly and needs to adapt to complex tasks, the existing solutions lead to a lack of further dynamic expansion ability of the multi-agent system, easily form performance bottlenecks and increase the risk of failures, and limit the flexibility and robustness of the multi-agent system. Summary of the Invention

[0008] The present invention provides a method and device for constructing a decentralized agent network with computing power in an intelligent computing center cloud platform, which is used to solve the problem that the existing multi-agent system adopting a centralized cooperation mechanism lacks further dynamic expansion ability, is prone to form a performance bottleneck and increase the failure risk, and limits the flexibility and robustness of the multi-agent system.

[0009] To solve the above technical problems, the present invention is implemented as follows: In a first aspect, the present invention provides a method for constructing a decentralized agent network with computing power in an intelligent computing center cloud platform, including: Step S1: The first agent sends broadcast information and receives broadcast information sent by other agents. The broadcast information includes the ability information of the agent sending the broadcast information, the routing table and the agent ability table maintained by the agent sending the broadcast information; Step S2: The first agent updates the routing table and the agent ability table maintained by itself according to the received broadcast information; Step S3: When the first agent receives a user task sent by an edge agent, the first agent decomposes the user task into multiple subtasks, selects a second agent for executing each subtask according to the agent ability table maintained by itself, and distributes the subtasks to the second agent according to the routing table maintained by itself; Step S4: The first agent receives the subtask execution results returned by the second agent, generates a task execution result based on all the subtask execution results, and sends the task execution result to the edge agent; Wherein, the routing table includes the third agents that the agent maintaining the routing table can reach, and the forwarding paths between the third agents, and the agent ability table includes the ability information of the third agents.

[0010] Optionally, the first agent includes at least one of the following: An agent deployed inside the intelligent computing center; An agent deployed outside the intelligent computing center; And / or, the other agents include at least one of the following: An agent deployed in a first intelligent computing center, where the first intelligent computing center is the intelligent computing center to which the first agent belongs; An agent deployed in a second intelligent computing center; An agent deployed outside the intelligent computing center.

[0011] Optionally, the step S3 includes: Step S31: When there are multiple candidate agents with the same or similar capabilities in the agent capability table maintained by the first agent, obtain the comprehensive value scores of the multiple candidate agents, and select a second agent to execute the subtask from the multiple candidate agents according to the comprehensive value scores of the multiple candidate agents; Among them, the comprehensive value score is determined based on at least one of the following target metrics: computing power node performance, network performance metric, historical performance metric, load status metric, cost-benefit metric.

[0012] Optionally, the comprehensive value score is the weighted sum of at least two of the target metrics.

[0013] Optionally, step S3 further includes: Step S32: The first agent collects the target metrics of each agent in the agent capability table it maintains, and calculates the comprehensive value scores of each agent in the agent capability table it maintains according to the collected target metrics; Or, step S33: The first agent receives the comprehensive value scores reported by other agents.

[0014] Optionally, before step S1, it further includes: Step S0: The first agent determines the capability information; where when the first agent is an agent for task decomposition and allocation in the first intelligent computing center, the capability information determined by the first agent includes the capability information of all agents that can be aggregated by the first intelligent computing center to which the first agent belongs.

[0015] In a second aspect, the present invention provides a device for constructing a decentralized agent network for computing power in an intelligent computing center cloud platform, including: A transceiver module, configured to send broadcast information and receive broadcast information sent by other agents, where the broadcast information includes the capability information of the agent sending the broadcast information, the routing table and the agent capability table maintained by the agent sending the broadcast information; An update module, configured to update the routing table and the agent capability table maintained by itself according to the received broadcast information; A scheduling module, configured to, when receiving a user task sent by an edge agent, decompose the user task into multiple subtasks, and select a second agent to execute each subtask according to the agent capability table maintained by itself, and distribute the subtasks to the second agent according to the routing table maintained by itself; A feedback module, configured to receive the subtask execution results returned by the second agent, generate a task execution result based on all the subtask execution results, and send the task execution result to the edge agent; Among them, the routing table contains third agents that the agent maintaining the routing table can reach, and forwarding paths between the third agents, and the agent capability table contains the capability information of the third agents.

[0016] In a third aspect, the present invention provides an electronic device, including: a processor, a memory, and a program stored on the memory and executable on the processor. When the program is executed by the processor, it implements the steps of the method for constructing a decentralized agent network with computing power of the intelligent computing center cloud platform as described in the first aspect above.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the method for constructing a decentralized agent network with computing power of the intelligent computing center cloud platform as described in the first aspect above.

[0018] In a fifth aspect, the present invention provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, it implements the steps of the method for constructing a decentralized agent network with computing power of the intelligent computing center cloud platform as described in the first aspect above.

[0019] In the present invention, through a dynamic routing mechanism, the agents discover and connect with each other, thereby dynamically constructing a multi-agent system with a decentralized network. Agents can join or leave the network at any time without knowing the global topology in advance, enabling the multi-agent system to have the ability of dynamic expansion, reducing performance bottlenecks and failure risks, and greatly improving the flexibility and robustness of the multi-agent system. At the same time, through the self-organizing cooperation mode between agents, the decomposition of complex tasks and the allocation of computing power resources are realized, so as to efficiently complete the processing of complex tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings: Figure 1 It is a schematic flowchart of the method for constructing a decentralized agent network with computing power of the intelligent computing center cloud platform of the present invention; Figure 2 It is a schematic structural diagram of the multi-agent system of the present invention; Figure 3 It is a schematic structural diagram of the device for constructing a decentralized agent network with computing power of the intelligent computing center cloud platform of the present invention; Figure 4 Schematic diagram of the structure of the electronic device of the present invention. Detailed implementation manners

[0021] The technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0022] First, the technical terms related to the present invention will be briefly described below.

[0023] The "computing power" referred to in the present invention means: the ability of a computer device or a computing / data center to process information, the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement, the computing ability to process information data and output a target result, a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, and mainly provides services to society through computing power infrastructure.

[0024] The "computational power" (Computational Power, CP) referred to in the present invention means: the ability of a data center server to process data and output a result, a comprehensive index 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 通用 + CP 智能 + CP 超级 .

[0025] The "carrying capacity" (Network Power, NP) referred to in the present invention means: the performance 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 index for measuring network transmission scheduling ability.

[0026] The "Storage Power" (SP) described in the present invention refers to the comprehensive ability of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and built-in storage devices of servers. The commonly used measurement unit for storage capacity is exabyte (EB, 1EB = 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.

[0027] The "computing power infrastructure" described in the present invention refers to a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage power, and can realize the centralized computing, storage, transmission, and application of information.

[0028] The "new type of information infrastructure" described in the present invention mainly includes network infrastructures such as 5G networks, fiber broadband networks, backbone networks, international communication networks, and satellite Internet, computing power infrastructures such as data centers, general computing power centers, intelligent computing centers, and supercomputing centers, and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0029] The "computing power" described in the present invention includes general computing power, intelligent computing power, and super computing power.

[0030] 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.

[0031] The "intelligent computing power" described in the present invention refers to a computing platform that is scaled up for various artificial intelligence innovation applications based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit), such as natural language processing, machine vision, and so on.

[0032] The "super computing power" described in the present invention mainly refers to the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and processes extremely complex or data-intensive problems through a dedicated operating system. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, gene analysis, and so on.

[0033] The "Intelligent Computing Center" described in the present invention refers to a facility that uses large-scale heterogeneous computing power resources, including general computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.), and 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). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enabling.

[0034] The "Intelligent Computing Center Cloud Platform" described in the present invention refers to a cloud computing platform that comprehensively serves based on the hardware resources and software resources of the intelligent computing center.

[0035] The "Intelligent Computing Center" described in the present invention includes, but is not limited to, the "Intelligent Computing Center".

[0036] The "Intelligent Computing Center" described in the present invention, namely the artificial intelligence computing center, is a type of computing power infrastructure that is based on artificial intelligence theory, adopts an artificial intelligence computing architecture, and provides computing power services, data services, and algorithm services required for artificial intelligence applications.

[0037] The "Computing Power Center" described in the present invention refers to a facility mainly composed of infrastructure such as wind, fire, water, and electricity and IT software and hardware devices, and has computing power, transportation power, and storage power, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0038] The "Supercomputing Center" described in the present invention, namely the supercomputing data center, is a data center based on supercomputers or large-scale computing clusters, which can provide 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.

[0039] The "Computing Power Resources" described in the present invention refers to technologies and facilities with information computing, transmission, storage, and application capabilities required for the development of the digital society, including, but not limited to, computing resources such as CPU and GPU, 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.

[0040] The "Computing Power Node" described in the present invention refers to the computing resources of servers / containers that can process computing tasks.

[0041] The "Large Language Model" described in the present invention refers to the 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.

[0042] The "Multimodal Large Models" described in the present invention refer to models that jointly train multimodal information such as text, images, videos, and audio, including but not limited to multimodal large language models.

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

[0044] To solve the problem that the existing multi-agent system adopting a centralized cooperation mechanism lacks further dynamic expansion ability, is prone to form a performance bottleneck and increase the risk of failure, and limits the flexibility and robustness of the system, please refer to Figure 1 , the present invention provides a method for constructing a decentralized agent network for the computing power of an intelligent computing center cloud platform, and the method includes: Step S1: The first agent sends broadcast information and receives the broadcast information sent by other agents. The broadcast information includes the ability information of the agent sending the broadcast information, the routing table maintained by the agent sending the broadcast information, and the agent ability table. In the present invention, optionally, the first agent includes at least one of the following: an agent deployed inside the intelligent computing center; an agent deployed outside the intelligent computing center. That is to say, the first agent can be an agent deployed inside the intelligent computing center or an agent deployed outside the intelligent center.

[0045] The first agent sends broadcast information to other agents. The broadcast information sent by the first agent includes: the ability information of the first agent, the routing table maintained by the first agent, and the agent ability table. Among them, the routing table sent by the first agent contains the third agent that the first agent can reach and the forwarding path between the first agent and the third agent, and the agent ability table sent by the first agent contains the ability information of the third agent.

[0046] In the present invention, the capability information of the intelligent agent is stored in a preset data table in the form of a structured description statement. The capability range of the intelligent agent covers two levels: basic function capabilities and business application capabilities. The basic function capabilities include, but are not limited to: document parsing and processing capabilities, code generation and analysis capabilities, image recognition and processing capabilities, and multi-modal information processing capabilities. The business application capabilities include, but are not limited to: specialized processing capabilities for specific business scenarios such as life scenarios and financial data scenarios.

[0047] In the present invention, the third intelligent agent refers to the destination node in the routing table, and the forwarding path between the first intelligent agent and the third intelligent agent refers to the forwarding path for the information sent by the first intelligent agent to reach the third intelligent agent.

[0048] Optionally, the first intelligent agent periodically sends broadcast information. The sending period can be set as needed.

[0049] The first intelligent agent can also receive broadcast information sent by other intelligent agents. The broadcast information sent by other intelligent agents includes: the capability information of the other intelligent agents, the routing table and the intelligent agent capability table maintained by the other intelligent agents. Among them, the routing table sent by other intelligent agents contains the third intelligent agents that the other intelligent agents can reach, and the forwarding paths between the other intelligent agents and the third intelligent agents. The intelligent agent capability table sent by other intelligent agents contains the capability information of the third intelligent agents.

[0050] In the present invention, the third intelligent agent refers to the destination node in the routing table, and the forwarding path between other intelligent agents and the third intelligent agent refers to the forwarding path for the information sent by other intelligent agents to reach the third intelligent agent.

[0051] In the present invention, the first intelligent agent receives broadcast information sent by adjacent other intelligent agents. It should be noted that in the present invention, adjacent intelligent agents can be connected by wire or by wireless means.

[0052] Optionally, the other intelligent agents periodically send broadcast information. The sending period can be set as needed. Further optionally, the sending periods of each intelligent agent (i.e., including the above-mentioned first intelligent agent and other intelligent agents) can be the same.

[0053] In the present invention, optionally, the other intelligent agents include at least one of the following: An intelligent agent deployed in the first intelligent computing center, where the first intelligent computing center is the intelligent computing center to which the first intelligent agent belongs; An intelligent agent deployed in the second intelligent computing center; An intelligent agent deployed outside the intelligent computing center.

[0054] The routing table in the present invention can be seen as shown in Table 1, and Table 1 is the routing table maintained by Agent A.

[0055] Table 1

[0056] The agent capability table in the present invention can be seen as shown in Table 2, and Table 2 is the agent capability table maintained by Agent A.

[0057] Table 2

[0058] Agent A can be the above-mentioned first agent or other agents mentioned above.

[0059] In the present invention, as shown in Table 3, the routing table and the agent capability table can also be integrated into one table.

[0060] Table 3

[0061] It should be noted that the formats of the above routing table and agent capability table are only examples and can also be other formats.

[0062] Step S2: The first agent updates the routing table and the agent capability table it maintains according to the received broadcast information; In the present invention, after the first agent receives the broadcast information sent by other agents, it can query whether the identities of the other agents are included in the routing table and the agent capability table it maintains. If not, it is necessary to add the identities of the other agents to the routing table and the agent capability table, and at the same time generate a forwarding path between the first agent and the other agents in the routing table, and record the capability information of the other agents in the agent capability table. In addition, it is also necessary to determine, according to the routing tables of all the other agents, the agents that the first agent can reach through the relay of the other agents, and the forwarding paths between the first agent and the agents that can be reached through the relay of the other agents, and update them to the routing table maintained by the first agent. At the same time, update the capability information of the agents that can be reached through the relay of the other agents to the agent capability table maintained by the first agent.

[0063] If the identities of the other agents are included in the routing table and the agent capability table it maintains, the first agent also needs to update the forwarding paths between itself and the other agents and the forwarding paths between itself and the agents that can be reached through the relay of the other intelligent machines in the routing table it maintains. At the same time, it is also necessary to update the capability information of the other agents and the agents that can be reached through the relay of the other agents recorded in the agent capability table it maintains.

[0064] So far, through the above dynamic routing mechanism, the agents can discover and connect with each other, thereby dynamically constructing a multi-agent system with a decentralized network.

[0065] Please refer to Figure 2 , Figure 2 , which is a schematic structural diagram of the multi-agent system of the present invention. Figure 2 In, each agent (A, B, C...) maintains its own routing table and agent capability table respectively, and periodically sends broadcast information. The agent that receives the broadcast information needs to update the routing table and agent capability table it maintains based on the received broadcast information to complete the discovery and connection between agents. The above first agent can be Figure 2 any agent in.

[0066] Figure 2 In, the intelligent computing center 1 and the intelligent computing center 2 respectively include a plurality of internal agents. At the same time, they can also include aggregable external agents. Among them, the agent K is an aggregable external agent of the intelligent computing center 1, and the agents I and J are aggregable external agents of the intelligent computing center 2. The "aggregable external agent" described in the present invention refers to an agent deployed outside the intelligent computing center and capable of collaborating with the agents inside the intelligent computing center to complete tasks.

[0067] The above agents include three main modules. One module is the routing module, which is used to send and receive broadcast information and maintain the routing table; the capability information module, which is used to determine its own capability information and maintain the agent capability table; the task execution module, which is used to execute tasks. In addition, some agents can also have a task decomposition module for performing task decomposition.

[0068] Step S3: When the first agent receives the user task sent by the edge agent, decompose the user task into multiple subtasks, and according to the agent capability table maintained by itself, select the second agent to execute each subtask, and distribute the subtasks to the second agent according to the routing table maintained by itself; It should be noted that the first agent in the present invention is an agent with the ability of task decomposition and computing power resource allocation, and can be any agent in the above constructed decentralized network. Preferably, considering security and computing power, it can be an agent inside the intelligent computing center. Further preferably, it is an agent inside the intelligent computing center that plays a role in connecting with external agents, such as Figure 2 agents A and F in. In the present invention, optionally, an agent for task decomposition and computing power resource allocation can be set for each intelligent computing center.

[0069] In the present invention, the edge agent can be deployed at the network edge or on the user side, and is responsible for the access and preliminary processing of user tasks.

[0070] In an alternative manner, when the edge agent receives a user task sent by a user, it can directly forward the user task to the first agent. In this case, the edge agent can select the first agent in various ways. For example, it can select the agent within the nearest intelligent computing center as the first agent, or select the agent within the intelligent computing center with the lowest cost as the first agent, etc.

[0071] In another alternative manner, when the edge agent receives a user task sent by a user, it can preprocess the user task. The preprocessing, for example, evaluates the capabilities required by the user task, so as to select the first agent adapted to the capabilities of the user task, or select an intelligent computing center adapted to the capabilities of the user task. Or, the user task can also be initially decomposed.

[0072] In the present invention, the edge agent needs to query the routing table maintained by itself, and send the user task to the first agent according to the forwarding path between the first agent selected and itself queried from the routing table.

[0073] In the present invention, the first agent can decompose the user task into multiple subtasks that can be independently executed. The multiple subtasks can include multiple subtasks that can be executed simultaneously, or can also include subtasks that need to wait for other subtasks to be executed before they can be executed. The first agent determines the agent capabilities required for each subtask, and selects a second agent for each of the subtasks from the maintained agent capability table. Among them, each second agent can execute one subtask or multiple subtasks. The first agent distributes the subtasks to the corresponding second agents according to the maintained routing table.

[0074] Step S4: The first agent receives the subtask execution results returned by the second agent, generates a task execution result based on all the subtask execution results, and sends the task execution result to the edge agent; According to the specific situation, the first agent can merge the subtask execution results returned by multiple second agents to obtain the task execution result, or analyze the subtask execution results returned by multiple second agents to obtain the task execution result.

[0075] The first agent sends the task execution result to the edge agent according to the query routing table and the forwarding path between the first agent and the edge agent in the routing table.

[0076] Among them, the routing table (including the routing tables of the above first agent and other agents) contains the third agents that the agent maintaining the routing table can reach, and the forwarding paths between the agent and the third agents, and the agent capability table contains the capability information of the third agents.

[0077] In the present invention, through a dynamic routing mechanism, the agents discover and connect with each other, thereby dynamically constructing a multi-agent system with a decentralized network. Agents can join or leave the network at any time without the need to know the global topology structure, enabling the multi-agent system to have the ability of dynamic expansion, reducing performance bottlenecks and failure risks, and greatly improving the flexibility and robustness of the multi-agent system. At the same time, through the self-organizing cooperation mode between agents, the decomposition of complex tasks and the allocation of computing power resources are realized, so as to efficiently complete the processing of complex tasks.

[0078] It should be noted that the agents in the present invention are agents based on large language models or multi-modal large models. The term "based on large language models or multi-modal large models" means that by organizing input information and designing prompt words, large language models or multi-modal large models are called to perform tasks such as reasoning and thinking, design, dialogue with users, generation of solution designs, and / or code generation.

[0079] In the present invention, optionally, the step S3 includes: Step S31: When the agent capability table maintained by the first agent contains multiple candidate agents with the same or similar capabilities, obtain the comprehensive value scores of the multiple candidate agents, and select a second agent to execute the subtask from the multiple candidate agents according to the comprehensive value scores of the multiple candidate agents; wherein, the comprehensive value scores are determined based on at least one of the following target indicators: computing power node performance, network performance indicators, historical performance indicators, load status indicators, cost-benefit indicators.

[0080] Among them, the computing power node performance may include at least one of the following: hardware indicators such as CPU performance, GPU performance, memory capacity, and storage speed; The network performance indicators may include at least one of the following: connection quality indicators such as access latency, bandwidth capacity, and network stability; The historical performance indicators may include at least one of the following: historical task completion success rate, historical task processing quality, etc.; The load status indicators may include at least one of the following: load conditions, available resource margins, etc.; The cost-benefit indicators may include at least one of the following: usage cost, cost performance of expected benefits, etc.

[0081] In the present invention, optionally, the comprehensive value score is the weighted sum of at least two of the above target metrics, that is, the comprehensive value of the agent is evaluated from multiple dimensions.

[0082] In the present invention, optionally, the agent with the highest comprehensive value score is selected as the second agent.

[0083] In the present invention, optionally, the process of selecting the second agent may be carried out for a certain subtask. For example, for a certain subtask, multiple candidate agents with the same or similar capabilities are queried from the agent capability table. At this time, the agent with the highest comprehensive value score needs to be selected as the second agent.

[0084] The process of selecting the second agent may also be carried out for multiple subtasks simultaneously. For example, when multiple subtasks all require the execution of the same or similar agents, and multiple candidate agents with the same or similar capabilities are queried from the agent capability table, and the number of candidate agents is more than the number of subtasks. At this time, multiple agents with the highest comprehensive value scores can be selected from the queried candidate agents as the second agents, and are respectively used to process the multiple subtasks.

[0085] In the present invention, optionally, step S3 further includes: Step S32: The first agent collects the target metrics of each agent in the agent capability table it maintains, and calculates the comprehensive value scores of each agent in the agent capability table it maintains according to the collected target metrics. In the present invention, optionally, the first agent may periodically collect the target metrics of each agent in the agent capability table it maintains, calculate the comprehensive value scores of each agent in the agent capability table it maintains, and record the comprehensive value scores of each agent in the agent capability table it maintains, so as to facilitate querying.

[0086] The first agent may also, when a user task needs to be executed, collect the target metrics of each agent in the agent capability table it maintains in real time, and calculate the comprehensive value scores of each agent in the agent capability table it maintains, so that the collected target metrics are more real-time.

[0087] In the present invention, optionally, step S3 further includes: Step S33: The first agent receives the comprehensive value scores reported by other agents.

[0088] In the present invention, other agents may collect their own respective target metrics by themselves and calculate their own comprehensive value scores based on the respective target metrics.

[0089] In the present invention, each agent can periodically send its respective comprehensive value score, for example, send the comprehensive value score through broadcast information. Each agent can also report its respective comprehensive value score based on the reporting request sent by the first agent.

[0090] In the present invention, optionally, before the step S1, the method further includes: Step S0: The first agent determines the capability information; wherein, when the first agent is an agent for performing task decomposition and allocation within the first intelligent computing center, the capability information determined by the first agent includes the capability information of all agents that can be aggregated by the first intelligent computing center to which the first agent belongs.

[0091] For example, please refer to Figure 2 , assuming that the first agent is agent A, the capability information determined by agent A includes the capability information of all agents that can be aggregated by intelligent computing center 1, that is, the capability information of intelligent computing center 1 as a whole, including the capability information of agents A, B, C, and K. Assuming that the first agent is agent F, the capability information determined by agent F includes the capability information of all agents that can be aggregated by intelligent computing center 2, that is, the capability information of intelligent computing center 2 as a whole, including the capability information of agents F, G, H, I, and J.

[0092] However, the capabilities determined by other agents except for the agents for performing task decomposition and allocation are their own capabilities. For example, Figure 2 the capability information determined by agent B in

[0093] Please refer to Figure 3 , the present invention also provides a device 10 for constructing a decentralized agent network for the computing power of an intelligent computing center cloud platform, including: A transceiver module 11, configured to send broadcast information and receive broadcast information sent by other agents, where the broadcast information includes the capability information of the agent sending the broadcast information, the routing table and the agent capability table maintained by the agent sending the broadcast information; An update module 12, configured to update the routing table and the agent capability table maintained by itself according to the received broadcast information; A scheduling module 13, configured to, when receiving a user task sent by an edge agent, decompose the user task into multiple subtasks, and select a second agent for executing each subtask according to the agent capability table maintained by itself, and distribute the subtasks to the second agent according to the routing table maintained by itself; A feedback module 14, configured to receive the sub-task execution results returned by the second agent, generate a task execution result based on all the sub-task execution results, and send the task execution result to the edge agent; Wherein, the routing table includes third agents that can be reached by the agent maintaining the routing table, and forwarding paths between the agent and the third agents, and the agent capability table includes the capability information of the third agents.

[0094] In the present invention, through a dynamic routing mechanism, the agents discover and connect to each other, thereby dynamically constructing a multi-agent system with a decentralized network. Agents can join or leave the network at any time without prior knowledge of the global topology, enabling the multi-agent system to have dynamic expansion capabilities, reducing performance bottlenecks and failure risks, and greatly improving the flexibility and robustness of the multi-agent system. At the same time, through the self-organizing cooperation mode between agents, the decomposition of complex tasks and the allocation of computing power resources are realized, so as to efficiently complete the processing of complex tasks.

[0095] Optionally, the first agent includes at least one of the following: An agent deployed inside the intelligent computing center; An agent deployed outside the intelligent computing center; And / or, the other agents include at least one of the following: An agent deployed in the first intelligent computing center, where the first intelligent computing center is the intelligent computing center to which the first agent belongs; An agent deployed in the second intelligent computing center; An agent deployed outside the intelligent computing center.

[0096] Optionally, the scheduling module 13 includes: A selection sub-module, configured to, when there are multiple candidate agents with the same or similar capabilities in the agent capability table maintained by the first agent, obtain the comprehensive value scores of the multiple candidate agents, and select a second agent to execute the sub-task from the multiple candidate agents according to the comprehensive value scores of the multiple candidate agents; Wherein, the comprehensive value score is determined based on at least one of the following target metrics: computing power node performance, network performance metric, historical performance metric, load status metric, cost-benefit metric.

[0097] Optionally, the comprehensive value score is the weighted sum of at least two of the target metrics.

[0098] Optionally, the scheduling module 13 includes: A calculation sub-module, configured to collect the target metrics of each agent in the agent capability table it maintains, and calculate the comprehensive value score of each agent in the agent capability table it maintains according to the collected target metrics; Alternatively, a receiving sub-module, configured to receive the comprehensive value scores reported by other agents.

[0099] Optionally, the device further includes: A determination module, configured to determine capability information; wherein, when the first agent is an agent for task decomposition and allocation in a first intelligent computing center, the capability information determined by the first agent includes the capability information of all agents that can be aggregated by the first intelligent computing center to which the first agent belongs.

[0100] Please refer to Figure 4 , the present invention also provides an electronic device 20, including a processor 21, a memory 22, and a computer program stored on the memory 22 and executable on the processor 21. When the computer program is executed by the processor 21, it implements each process of the method embodiment of constructing a decentralized agent network for the computing power of the intelligent computing center cloud platform as described above, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0101] 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 of constructing a decentralized agent network for the computing power of the intelligent computing center cloud platform as described above, 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.

[0102] The present invention also provides a computer program product, including computer instructions, which implement each process of the method embodiment of constructing a decentralized agent network for the computing power of the intelligent computing center cloud platform as shown above when the computer instructions are executed by a processor, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here. Figure 1 and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0103] It should be noted that in this text, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of additional identical elements in the process, method, article or device comprising that element.

[0104] From the description of the above embodiments, those skilled in the art can clearly understand that the above-described embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on this understanding, the technical solution of the present invention, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc) and includes several instructions for causing a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0105] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the claims of the present invention, and all of them fall within the protection scope of the present invention.

Claims

1. A method for constructing a decentralized intelligent agent network with computing power in an intelligent computing center cloud platform, characterized in that, Including: Step S1: The first agent sends broadcast information and receives broadcast information sent by other agents. The broadcast information includes the capability information of the agent sending the broadcast information, the routing table maintained by the agent sending the broadcast information, and the agent capability table. Step S2: The first agent updates the routing table and the agent capability table it maintains according to the received broadcast information. Step S3: When the first agent receives a user task sent by an edge agent, it decomposes the user task into multiple subtasks, selects a second agent to execute each subtask according to the agent capability table it maintains, and distributes the subtasks to the second agent according to the routing table it maintains. Step S4: The first agent receives the subtask execution results returned by the second agent, generates a task execution result based on all the subtask execution results, and sends the task execution result to the edge agent. Wherein, the routing table includes the third agents that the agent maintaining the routing table can reach, and the forwarding paths between the third agents. The agent capability table includes the capability information of the third agents.

2. The method according to claim 1, wherein: The first agent includes at least one of the following: An agent deployed inside the intelligent computing center; An agent deployed outside the intelligent computing center; And / or, the other agents include at least one of the following: An agent deployed in the first intelligent computing center, where the first intelligent computing center is the intelligent computing center to which the first agent belongs; An agent deployed in the second intelligent computing center; An agent deployed outside the intelligent computing center.

3. The method according to claim 1, wherein Step S3 includes: Step S31: When there are multiple candidate agents with the same or similar capabilities in the agent capability table maintained by the first agent, obtain the comprehensive value scores of the multiple candidate agents, and select a second agent to execute the subtask from the multiple candidate agents according to the comprehensive value scores of the multiple candidate agents. Wherein, the comprehensive value score is determined based on at least one of the following target metrics: computing power node performance metric, network performance metric, historical performance metric, load status metric, cost-benefit metric.

4. The method according to claim 3, characterized in that The comprehensive value score is the weighted sum of at least two of the target metrics.

5. The method according to any one of claims 3 or 4, characterized in that, Step S3 further includes: Step S32: The first agent collects the target metrics of each agent in the agent capability table it maintains, and calculates the comprehensive value scores of each agent in the agent capability table it maintains according to the collected target metrics. Or, Step S33: The first agent receives the comprehensive value scores reported by other agents.

6. The method according to claim 1, characterized in that, Before Step S1, it further includes: Step S0: The first agent determines the capability information. Wherein, when the first agent is an agent for task decomposition and allocation in the first intelligent computing center, the capability information determined by the first agent includes the capability information of all agents that can be aggregated by the first intelligent computing center to which the first agent belongs.

7. An apparatus for constructing a decentralized intelligent agent network with computing power in a cloud platform of an intelligent computing center, characterized in that, Comprising: A transceiver module, configured to send broadcast information and receive broadcast information sent by other agents, where the broadcast information includes the capability information of the agent sending the broadcast information, the routing table maintained by the agent sending the broadcast information, and the agent capability table. An update module, configured to update the routing table and the agent capability table maintained by itself according to the received broadcast information. A scheduling module, configured to, when receiving a user task sent by an edge agent, decompose the user task into multiple subtasks, select a second agent for executing each subtask according to the agent capability table maintained by itself, and distribute the subtasks to the second agent according to the routing table maintained by itself. A feedback module, configured to receive the execution results of the subtasks returned by the second agent, generate a task execution result based on the execution results of all subtasks, and send the task execution result to the edge agent. Wherein, the routing table includes the third agents that the agent maintaining the routing table can reach, and the forwarding paths between the third agents, and the agent capability table includes the capability information of the third agents.

8. An electronic device, characterized in that, Comprising: 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 constructing a decentralized agent network with the computing power of the intelligent computing center cloud platform according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the steps of the method for constructing a decentralized agent network with the computing power of the intelligent computing center cloud platform according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Including computer instructions, where when the computer instructions are executed by a processor, it implements the steps of the method for constructing a decentralized agent network with the computing power of the intelligent computing center cloud platform according to any one of claims 1 to 6.

Images