Method for realizing large-scale language model training through computing power of intelligent computing center

Abstract

The invention provides a method for realizing large-scale language model training through computing power of an intelligent computing center, and relates to the technical field of computing power infrastructure and artificial intelligence, and the method comprises the following steps: S1, a first agent receives and understands a large-scale language model training task given by a user on a dialogue interface; s2, the first intelligent agent divides the task into a plurality of sub-tasks, and the sub-tasks are distributed to a plurality of second intelligent agents; s3, the first agent determines an execution plan of the task and notifies the second agent to execute the allocated sub-task based on the execution plan; and S4, the first agent or the second agent generates a report result of the task and reports the report result to the user. Due to the adoption of cooperation of multiple agents, the computing power of the intelligent computing center can be fully utilized to process a relatively complex task for training the large language model, and meanwhile, in the process of executing the task for training the large language model, a dialogue interface is adopted to display a communication process and a decision process, so that the processing process is clearer.

Description

Technical Field

[0001] Embodiments of the present invention relate to the fields of computing power infrastructure and artificial intelligence technology, and in particular, to a method for implementing large-scale language model training through the computing power of an intelligent computing center. Background Art

[0002] With the development of artificial intelligence technology and computing power technology, the concept of intelligent computing center has emerged. "Intelligent computing center" refers to the use of large-scale heterogeneous computing power resources, including general computing power and intelligent computing power, mainly for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning and other scenarios) to provide the required computing power, data and algorithms. Intelligent computing center covers facilities, hardware, software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0003] An agent is an agent that can perceive the environment and take actions to achieve specific goals. It can be software, hardware or a system with autonomy, adaptability and interaction. Agents perceive changes in the environment (such as through sensors or data input), make judgments and decisions based on their own learned knowledge and algorithms, and then perform actions to affect the environment or achieve predetermined goals. Agents are widely used in the field of artificial intelligence and are commonly seen in automated systems, robots, virtual assistants and game characters. The core of agents is the ability to learn autonomously and continuously evolve to better complete tasks and adapt to complex environments.

[0004] However, the intelligent agents currently realized through the computing power of intelligent computing centers can usually only complete a simple task alone and find it difficult to complete more complex tasks. Summary of the invention

[0005] The embodiment of the present invention provides a method for implementing large-scale language model training through the computing power of an intelligent computing center, which is used to solve the problem that the existing intelligent agent implemented through the computing power of an intelligent computing center can only complete a simple task alone and has difficulty in completing more complex tasks.

[0006] In order to solve the above-mentioned technical problems, the present invention is achieved as follows: In a first aspect, an embodiment of the present invention provides a method for implementing large-scale language model training through the computing power of an intelligent computing center, which is applied to a multi-agent collaborative system, wherein the multi-agent collaborative system includes a dialogue interface and multiple agents, each of which has specific capabilities, and the multiple agents can form a team and collaborate to complete a specified task in the form of dialogue communication in the dialogue interface, and the dialogue interface is used to display the communication process between the multiple agents in the process of performing the specified task, as well as part or all of the decision-making process of the agents, and the multiple agents include a first agent and multiple second agents; the method includes: Step S1: the first agent receives and understands a specified task given by a user in the dialogue interface, where the specified task is a task for training a large language model; Step S2: the first agent divides the specified task into a plurality of subtasks, and assigns the plurality of subtasks to the plurality of second agents; the plurality of subtasks include at least one of the following: confirming relevant parameters for training the large language model, creating an environment for training the large language model, processing data for training the large language model, training the large language model, evaluating the trained large language model, and sending a report; Step S3: the first agent formulates an execution plan for the designated task and notifies the second agent to execute the assigned subtask based on the execution plan; Step S4: The first agent generates a report result of the specified task based on the training results and evaluation results returned by the second agent and reports it to the user, or instructs the second agent to generate a report result of the specified task and reports it to the user.

[0007] Optionally, step S2 includes: Step S21: The first agent obtains the capabilities of the multiple second agents, divides the designated task into multiple subtasks according to the capabilities of the multiple second agents, and assigns the multiple subtasks to the multiple second agents respectively.

[0008] Optionally, the second agent includes at least one of the following: an agent that confirms relevant parameters for training the large language model, an agent that creates an environment for training the large language model, an agent that processes data for training the large language model, an agent that trains the large language model, an agent that evaluates the trained large language model and sends a report.

[0009] Optionally, the relevant parameters include at least one of the following: a data set used to train the large language model, a purpose of the large language model, an algorithm for training the large language model, a framework for training the large language model, an evaluation index of the large language model, an initial large language model, and computing power required for training the large language model.

[0010] Optionally, also include: Step S5: The second agent confirming the relevant parameters reports the relevant parameters on the dialogue interface; Step S6: The second agent that confirms the relevant parameters receives and understands the user's modification suggestions for the relevant parameters in the dialogue interface, and modifies the relevant parameters based on the modification suggestions.

[0011] Optionally, also include: Step S7: The second agent that creates the environment checks the necessary packages required to run the large language model and the specified task based on the assigned subtasks, and creates a virtual environment for training and evaluating the large language model.

[0012] Optionally, also include: Step S8: The second agent that processes the data checks the format of the data in the data set based on the assigned subtask, performs data format conversion, and divides the data set into a training set and a test set.

[0013] Optionally, also include: Step S9: The second agent that trains the large language model configures a training script based on the assigned subtasks and relevant information provided by other agents, performs training of the large language model according to the training script, and reports the training results.

[0014] Optionally, also include: Step S10: The second agent that evaluates the trained large language model configures an evaluation script based on the assigned subtask and the relevant information provided by other agents, evaluates the trained large language model according to the evaluation script, and reports the evaluation results.

[0015] Optionally, also include: Step S0: The first agent selects at least one second agent from other agents to perform the specified task according to the specified task, and the at least one second agent is all or part of the other agents.

[0016] In an embodiment of the present invention, a multi-agent collaboration system of an intelligent computing center is provided with multiple agents, and the multiple agents can form a team and collaborate to complete the task of training a large language model in the form of dialogue communication in a dialogue interface. Since multiple agents are used for collaboration, the computing power of the intelligent computing center can be fully utilized to handle the more complex task of training a large language model. At the same time, in the process of executing the task of training a large language model, a dialogue interface is used to display the communication process between multiple agents and part or all of the decision-making process of the agents, so that users can have a clearer understanding of the processing process of the specified task, thereby improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 A flowchart of a method for implementing large-scale language model training through the computing power of an intelligent computing center according to an embodiment of the present invention; Figure 2 A schematic diagram of a dialogue interface of a multi-agent collaboration system according to an embodiment of the present invention; Figure 3 is another schematic diagram of a dialogue interface of a multi-agent collaboration system in an embodiment of the present invention; Figure 4 A directed acyclic graph corresponding to multiple agents of a multi-agent collaboration system according to an embodiment of the present invention; Figure 5 A directed acyclic graph corresponding to multiple agents for training a large language model according to an embodiment of the present invention; Figure 6 Another schematic diagram of the dialogue interface of the multi-agent collaboration system in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] First, the technical terms involved in the present invention are briefly explained below.

[0020] The "computing power" mentioned in the present invention is the ability of computer equipment or computing / data center to process information. It is the ability of computer hardware and software to work together to execute certain computing requirements. It is the computing power to achieve target result output by processing information data. It is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to the society through computing power infrastructure.

[0021] The "computing power" (CP) described in the present invention is the ability of a data center server to process data and output results. It is a comprehensive indicator for measuring the computing power of a data center, including general computing power, super computing power, and intelligent computing power. The commonly used unit of measurement is the number of floating point operations performed per second (FLOPS: Floating Point Operations Per Second, 1EFLOPS=10^18 FLOPS). The larger the value, the stronger the comprehensive computing power. According to calculations, 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs or 2 million mainstream notebooks. The calculation formula is: CP=CP 通用 +CP 智能 +CP 超级 .

[0022] The "Network Power" (NP) described in the present invention is a manifestation of the data transmission capability of computing facilities, including comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, etc. The carrying capacity involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling capabilities. In the embodiment of the present invention, the carrying capacity uses the memory bandwidth.

[0023] The "Storage Power" (SP) described in the present invention is the comprehensive ability of a data center in terms of data storage capacity, performance, safety 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 server built-in storage devices. The commonly used unit of measurement for storage capacity is exabyte (EB, 1EB=2^60bytes), and the commonly used unit of measurement for performance is the number of reads and writes per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB). The disaster recovery ratio is an important manifestation of safety and reliability.

[0024] The "computing power infrastructure" described in the present invention is a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity. It can realize centralized computing, storage, transmission and application of information, and presents characteristics such as diversity and ubiquity, intelligence and agility, safety and reliability, green and low-carbon. It is of great significance to promote industrial transformation and upgrading, enable my country's scientific and technological innovation, meet people's better life and realize high-efficiency social governance.

[0025] The “computing power” mentioned in the present invention includes general computing power, intelligent computing power and super computing power.

[0026] The “general computing power” mentioned 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.

[0027] The "intelligent computing power" described in the present invention is a computing platform for large-scale deployment of special chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit) for various innovative applications of artificial intelligence, such as natural language processing and machine vision.

[0028] The "super computing power" mentioned in the present invention is 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 uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for calculations in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0029] The "intelligent computing center" mentioned in the present invention refers to a facility that provides the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning scenarios) by using large-scale heterogeneous computing power resources, including general computing power (CPU: Central Processing Unit) and intelligent computing power (GPU: Graphics Processing Unit, FPGA: Field Programmable Gate Array, ASIC: Application Specific Integrated Circuit, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0030] The "computing resources" mentioned in the present invention refer to technologies and facilities with information calculation, transmission, storage and application capabilities required for the development of a digital society, including but not limited to computing resources such as CPU (Central Processing Unit), GPU (Graphics Processing Unit), 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 supporting and guarantee resources such as wind, fire, water and electricity.

[0031] The agent described in the embodiments of the present invention refers to an agent that can perceive the environment and take actions to achieve specific goals. It can be software, hardware or a system with autonomy, adaptability and interaction capabilities. 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 performs actions to affect the environment or achieve predetermined goals. Agents are widely used in the field of artificial intelligence and are commonly seen in automated systems, robots, virtual assistants, and game characters. The core of agents is the ability to learn autonomously and continuously evolve to better complete tasks and adapt to complex environments.

[0032] The natural language processing (NLP) described in the embodiments of the present invention is a core technology that enables intelligent agents to understand, generate and interact with human natural language. Through NLP, intelligent agents can parse the semantics, grammar and context of human language, so as to perform tasks such as text understanding, language translation, dialogue system, sentiment analysis, etc. This enables intelligent agents to communicate with humans in natural language, understand the user's intentions, and provide corresponding responses and services.

[0033] The “large language model (also referred to as a large language model)” described in the embodiments of the present invention refers to a large language model (LLM), which is a language model with a large parameter scale that is designed to understand and generate human language. It is trained with a large amount of text data and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc.

[0034] The directed acyclic graph described in the embodiment of the present invention means that in graph theory, if a directed graph cannot start from a vertex and return to the vertex through several edges, then the graph is a directed acyclic graph (DAG graph).

[0035] In order to solve the problem that a single existing intelligent agent can only complete a simple task alone and has difficulty in completing more complex tasks, an embodiment of the present invention provides a method for implementing large-scale language model training through the computing power of an intelligent computing center, which is applied to a multi-agent collaborative system. The multi-agent collaborative system includes a dialogue interface and multiple intelligent agents, each of which has specific capabilities. The multiple intelligent agents can form a team and collaborate to complete a specified task in the form of dialogue communication in the dialogue interface. The dialogue interface is used to display the communication process between the multiple intelligent agents in the process of executing the specified task, as well as part or all of the decision-making process of the intelligent agent. The multiple intelligent agents include a first intelligent agent and multiple second intelligent agents; please refer to Figure 1 , the method comprising: Step S1: the first agent receives and understands a specified task given by a user in the dialogue interface, where the specified task is a task for training a large language model; Step S2: the first agent divides the specified task into a plurality of subtasks, and assigns the plurality of subtasks to the plurality of second agents; the plurality of subtasks include at least one of the following: confirming relevant parameters for training the large language model, creating an environment for training the large language model, processing data for training the large language model, training the large language model, evaluating the trained large language model, and sending a report; Step S3: the first agent formulates an execution plan for the designated task and notifies the second agent to execute the assigned subtask based on the execution plan; Step S4: The first agent generates a report result of the specified task based on the training results and evaluation results returned by the second agent and reports it to the user, or instructs the second agent to generate a report result of the specified task and reports it to the user.

[0036] Please refer to Figure 2 , Figure 2 A schematic diagram of a dialogue interface of a multi-agent collaboration system according to an embodiment of the present invention. Figure 2 It can be seen that the multi-agent collaboration system in this embodiment includes multiple agents, including: NANA (project manager), Hu (PM assistant), Wang (operation and maintenance engineer), Doudou (data engineer), Li (large language model training engineer), Guo (large language model evaluation engineer), where each agent is built on the basis of a large language model, and each agent has specific capabilities. For example, "Hu (PM assistant)" can be responsible for confirming necessary parameters and recording parameters for reference by other users. Each agent can communicate with other agents and users (i.e. humans, such as Figure 1In the dialogue interface, the communication process between multiple agents in the process of executing a specified task, as well as part or all of the decision-making process of the agents can be displayed.

[0037] In an embodiment of the present invention, a multi-agent collaboration system of an intelligent computing center is provided with multiple agents, and the multiple agents can form a team and collaborate to complete the task of training a large language model in the form of dialogue communication in a dialogue interface. Since multiple agents are used for collaboration, the computing power of the intelligent computing center can be fully utilized to handle the more complex task of training a large language model. At the same time, in the process of executing the task of training a large language model, a dialogue interface is used to display the communication process between multiple agents and part or all of the decision-making process of the agents, so that users can have a clearer understanding of the processing process of the specified task, thereby improving the user experience.

[0038] In the embodiment of the present invention, optionally, step S2 includes: Step S21: The first agent obtains the capabilities of the multiple second agents, divides the designated task into multiple subtasks according to the capabilities of the multiple second agents, and assigns the multiple subtasks to the multiple second agents respectively.

[0039] In the embodiment of the present invention, optionally, the capabilities of the agent include decision-making capabilities and the ability to use tools. The decision-making capabilities are the decision-making capabilities possessed by the large language model corresponding to the agent. The tools may include the tools of the agent itself, and may also include external tools.

[0040] Please refer to Figure 3 , Figure 3 is another schematic diagram of the dialogue interface of the multi-agent collaboration system in an embodiment of the present invention, Figure 3 As can be seen in FIG. 1 , the first agent (“Nana”) executes the above steps S1 to S4.

[0041] In the embodiment of the present invention, optionally, the second agent includes at least one of the following: an agent that confirms relevant parameters for training the large language model, an agent that creates an environment for training the large language model, an agent that processes data for training the large language model, an agent that trains the large language model, an agent that evaluates the trained large language model and sends a report.

[0042] In an embodiment of the present invention, optionally, the relevant parameters include at least one of the following: a data set used to train the large language model, a purpose of the large language model, an algorithm for training the large language model, a framework for training the large language model, an evaluation index of the large language model, an initial large language model, and computing power required for training the large language model.

[0043] In an embodiment of the present invention, the computing power requirement may include computing power requirements such as the number of GPUs.

[0044] For example, see Figure 2 , Figure 2 In the multi-agent collaborative system shown, "Hu (PM assistant)" is an agent that confirms the relevant parameters for training the large language model, "Wang (operation and maintenance engineer)" is an agent that creates an environment for training the large language model, "Dou Dou (data engineer)" is an agent that processes the data for training the large language model, "Li (large language model training engineer)" is an agent that trains the large language model, and "Guo (large language model evaluation engineer)" is an agent that evaluates the trained large language model and sends a report.

[0045] In an embodiment of the present invention, optionally, the multiple agents are organized via a directed acyclic graph, wherein each node in the directed acyclic graph represents one of the agents, and each edge in the directed acyclic graph represents the communication between two of the agents located at both ends of the edge.

[0046] Please refer to Figure 4 , Figure 4 A directed acyclic graph corresponding to multiple agents of a multi-agent collaboration system of an embodiment of the present invention, wherein the directed acyclic graph represents an organization with an overall goal, includes five nodes (A1, A2, A3, A4 and A5), each node represents an agent with a specific capability, and each edge in the directed acyclic graph represents the communication between two agents located at both ends of the edge, and communication refers to sending information, such as Figure 4 As shown, A1 can communicate with A2, A3, A4, and A5, A2 can communicate with A1 and A4, and so on.

[0047] In the embodiment of the present invention, optionally, in the process of executing the specified task, each of the agents can follow the task instructions given by the agent at the previous level or the user, or give feedback suggestions to the agent at the previous level or the user. Figure 4 , Figure 4 Among them, A1 is the parent agent of A2, A3, A4 and A5, and A2 is the parent agent of A4.

[0048] In the embodiment of the present invention, optionally, agents at the same level can simultaneously execute the assigned tasks, thereby improving the execution efficiency of the tasks. Figure 4 , A2 and A3 are agents of the same level. A1 can assign tasks to A2 and A3, and A2 and A3 can perform the assigned tasks at the same time.

[0049] Please refer to Figure 5 , Figure 5 A directed acyclic graph corresponding to multiple agents for training a large language model according to an embodiment of the present invention.

[0050] In the embodiment of the present invention, optionally, the method further includes: Step S5: The second agent confirming the relevant parameters reports the relevant parameters on the dialogue interface; Step S6: The second agent that confirms the relevant parameters receives and understands the user's modification suggestions for the relevant parameters in the dialogue interface, and modifies the relevant parameters based on the modification suggestions.

[0051] Please refer to Figure 6 , Figure 6 is another schematic diagram of the dialogue interface of the multi-agent collaboration system in an embodiment of the present invention, Figure 6 It can be seen that the second agent ("Hu (PM Assistant)") that confirms the relevant parameters can communicate with the user once or multiple times to modify and determine the relevant parameters.

[0052] Through the second agent that confirms the relevant parameters, the multi-agent collaboration system in the embodiment of the present invention can be applied to users with different professional levels. If the user is a non-professional user, the user only needs to provide simple task requirements, such as: fine-tune the code model (LLM) for me, my data is at a certain location. If the user is a professional user, the user can provide more professional task requirements, such as: fine-tune the code model (LLM) for me, use the DeepSeek model, the training framework should be fire-fly, use lara to train the model, etc. When the second agent that confirms the relevant parameters gives recommended relevant parameters, the user can also communicate with the second agent that confirms the relevant parameters once or multiple times through the dialogue interface to modify the recommended relevant parameters.

[0053] In the embodiment of the present invention, optionally, the method further includes: Step S7: The second agent that creates the environment checks the necessary packages required to run the large language model and the specified task based on the assigned subtasks, and creates a virtual environment for training and evaluating the large language model.

[0054] In the embodiment of the present invention, optionally, the method further includes: Step S8: The second agent that processes the data checks the format of the data in the data set based on the assigned subtask, performs data format conversion, and divides the data set into a training set and a test set.

[0055] In the embodiment of the present invention, optionally, the method further includes: Step S9: The second agent that trains the large language model configures a training script based on the assigned subtasks and relevant information provided by other agents, performs training of the large language model according to the training script, and reports the training results.

[0056] In the embodiment of the present invention, optionally, the method further includes: Step S10: The second agent that evaluates the trained large language model configures an evaluation script based on the assigned subtask and the relevant information provided by other agents, evaluates the trained large language model according to the evaluation script, and reports the evaluation results.

[0057] In the embodiment of the present invention, optionally, the method further includes: Step S0: The first agent selects at least one second agent from other agents to perform the specified task according to the specified task, and the at least one second agent is all or part of the other agents.

[0058] That is, the agents in the team used to complete the specified task are all or part of the multiple agents. In other words, when performing a specified task, not all agents in the system are needed. The agents required to perform the specified task can be determined as needed and formed into a team. For example, Figure 4 The agents A1, A2, and A4 in the system form a team to complete the specified task. It should be noted that the selected agents also need to be organized according to the predetermined directed acyclic graph.

[0059] In the embodiment of the present invention, optionally, each of the intelligent agents can be used to perceive the user's dialogue instructions in real time during the execution of the specified task, and make decisions and take corresponding actions based on the user's dialogue instructions. That is to say, in the embodiment of the present invention, in the process of multiple intelligent agents collaborating to complete the specified task, humans can join in at any time, give suggestions on the execution of the specified task, or terminate the operation of the specified task, etc., with high flexibility. Of course, the user's dialogue instructions may also be unrelated to the specified task. For example, the specified task is to train a large language model. During the training of the large language model, the user can ask a certain intelligent agent in the team (such as Hu (PM Assistant)), such as "Can you sing?" "Hu (PM Assistant)" can respond accordingly based on the user's question without affecting the normal process of training the large language model.

[0060] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0061] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course 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, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0062] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for implementing large-scale language model training through the computing power of an intelligent computing center, characterized in that: Applied to a multi-agent collaboration system, the multi-agent collaboration system includes a dialogue interface and multiple agents, each of the agents has specific capabilities, the multiple agents can form a team and collaborate to complete a specified task in the form of dialogue communication in the dialogue interface, the dialogue interface is used to display the communication process between the multiple agents in the process of performing the specified task, and part or all of the decision-making process of the agents, the multiple agents include a first agent and multiple second agents; the method includes: Step S1: the first agent receives and understands a specified task given by a user in the dialogue interface, where the specified task is a task for training a large language model; Step S2: the first agent divides the specified task into a plurality of subtasks, and assigns the plurality of subtasks to the plurality of second agents; the plurality of subtasks include at least one of the following: confirming relevant parameters for training the large language model, creating an environment for training the large language model, processing data for training the large language model, training the large language model, evaluating the trained large language model, and sending a report; Step S3: the first agent formulates an execution plan for the designated task and notifies the second agent to execute the assigned subtask based on the execution plan; Step S4: The first agent generates a report result of the specified task based on the training results and evaluation results returned by the second agent and reports it to the user, or instructs the second agent to generate a report result of the specified task and reports it to the user.

2. The method according to claim 1, characterized in that The step S2 comprises: Step S21: The first agent obtains the capabilities of the multiple second agents, divides the designated task into multiple subtasks according to the capabilities of the multiple second agents, and assigns the multiple subtasks to the multiple second agents respectively.

3. The method according to claim 1, characterized in that The second agent includes at least one of the following: an agent that confirms relevant parameters for training the large language model, an agent that creates an environment for training the large language model, an agent that processes data for training the large language model, an agent that trains the large language model, an agent that evaluates the trained large language model and sends a report.

4. The method according to claim 3, characterized in that: The relevant parameters include at least one of the following: a data set used to train the large language model, a purpose of the large language model, an algorithm for training the large language model, a framework for training the large language model, an evaluation index of the large language model, an initial large language model, and computing power required for training the large language model.

5. The method according to claim 3 or 4, characterized in that: Also includes: Step S5: The second agent confirming the relevant parameters reports the relevant parameters on the dialogue interface; Step S6: The second agent that confirms the relevant parameters receives and understands the user's modification suggestions for the relevant parameters in the dialogue interface, and modifies the relevant parameters based on the modification suggestions.

6. The method according to claim 3, characterized in that Also includes: Step S7: The second agent that creates the environment checks the necessary packages required to run the large language model and the specified task based on the assigned subtasks, and creates a virtual environment for training and evaluating the large language model.

7. The method according to claim 3, characterized in that Also includes: Step S8: The second agent that processes the data checks the format of the data in the data set, performs data format conversion, and divides the data set into a training set and a test set based on the assigned subtask.

8. The method according to claim 3, characterized in that Also includes: Step S9: The second agent that trains the large language model configures a training script based on the assigned subtasks and relevant information provided by other agents, performs training of the large language model according to the training script, and reports the training results.

9. The method according to claim 3, characterized in that: Also includes: Step S10: The second agent that evaluates the trained large language model configures an evaluation script based on the assigned subtask and the relevant information provided by other agents, evaluates the trained large language model according to the evaluation script, and reports the evaluation results.

10. The method according to claim 1, characterized in that Also includes: Step S0: The first agent selects at least one second agent from other agents to perform the specified task according to the specified task, and the at least one second agent is all or part of the other agents.

Images