Task processing method and device

Abstract

The embodiment of the invention provides a task processing method and device, computer equipment, a computer readable storage medium, a computer program product and an intelligent agent, and relates to the technical field of computers. The method comprises the steps of determining a target task type according to input target task description; according to the target task description, splitting the target task type into a plurality of target sub-tasks, and determining a target sub-task sequence, the target sub-task sequence comprising a plurality of target sub-tasks; according to the multiple target sub-tasks, multiple segments of target codes are generated, and one target sub-task corresponds to one segment of target code; according to the target subtask sequence, running the multiple segments of target codes to obtain multiple running results; and generating a target task result according to the multiple operation results. According to the technical scheme of the embodiment of the invention, the task result of the target task can be determined according to the results of the multiple sub-tasks, so that the capability of the intelligent agent for solving the complex task is improved, and the task processing efficiency and accuracy of the intelligent agent are improved.

Description

Technical Field

[0001] Embodiments of the present application relate to the field of computer technologies, and in particular, to a task processing method, apparatus, computer device, computer-readable storage medium, computer program product, and agent. Background Art

[0002] With the increasing complexity of the business environment and the continuous growth of data volume, the importance of business analysis in enterprise decision-making has become increasingly prominent. The business analysis process usually covers links such as data research, query, summary, visualization, understanding, and conclusion extraction, and finally forms an analysis report. The manual participation in the business analysis process is high, which not only consumes a large amount of labor costs, but also the accuracy and stability of the analysis results fluctuate greatly.

[0003] With the rapid development of large model technologies, their ability to think deeply and reason provides new ideas for optimizing the business analysis process. However, large models and agents developed based on them still have limitations in business analysis applications, mainly focusing on solving single tasks, and having low efficiency and accuracy in processing complex tasks.

[0004] It should be noted that the above content is not necessarily prior art and is not used to limit the patent protection scope of the present application. Summary of the Invention

[0005] Embodiments of the present application provide a task processing method, apparatus, computer device, computer-readable storage medium, computer program product, and agent to solve or alleviate one or more of the above technical problems.

[0006] One aspect of embodiments of the present application provides a task processing method for an agent based on a large language model. The method includes: Determine a target task type according to the input target task description; Split the target task type into multiple target subtasks according to the target task description, and determine a target subtask sequence, where the target subtask sequence includes multiple target subtasks; Generate multiple segments of target code according to the multiple target subtasks, where one target subtask corresponds to one segment of target code; Run the multiple segments of target code according to the target subtask sequence to obtain multiple running results; Generate a target task result according to the multiple running results.

[0007] Optionally, determining a target task type according to the input target task description includes: Obtain multiple task type-task description pairs, where each task type-task description pair includes a task description and a task type; Determine a target task type-task description pair according to multiple of the task type-task description pairs and the target task description, where the target task description is the task description in the target task type-task description pair; Determine the target task type according to the target task type-task description pair.

[0008] Optionally, determine a target subtask sequence according to the target task and the target task description, including: Determine multiple of the target subtasks and the running order of the multiple target subtasks according to the target task type and the target task description; Determine the target subtask sequence according to the multiple target subtasks and the running order.

[0009] Optionally, run multiple segments of the target code according to the target subtask sequence, including: Determine the running order corresponding to the multiple target subtasks according to the target subtask sequence; Run multiple segments of the target code according to the running order.

[0010] Optionally, generate a target task result according to multiple of the running results, including: Verify whether multiple of the running results are correct; When multiple of the running results are all correct, generate the target task result according to multiple of the running results.

[0011] Optionally, the method further includes: When a target running result among multiple of the running results is incorrect, determine the abnormal target subtask corresponding to the target running result; wherein, the abnormal target subtask is the task that generates the target running result; Generate an update code according to the abnormal target subtask; Run the update code to obtain an update result corresponding to the abnormal target subtask; Generate the target task result according to multiple of the running results verified to be correct and the update result.

[0012] Optionally, the target task result includes a task analysis report; generating a target task result according to multiple of the running results includes: Determine a target report writing habit according to the target task type; Generate the task analysis report according to the target report writing habit and multiple of the running results.

[0013] Another aspect of the embodiments of the present application provides a task processing device, the device comprising: A first determination module, configured to determine a target task type according to an input target task description; A second determination module, configured to split the target task type into a plurality of target subtasks according to the target task description, and determine a target subtask sequence, the target subtask sequence including a plurality of the target subtasks; A first generation module, configured to generate multiple segments of target code according to the plurality of target subtasks, one target subtask corresponding to one segment of the target code; An execution module, configured to run the multiple segments of target code according to the target subtask sequence to obtain a plurality of running results; A second generation module, configured to generate a target task result according to the plurality of running results.

[0014] Another aspect of the embodiments of the present application provides a computer device, comprising: At least one processor; and A memory communicatively connected to the at least one processor; Wherein: the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method as described above.

[0015] Another aspect of the embodiments of the present application provides a computer-readable storage medium, in which computer instructions are stored, and when the computer instructions are executed by a processor, the method as described above is implemented.

[0016] Another aspect of the embodiments of the present application provides a computer program product, comprising a computer program, and when the computer program is executed by a processor, the method as described above is implemented.

[0017] Another aspect of the embodiments of the present application provides an agent based on a large language model, comprising: An identifier, configured to determine a target task type according to an input target task description; A planner, configured to determine a target subtask sequence according to the target task type and the target task description, the target subtask sequence including a plurality of target subtasks; and generate multiple segments of target code according to the plurality of target subtasks, one target subtask corresponding to one segment of the target code; An executor, configured to run the multiple segments of target code according to the target subtask sequence to obtain a plurality of running results; An interpreter, configured to generate a target task result according to the plurality of running results.

[0018] The embodiments of the present application adopting the above technical solutions may include the following advantages: By dividing the recognition (determination), planning (forming a task sequence and generating task-based target code), execution (running the code of the task), and interpretation (generating and outputting the task running result) of tasks in the large language model-based agent, the processing efficiency and accuracy of the agent for responsible tasks are improved. It can be seen that the agent provided in this embodiment can recognize, split, plan, and execute multiple subtasks, and form the final result, which can solve complex tasks, improving the task processing efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings exemplarily show the embodiments and form part of the specification, and are used together with the written description of the specification to explain the exemplary implementation manners of the embodiments. The shown embodiments are only for illustrative purposes and do not limit the scope of the claims. In all the drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0020] Figure 1 Schematically shows the operating environment diagram of the task processing method according to Embodiment 1 of the present application; Figure 2 Schematically shows the flowchart of the task processing method according to Embodiment 1 of the present application; Figure 3 Schematically shows Figure 2 The sub-step flowchart of step S200 in; Figure 4 Schematically shows Figure 2 The sub-step flowchart of step S202 in; Figure 5 Schematically shows Figure 2 The sub-step flowchart of step S206 in; Figure 6 Schematically shows Figure 2 The sub-step flowchart of step S208 in; Figure 7 Schematically shows the new flowchart of the task processing method according to Embodiment 1 of the present application; Figure 8 Schematically shows Figure 2 The sub-step flowchart of step S208 in; Figure 9 Schematically shows the application example flowchart of the task processing method according to Embodiment 1 of the present application; Figure 10 Schematically shows the implementation effect diagram of determining the target task type according to the target task description; Figure 11 Schematically shows the implementation effect diagram of determining the target subtask sequence according to the target task type; Figure 12 Schematically shows the implementation effect diagram of obtaining the running result of the target subtask; Figure 13 Schematically shows the implementation effect diagram of verifying whether the running result is correct; Figure 14 Schematically shows the implementation effect diagram of achieving the target task result; Figure 15 Schematically shows the block diagram of the task processing device according to Embodiment 2 of the present application; Figure 16 Schematically shows the hardware architecture diagram of the computer device according to Embodiment 3 of the present application; Figure 17 Schematically shows the product interface diagram of the large language model-based agent according to Embodiment 6 of the present application; Figure 18 Schematically shows the running effect diagram of the large language model-based agent according to Embodiment 6 of the present application; and Figure 19 Schematically shows the product architecture diagram of the large language model-based agent according to Embodiment 6 of the present application. Detailed implementation manners

[0021] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0022] It should be noted that the descriptions involving "first", "second", etc. in the embodiments of the present application are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the fact that those of ordinary skill in the art can implement them. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present application.

[0023] In the description of the present application, it should be understood that the numerical labels before the steps do not identify the order of execution of the steps, but are only used to facilitate the description of the present application and distinguish each step, so they cannot be understood as a limitation to the present application.

[0024] First, provide the term explanations involved in this application: Large language model: An artificial intelligence model built based on deep learning technology. By learning from a vast amount of text data, it can understand and generate natural language text, and has multiple functions such as language understanding, text generation, and dialogue interaction. It should be noted that the large language model in this embodiment can be a single-modal or multi-modal model.

[0025] Agent: An entity that can perceive the environment, make decisions, and execute actions. It can make autonomous decisions and actions based on environmental information to complete specific tasks. An agent has capabilities such as perception, decision-making, action, and learning, and can operate autonomously to a certain extent.

[0026] Embodied agent: It has a physical entity and can perceive the surrounding environment and interact with it through actuators, such as robots and autonomous driving vehicles. It emphasizes the importance of the interaction between the body and the environment for intelligence and can perform tasks in the real world.

[0027] Virtual agent: It exists in a virtual environment or digital system and is implemented through software. It can understand and process natural language and interact with users, such as intelligent customer service and virtual assistants. Its advantage lies in processing information and providing services.

[0028] AIGC (Artificial Intelligence Generated Content): It uses artificial intelligence technology to automatically generate content such as text, images, and audio through machine learning algorithms and models.

[0029] Data analysis: It refers to the process of collecting, organizing, cleaning, transforming, and modeling data. By applying various statistical and analytical methods, valuable information and knowledge are extracted from a large amount of data to support decision-making, discover patterns, predict trends, and solve problems.

[0030] Data presentation funnel: It refers to the phenomenon that during the process of data from the bottom layer to the final presentation to users, due to multiple layers of processing, screening, and transformation, the amount of information gradually decreases or is distorted.

[0031] User understanding funnel: It is a concept used in information transmission and user behavior analysis, describing the phenomenon that during the process of users contacting, understanding, and finally taking actions, information gradually decreases or is lost.

[0032] User presentation funnel: It refers to the process of information transformation and understanding from professional analysts to users during the information transmission process. During this process, information may be lost or deviated due to factors such as the expression method, user understanding ability, and information presentation form.

[0033] Natural language: It is an unstructured language commonly used by humans in daily life, such as Chinese, English, etc. It is a communication tool that has naturally formed and developed in humans' long-term social life.

[0034] Structured data: It refers to data with a fixed format and a clear organization method, stored in a table form, containing rows and columns. Each column corresponds to an attribute or field, and each row represents a record or instance.

[0035] Knowledge graph: It is a structured semantic knowledge base that represents knowledge in the form of a graph, where nodes represent entities (such as people, places, concepts, etc.) and edges represent the relationships between entities.

[0036] Pattern matching: It is a technique for finding specific patterns or rules in data. By comparing the input data with a predefined pattern, it determines whether there is a matching relationship.

[0037] Mapped dataset: It is a data set that establishes an association between input data and corresponding output data, used to train a model to understand the mapping relationship between input and output, enabling it to accurately predict or generate corresponding output results based on new input data.

[0038] Self-supervision: It is a machine learning technique, which refers to the process in which an agent compares its own output results with preset goals or standards, thereby automatically detecting errors and adjusting its own behavior or strategy.

[0039] SFT (Supervised Fine-Tuning): It is a technique for fine-tuning parameters for a specific task using labeled data based on a pre-trained model.

[0040] Distributed storage system: It is a system that disperses data storage across multiple storage nodes and collaborates through network connections.

[0041] CoT (Chain of Thought): That is, the chain of thought, which is a technique for guiding a model to solve complex problems through step-by-step reasoning.

[0042] DB (Database) knowledge base: It is an integrated and structured information repository for storing content related to databases, including but not limited to database schema information, data dictionaries, data models, database design documents, database performance optimization strategies, database security configurations, database management best practices, database-related technical articles and tutorials, etc.

[0043] DS (Data Science) tool library: It is a tool set for data analysis and processing, providing various functions to help users efficiently perform data operations, including data cleaning, data transformation, data visualization, data modeling, etc.

[0044] SQL (Structured Query Language): It is the standard language for managing and operating relational databases. It provides powerful functions, including operations such as data query, data update, data definition, and data control.

[0045] RAG (Retrieval Augmented Generation) is an artificial intelligence method that combines retrieval technology with a generation model. It guides the generation model by retrieving information from external knowledge sources, thereby generating more accurate and well-founded content.

[0046] API (Application Programming Interface): It is a set of predefined functions, rules, and definitions that allow different software applications to interact and communicate. It provides developers with a standardized way to access and operate the functions of a software system through a specific instruction set.

[0047] Long context: It refers to the ability of a model in natural language processing to process longer text information, being able to refer to extensive previous content to more accurately reply or complete tasks.

[0048] Long conversation: In the field of natural language processing, it means that the model can process and understand longer conversation texts, not only being able to refer to extensive previous content but also accurately reply or complete tasks.

[0049] Interactive report: It is a dynamic report form that allows users to interact with the report content. Users can explore data and information through operations such as clicking, filtering, sorting, and zooming, and delve into specific content according to their own needs and interests.

[0050] Secondly, to facilitate the understanding of the technical solutions provided by the embodiments of the present application by those skilled in the art, the related technologies are described below: In the business data analysis task process, multiple steps such as data research, data query, data cleaning, data mining, data visualization, data analysis, and conclusion extraction are required, and a report is finally formed. This process not only consumes time and effort, which is not conducive to improving human efficiency; but also from the underlying data to the final report presentation, information needs to go through multiple funnels (such as data presentation funnel, user understanding funnel, user presentation funnel), resulting in information loss and low efficiency. In addition, data analysis methods usually rely on static rules, fixed models, or predefined templates, resulting in poor flexibility and adaptability of the analysis, and it is difficult to meet the needs of the dynamically changing business environment. At the same time, because of the manual participation in the analysis reports produced by traditional methods, the complete transmission of information requires long-term learning and running-in.

[0051] Although the large model AIGC technology known to the present inventor performs well in single-task processing, there are still obvious limitations in dealing with complex and multi-step business analysis tasks. First of all, general large models mainly focus on solving single tasks and are difficult to complete the complete business analysis and data analysis report work of multi-steps and multi-modules, especially in ensuring information transfer and mutual call between technical modules. Secondly, in terms of data module arrangement, information transfer, technical module adjustment, etc., there is a lack of effective mechanisms to ensure data consistency and accuracy. Finally, in business scenarios with domain-specific knowledge, it is often difficult to accurately understand and apply the knowledge unique to the business domain, resulting in limited credibility and practicality of the conclusions.

[0052] Therefore, the embodiments of the present application provide a task processing technical solution. In this technical solution, (1) using the code generation ability of the large language model to directly generate code for obtaining data from the database can ensure the integrity and accuracy of data acquisition and reduce the problem of data information loss caused by the data presentation funnel; (2) by executing the update and re-run of the target code when an error is found, the accuracy and reliability of the target task result can be ensured, the analysis depth of the intelligent agent can be improved, as well as the intelligent level and automation degree when dealing with complex tasks; (3) using the deep thinking and reasoning ability of the current large language model, combined with AI Agent (intelligent agent) technology, so as to realize the autonomous arrangement of the target subtask sequence and automatically generate and execute the target code, and at the same time can also intelligently interpret the data to generate the operation result, which helps to improve the output efficiency and data accuracy; (4) using the interpretation ability of the large language model and the professional knowledge base to generate high-quality and highly readable target task results, which can achieve end-to-end automation and reduce the loss caused by information simplification or misunderstanding, ensuring the integrity and accuracy of the target task result. See the following for details.

[0053] Finally, for easy understanding, an exemplary operating environment is provided below.

[0054] Such as Figure 1As shown, the operating environment diagram includes: computer device 2.

[0055] Computer device 2 includes various types of electronic devices, such as portable handheld devices, general-purpose computers (such as personal computers and laptop computers), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, game systems, thin clients, various messaging devices, or other electronic devices, etc. These computer devices can run various types and versions of software applications and operating systems, such as MICROSOFT Windows, APPLE iOS, UNIX-like operating systems, Linux or Linux-like operating systems (such as GOOGLE Chrome OS); or include various mobile operating systems, such as MICROSOFT Windows, Mobile OS, iOS, Windows Phone, Android. Portable handheld devices can include cellular phones, smartphones, tablets, personal digital assistants, etc. Wearable devices can include head-mounted displays (such as smart glasses), etc. Game systems can include various handheld game devices, Internet-enabled game devices, etc. Client devices can execute various different applications, such as various Internet-related applications, communication applications (such as email applications), short message service (SMS) applications, and can use various communication protocols. It should be noted that computer device 2 can also be a server that provides the operation of the agent or a Client / Server (C / S) architecture entity.

[0056] Based on the operating environment described above, computer device 2 can be installed with one or more applications, such as an agent or an interface to access the agent.

[0057] Next, taking computer device 2 as the execution subject, the technical solutions of the present application will be introduced through multiple embodiments. It should be noted that these embodiments can be implemented in many different forms and should not be construed as being limited only to the embodiments described herein.

[0058] Embodiment 1 Figure 2 Schematically shows a flowchart of a task processing method according to Embodiment 1 of the present application.

[0059] As Figure 2 shown, the task processing method can include steps S200 to S208, where: Step S200, determine the target task type according to the input target task description; Step S202: According to the described target task, split the target task type into multiple target subtasks, and determine a target subtask sequence, where the target subtask sequence includes multiple said target subtasks; Step S204: Generate multiple segments of target code according to the multiple said target subtasks, with one said target subtask corresponding to one segment of said target code; Step S206: Run the multiple segments of said target code according to the target subtask sequence to obtain multiple running results; Step S208: Generate a target task result according to the multiple said running results.

[0060] The task processing method provided in this embodiment improves the processing efficiency and accuracy of the intelligent agent for responsible tasks by dividing the identification (determination), planning (forming a task sequence and generating task-based target code), execution (running the code of the task), and interpretation (generating and outputting the task running result) of tasks in the intelligent agent based on the large language model. It can be seen that the intelligent agent provided in this embodiment can identify, split, plan, and execute multiple subtasks, and form a final result, with the ability to solve complex tasks, improving the task processing efficiency and accuracy. By dividing the identification (determination), planning (forming a task sequence and generating task-based target code), execution (running the code of the task), and interpretation (generating and outputting the task running result) of tasks in the intelligent agent based on the large language model, the processing efficiency and accuracy of the intelligent agent for responsible tasks are improved. It can be seen that the intelligent agent provided in this embodiment can identify, split, plan, and execute multiple subtasks, and form a final result, with the ability to solve complex tasks, improving the task processing efficiency and accuracy.

[0061] The following Figure 2 elaborates in detail on each step in steps S200 to S208 and optional other steps.

[0062] Step S200 , determine the target task type according to the input target task description.

[0063] In some embodiments, the target task description can be natural language input by the user, structured data description, or a description method combining natural language and structured data. In other embodiments, the target task description can be text description, voice input, or a combination of text and image. The target task description can include information in multiple dimensions such as time, product, and business. The target task type can be data query and extraction, data analysis and statistics, business process optimization, or decision support and advice, etc.

[0064] In some embodiments, keywords or key phrases can be extracted from the target task description through natural language processing algorithms and the like, and the meaning and intention of the target task description can be understood by combining semantic analysis, so as to determine the target task type according to the understanding result. In other embodiments, the target task type can also be determined by constructing a knowledge graph matching or task template and pattern matching.

[0065] For example, in the case of using a natural language processing algorithm to determine the target task type, if a user inputs a target task description: "Summarize the sales data of the last quarter", which includes the keyword "sales data", the task type can be determined as a "market analysis task".

[0066] Another example, in the case of determining the target task type by pattern matching, if there is a task template "Analyze the sales situation of a certain product in a certain region during a certain period", when the target task description is "Analyze the sales situation of product A of our company in the East China region in the first quarter of 2025", and it matches the template successfully, the task type can be determined as a "product sales analysis task".

[0067] In this embodiment, the true intention is recognized from the input target task description to locate the target task type that needs to be executed. Thus, clear guidance can be provided for subsequent task processing steps, ensuring accurate understanding of user requirements and improving the task execution efficiency and result accuracy of the intelligent agent.

[0068] As described above, there are multiple methods to determine the target task type. The following provides an exemplary determination method.

[0069] In an alternative embodiment, as Figure 3 shown, step S200 includes: S300, obtain multiple task type-task description pairs, each of the task type-task description pairs including a task description and a task type.

[0070] S302, according to the multiple task type-task description pairs and the target task description, determine a target task type-task description pair, the target task description being the task description in the target task type-task description pair.

[0071] S304, according to the target task type-task description pair, determine the target task type.

[0072] The task type-task description pairs can be collected manually or by machine from multiple channels such as historical task records, business requirement documents, user feedback, etc., can also be defined by relevant business personnel, or can be imported industry standards, specification documents, task templates in professional fields, etc.

[0073] The task description stored in the task type-task description pair can be a complete sentence, or the key elements or keywords that the task description should contain, such as business area, analysis object, analysis dimension, output form, etc.

[0074] For example, if the input target task description is "How about the current revenue on the traffic side?", the corresponding target task type-task description pair can be "Conduct a review analysis of the revenue on the traffic side - How about the revenue on the traffic side" or "Conduct a review analysis of the revenue on the traffic side - Traffic side (business area) / Revenue situation (analysis object)".

[0075] In specific implementation, the understanding ability of the large language model can also be utilized to supplement the target task type according to the target task description and context content. For example, the previously mentioned "Conduct a review analysis of the revenue on the traffic side" can be further supplemented with time information and data domain information to obtain the final target task type of "Conduct a review analysis of the commercial traffic side revenue in January 2025".

[0076] In some embodiments, the task type-task description pair can be stored in a database in the form of a mapping data set, and the large language model can be adjusted through the SFT (Supervised Fine-Tuning) technology to enable the large language model to understand the natural language related to the business in the target business description. In other embodiments, a graph database can also be constructed in the form of a knowledge graph to store the task type-task description pair.

[0077] In large-scale applications or high-concurrency scenarios, a distributed storage system (such as the Hadoop Distributed File System HDFS, etc.) can also be used to store a large amount of task type-task description pair data, and the commonly used or hot task type-task description pair data can be cached in memory to improve the data access speed.

[0078] In this embodiment, establishing the correspondence between the task type and the task description and using it to determine the target task type can achieve understanding the natural language in the target task description and accurately classifying and identifying the target task type, providing an accurate basis for subsequent task processing and result generation.

[0079] Step S202 , according to the target task description, split the target task type into multiple target subtasks, and determine the target subtask sequence, where the target subtask sequence includes multiple said target subtasks.

[0080] For example, if the target task type is "Conduct a review analysis of the commercial traffic side revenue in January 2025", the target subtask sequence can be determined as "Analysis of traffic reserve efficiency, Analysis of traffic allocation efficiency, Analysis of traffic monetization efficiency".

[0081] In some embodiments, the deep thinking and CoT (Chain of Thought) technology of large language models can be utilized. Specifically, according to the target task type and the target task description, the dynamic planning and orchestration of the target subtask sequence are autonomously completed through CoT.

[0082] In other embodiments, the knowledge base of a specific business domain and the context content of the target task description can also be combined to help the agent better understand the professional terms and background information in the target task description, so as to determine the target subtask sequence.

[0083] In other embodiments, the ability of the agent to autonomously discover and correct errors can also be cultivated through reinforcement learning, self-supervision, etc. During the determination process of the target subtask sequence, if the agent discovers that there is an error or omission in the setting of a certain target subtask, it can autonomously re-determine part of the target subtasks or the entire target subtask sequence.

[0084] Decompose the target task type to obtain a target task sequence including multiple target subtasks. Thus, complex tasks can be divided into multiple simple task steps, improving the task processing efficiency of the agent and its processing ability for complex tasks.

[0085] As mentioned above, the target subtask sequence can be determined by various methods. The following provides an exemplary method for determining the target subtask sequence.

[0086] In an alternative embodiment, as Figure 4 shown, step S202 includes: S400, according to the target task type and the target task description, determine multiple target subtasks and the running order of multiple target subtasks.

[0087] S402, according to multiple target subtasks and the running order, determine the target subtask sequence.

[0088] In some embodiments, the deep thinking and reasoning ability of large language models and CoT technology can be utilized to determine the running order according to factors such as the complexity of the task and the target requirements. The execution situation of historical similar tasks can also be referred to, and big data analysis technology can be used to mine the rules and patterns therein to determine the running order of multiple target subtasks. The running order can also be determined by analyzing the dependency relationships such as data dependency and logical dependency between target subtasks, and the running order can be adjusted and optimized according to system resources (such as computing resources, storage resources, network bandwidth, etc.).

[0089] In this alternative embodiment, the target task type is decomposed to obtain multiple target subtasks, and the execution order of the multiple target subtasks is determined. Thereby, it is convenient to sequentially execute the multiple target subtasks in the execution order during the subsequent task processing, improving the task processing efficiency of the agent and facilitating ensuring the accuracy of the processing result.

[0090] Step S204 According to the multiple target subtasks, multiple segments of target code are generated, with one target subtask corresponding to one segment of the target code.

[0091] The target code can be database query code (SQL) generated by calling the DB knowledge base and the DS tool library, or code in different languages such as Python, Java, C++ automatically generated according to the requirements of the target subtasks. The target code can be generated according to preset code rules, or the code rules can be adaptively adjusted according to the resource status of the agent running device to generate efficient and low-resource-consuming code.

[0092] In some embodiments, the code generation ability of a single large language model can be used to generate the target code. In other embodiments, the large language model can also be combined with other types of models, such as models dedicated to code generation, models enhanced by specific domain knowledge graphs, etc., to jointly generate the target code. This is not limited here.

[0093] Automatically generating target code for multiple target subtasks by the agent can improve the execution efficiency and accuracy of the target subtasks, enhance the intelligence and automation level of the task processing method, reduce manual intervention, and achieve end-to-end automated task processing. At the same time, using the code generation ability of the large language model to directly generate code for obtaining data from the database can ensure the integrity and accuracy of data acquisition and reduce the problem of data information loss caused by the data presentation funnel.

[0094] Step S206 According to the target subtask sequence, multiple segments of the target code are run to obtain multiple running results.

[0095] In some embodiments, a task dependency graph can be constructed based on the dependencies between multiple target subtasks, and then task scheduling can be performed according to the task dependency graph. An asynchronous programming mechanism or a multithreaded parallel mechanism can also be adopted to run multiple segments of target code simultaneously. When running the target code, an effective memory management strategy can be adopted according to the memory usage. Computational resources can also be dynamically allocated according to the computational intensity and priority of the target subtasks. During the process of running the target code, the automatic error correction ability of the agent can be utilized to monitor the running state of the code in real time and detect errors in a timely manner. Verification mechanisms such as voting and consensus can also be introduced to execute the same task on multiple replicas or nodes, and the final correct result can be determined by comparing the results.

[0096] In some embodiments, based on reinforcement learning, rewards based on code execution success and agent accuracy can be designed to enable the agent to train the ability to automatically generate database query and data analysis code. At the same time, RAG or long-context business knowledge input can also be utilized to help the agent produce analysis results with business value.

[0097] Utilize the deep thinking and reasoning capabilities of the current large language model, combined with AI Agent technology, to autonomously orchestrate the target subtask sequence, automatically generate and execute the target code, and at the same time, can also intelligently interpret data to generate operation results, which helps to improve production efficiency and data accuracy.

[0098] As mentioned above, in some embodiments, the target subtask sequence is also configured with a corresponding running order. The following provides an exemplary method for running the target code when the running order is configured in the target subtask sequence.

[0099] In an alternative embodiment, as Figure 5 shown, step S206 includes: S500, determine the running order corresponding to multiple said target subtasks according to the said target subtask sequence.

[0100] S502, run multiple segments of the said target code according to the running order.

[0101] In some embodiments, an error recovery mechanism can be introduced to rearrange the running order when an error occurs during the running of the target code. In other embodiments, the performance metrics of the agent running device, such as CPU usage, memory occupancy, network latency, etc., can also be monitored in real time, and the running order can be adjusted according to the changes in the performance metrics.

[0102] In this alternative embodiment, running the target code in the running order can ensure the orderly execution of each target subtask, improve the task processing efficiency, reduce human errors at the same time, enhance the accuracy and stability of the task processing result, and increase the automation and intelligence level of the task processing method.

[0103] Step S208 Generate a target task result according to the multiple running results.

[0104] The target task result can be in the form of a text report, a visualization chart, a dynamic picture, a video, etc. With the user's authorization, the output form and content preference of the target task result can also be customized according to the user's needs and preferences.

[0105] After generating the target task result, the target task result can also be output to the user. The user can judge whether to adjust or add the input words entered into the intelligent agent according to the obtained target task result to implement the long conversation function. The user can also give feedback on the target task result, and the intelligent agent can perform iterative optimization based on these feedbacks.

[0106] Utilizing the interpretation ability of the large language model and the professional knowledge base to generate a high-quality and highly readable target task result can achieve end-to-end automation, reduce losses caused by information simplification or misunderstanding, and ensure the integrity and accuracy of the target task result.

[0107] In the actual processing process, various verification or optimization mechanisms can also be added during the generation of the target task result to improve the accuracy of the target task result. The following provides an exemplary verification mechanism.

[0108] In an alternative embodiment, as Figure 6 shown, step S208 includes: S600, verify whether the multiple running results are correct.

[0109] S602, when the multiple running results are all correct, generate the target task result according to the multiple running results.

[0110] In some embodiments, methods such as multi-model cross-verification, adversarial training of large models, and optimization of prompt engineering can be used to improve the intelligent agent's verification ability for the rationality of the results. It is also possible to verify the running results by establishing a verification rule library including rules such as data format verification, logical consistency check, and business rule matching according to the business domain and task type.

[0111] In some embodiments, the current running result can also be compared and analyzed with the running results of historical similar tasks, and the analysis results can be used as a reference for judging the correctness of the current running result.

[0112] Generate the target task result based on the verified correct operation result. By verifying and integrating the operation result, the accuracy and integrity of the target task result are ensured, and the stability and reliability of the task processing method are improved.

[0113] The above describes the situation where the operation results are all verified as correct. When there are operation results verified as incorrect, various methods can also be used for processing.

[0114] The following provides an exemplary processing method.

[0115] In an alternative embodiment, as Figure 7 shown, the method further includes: S700, in the case where the target operation result among multiple said operation results is incorrect, determine the abnormal target subtask corresponding to the target operation result; wherein, the abnormal target subtask is the task that generates the target operation result.

[0116] S702, generate an update code according to the abnormal target subtask.

[0117] S704, run the update code to obtain the update result corresponding to the abnormal target subtask.

[0118] S706, generate the target task result according to multiple said operation results verified as correct and the update result.

[0119] In some embodiments, the correction records in previous similar abnormal situations can be referred to, and the patterns and rules in the correction records can be used to guide the generation of the current update code. In other embodiments, existing code libraries and API documents can also be accessed, and appropriate code snippets or API calls can be matched through semantic understanding to improve the accuracy and efficiency of generating the update code.

[0120] When there are operation results verified as incorrect, generate an update code and run the update code to obtain the update result. By performing the update and re - running of the target code when an error is found, the accuracy and reliability of the target task result can be ensured, the analysis depth of the intelligent agent can be enhanced, as well as the intelligent level and automation degree when dealing with complex tasks.

[0121] As described above, the target task result can have multiple output forms, and there are also different generation methods corresponding to each output form. The following provides an exemplary method for generating the target task result.

[0122] In an alternative embodiment, as Figure 8 shown, the target task result includes a task analysis report, and step S208 includes: S800. Determine the target report writing habit according to the target task type; S802. Generate the task analysis report according to the target report writing habit and multiple operation results.

[0123] The task analysis report can be a text report, a chart report, a presentation, an interactive report, etc. The task analysis report can include detailed data, analysis methods and processes, main conclusions and suggestions of the analysis, etc.

[0124] The target report writing habit can be the report writing habit and format of business experts or business analysts in the business field corresponding to the target task type, such as language style, format layout, content order and hierarchical structure, etc.

[0125] In some embodiments, the method of scoring by business experts plus reinforcement learning training can be used to enable the intelligent agent to obtain the target report writing habit corresponding to the target task type during the training process. In other embodiments, it is also possible to input case processing of the target task type of multiple business experts for the intelligent agent to analyze and summarize, so as to obtain the target report writing habit corresponding to the target task type.

[0126] In other embodiments, the user can also select a preferred report style from a variety of pre-configured optional report styles, or clearly inform the intelligent agent of their report writing habits and preferences in the prompt words input to the intelligent agent, so that the intelligent agent can determine the target report writing habit.

[0127] Generate the task analysis report as the target task result according to the target report writing habit corresponding to the target task type. Thereby, the target task result can be made more in line with the writing habits and formats of professionals in the corresponding business field, thereby improving the readability, usability and reliability of the target task result.

[0128] To make the present application easier to understand, the following is combined with Figures 9 - 14 and Figure 19 Provide an exemplary application.

[0129] In this exemplary application, the intelligent agent is configured with an identifier, a planner, an executor, and an interpreter. Among them: S11. The user inputs a natural language (NL) question "How about the revenue on the traffic side currently?" (i.e., the target task description, Input) to the intelligent agent (i.e., the intelligent agent based on the large language model, InsightAgent); S12. The intelligent agent determines through the identifier (Indentifier) that the task A (Task) to be executed according to the user's description is "Review and analyze the commercial traffic side revenue in January 25" (i.e., the target task type); In S13, the intelligent agent combines the business knowledge base through a Planner to decompose Task A (Task Decomposition), obtaining two subtasks, Subtask1 and Subtask2; In S14, the intelligent agent discovers a problem with the setting of Subtask2. Then it calls the recognizer to re-identify the task and re-partitions the re-identified task, obtaining three target subtasks: analysis of traffic reserve efficiency (Subtask1), analysis of traffic allocation efficiency (Subtask2), and analysis of traffic monetization efficiency (Subtask3) (i.e., the target subtask sequence, Planning Graph); In S15, the intelligent agent calls the DB knowledge base and the DS tool library through an Executor to write code for each target subtask (i.e., the target code, Code); In S16, the intelligent agent runs the code of each target subtask in sequence through an Executor to obtain the analysis results of each target subtask, such as Solution1, Solution2, etc. (i.e., the running results); In S17, the intelligent agent uses a Verifier to verify whether the analysis results of each target subtask are correct and discovers that Solution2 is incorrect. Then it re-executes the target subtask Subtask2, re-codes and executes it to obtain a new Solution2; In S18, the intelligent agent calls the business knowledge base through an Interpreter to summarize and analyze the analysis results of each target subtask and returns the result of "conducting a review and analysis of the commercial traffic-side revenue in January 2025" to the user (i.e., the target task result, Output); In S19, if the user inputs new content (Prompt Tuning) after viewing the result returned by the intelligent agent, then according to the new content input by the user, step S12 is re-executed.

[0130] Embodiment 2 Figure 15 A block diagram of a task processing device according to Embodiment 2 of the present application is schematically shown. The device can be divided into one or more program modules. One or more program modules are stored in a storage medium and executed by one or more processors to complete the embodiments of the present application. The program modules referred to in the embodiments of the present application refer to a series of computer program instruction segments that can complete specific functions. The following description will specifically introduce the functions of each program module in this embodiment. As Figure 15As shown in the figure, the device 1000 may include: a first determination module 1100, a second determination module 1200, a first generation module 1300, a running module 1400, and a second generation module 1500, where: The first determination module 1100 is configured to determine a target task type according to the input target task description. The second determination module 1200 is configured to split the target task type into multiple target subtasks according to the target task description, and determine a target subtask sequence, where the target subtask sequence includes multiple target subtasks. The first generation module 1300 is configured to generate multiple segments of target code according to multiple target subtasks, and one target subtask corresponds to one segment of target code. The running module 1400 is configured to run multiple segments of target code according to the target subtask sequence to obtain multiple running results. The second generation module 1500 is configured to generate a target task result according to multiple running results.

[0131] As an optional embodiment, the first determination module 1100 is further configured to: Obtain multiple task type-task description pairs, where each task type-task description pair includes a task description and a task type. Determine a target task type-task description pair according to multiple task type-task description pairs and the target task description, where the target task description is the task description in the target task type-task description pair. Determine the target task type according to the target task type-task description pair.

[0132] As an optional embodiment, the second determination module 1200 is further configured to: Determine multiple target subtasks and the running order of multiple target subtasks according to the target task type and the target task description. Determine the target subtask sequence according to multiple target subtasks and the running order.

[0133] As an optional embodiment, the running module 1400 is further configured to: Determine the running order corresponding to multiple target subtasks according to the target subtask sequence. Run multiple segments of target code according to the running order.

[0134] As an optional embodiment, the second generation module 1500 is further configured to: Verify whether multiple running results are correct. When multiple said operation results are all correct, generate the target task result according to the multiple said operation results.

[0135] As an optional embodiment, the apparatus 1000 further includes an exception handling module, configured to: When the target operation result among the multiple said operation results is incorrect, determine the abnormal target subtask corresponding to the target operation result; wherein, the abnormal target subtask is the task for generating the target operation result; Generate an update code according to the abnormal target subtask; Run the update code to obtain an update result corresponding to the abnormal target subtask; Generate the target task result according to the multiple operation results verified to be correct and the update result.

[0136] As an optional embodiment, the second generation module 1500 is further configured to: Determine a target report writing habit according to the target task type; Generate the task analysis report according to the target report writing habit and the multiple said operation results.

[0137] Embodiment III Figure 16 Schematically shows a hardware architecture diagram of a computer device 10000 suitable for implementing the task processing method according to Embodiment III of the present application. In some embodiments, the computer device 10000 may be a terminal device such as a smart phone, a wearable device, a tablet computer, a personal computer, a vehicle-mounted terminal, a game console, a virtual device, a workbench, a digital assistant, a set-top box, a robot, etc. In other embodiments, the computer device 10000 may be a rack server, a blade server, a tower server or a cabinet server (including an independent server, or a server cluster composed of multiple servers), etc. As Figure 16 shown, the computer device 10000 includes but is not limited to: a memory 10010, a processor 10020, and a network interface 10030 that can be communicatively linked to each other through a system bus. Among them: The memory 10010 includes at least one type of computer-readable storage medium. The readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (such as SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disc, etc. In some embodiments, the memory 10010 may be an internal storage module of the computer device 10000, such as the hard disk or memory of the computer device 10000. In other embodiments, the memory 10010 may also be an external storage device of the computer device 10000, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. equipped on the computer device 10000. Of course, the memory 10010 may also include both the internal storage module and the external storage device of the computer device 10000. In this embodiment, the memory 10010 is generally used to store the operating system and various application software installed on the computer device 10000, such as the program code of the task processing method, etc. In addition, the memory 10010 can also be used to temporarily store various types of data that have been output or will be output.

[0138] In some embodiments, the processor 10020 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other chip. The processor 10020 is generally used to control the overall operation of the computer device 10000, such as performing control and processing related to data interaction or communication with the computer device 10000. In this embodiment, the processor 10020 is used to run the program code stored in the memory 10010 or process data.

[0139] The network interface 10030 may include a wireless network interface or a wired network interface, which is generally used to establish a communication link between the computer device 10000 and other computer devices. For example, the network interface 10030 is used to connect the computer device 10000 to an external terminal through a network, and establish a data transmission channel and a communication link between the computer device 10000 and the external terminal. The network may be a wireless or wired network such as an enterprise intranet (Intranet), the Internet, Global System of Mobile communication (GSM for short), Wideband Code Division Multiple Access (WCDMA for short), 4G network, 5G network, Bluetooth, Wi-Fi, etc.

[0140] It should be noted that Figure 16 Only the computer device with components 10010 - 10030 is shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively.

[0141] In this embodiment, the task processing method stored in the memory 10010 can also be divided into one or more program modules and executed by one or more processors (such as the processor 10020) to complete the embodiments of this application.

[0142] Embodiment 4 The embodiments of this application also provide a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the task processing method in the embodiment are implemented.

[0143] In this embodiment, the computer-readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disc, etc. In some embodiments, the computer-readable storage medium may be an internal storage unit of a computer device, such as the hard disk or memory of the computer device. In other embodiments, the computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. equipped on the computer device. Of course, the computer-readable storage medium may also include both the internal storage unit and the external storage device of the computer device. In this embodiment, the computer-readable storage medium is generally used to store the operating system and various application software installed on the computer device, such as the program code of the task processing method in the embodiment. In addition, the computer-readable storage medium may also be used to temporarily store various data that have been output or will be output.

[0144] Embodiment Five The embodiment of the present application also provides a computer program product, including a computer program, which implements the method in the above embodiment when executed by a processor.

[0145] Embodiment Six The purpose of this embodiment is to provide an agent product. For specific technical details and effects, reference can be made to, introduced, and combined with Embodiment One.

[0146] The embodiment of the present application provides an agent based on a large language model. The agent can be an embodied agent or a virtual agent, which is not limited in this application. The agent based on the large language model provided by the embodiment of the present application includes: An identifier, configured to determine the target task type according to the input target task description (Task Decompostition).

[0147] A planner, configured to split the target task type into multiple target subtasks (Subtask1, Subtask2,...) according to the target task description, and determine a target subtask sequence, where the target subtask sequence includes multiple target subtasks; and generate multiple segments of target code according to the multiple target subtasks, and one target subtask corresponds to one segment of the target code.

[0148] An Executor, which is used to run multiple segments of the target code according to the target subtask sequence to obtain multiple running results.

[0149] Optionally, a Verifier, which is used to verify whether multiple running results (Solution1, Solution2, …) are correct.

[0150] When multiple running results are all correct, generate the target task result according to the multiple running results.

[0151] When the target running result among multiple running results is incorrect, determine the abnormal target subtask corresponding to the target running result; wherein, the abnormal target subtask is the task that generates the target running result.

[0152] Generate an updated code according to the abnormal target subtask; Run the updated code to obtain the updated result corresponding to the abnormal target subtask; Generate the target task result according to the multiple running results verified to be correct and the updated result.

[0153] An Interpreter, which is used to generate a target task result according to multiple running results.

[0154] The intelligent agent based on the large language model provided in this embodiment can reduce the loss of data information caused by multiple-layer funnels, improve the analysis depth, and achieve efficient data analysis through modular recognizers, planners, executors, interpreters, and verifiers; ensure the reliability and stability of data analysis in a commercial context through the call, loop, and termination strategies of the above modules; and can also utilize the deep thinking and reasoning capabilities of the current large language model, combine with AIAgent (intelligent agent) technology, autonomously arrange and execute tasks, automatically generate and execute code, intelligently interpret data and generate analysis results, improving the output efficiency and data accuracy.

[0155] The schematic diagram of the product interface, the schematic diagram of the running effect, and the schematic diagram of the product architecture of the intelligent agent can be as Figure 17 、 Figure 18 and Figure 19 shown.

[0156] Obviously, those skilled in the art should understand that the various modules or steps of the embodiments of the present application described above can be implemented by general computer devices. They can be concentrated on a single computer device or distributed over a network composed of multiple computer devices. Optionally, they can be implemented with program codes executable by the computer devices, so that they can be stored in a storage device and executed by the computer devices. And in some cases, the steps shown or described can be executed in a sequence different from that here, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module to be implemented. Thus, the embodiments of the present application are not limited to any specific combination of hardware and software.

[0157] It should be noted that the above are only the preferred embodiments of the present application, and do not limit the patent protection scope of the present application. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present application.

Claims

1. A task processing method, characterized in that, In an agent based on a large language model, the method includes: Determine a target task type according to the input target task description; According to the target task description, split the target task type into multiple target subtasks, and determine a target subtask sequence, where the target subtask sequence includes multiple target subtasks; Generate multiple segments of target code according to the multiple target subtasks, with one target subtask corresponding to one segment of target code; Run the multiple segments of target code according to the target subtask sequence to obtain multiple running results; Generate a target task result according to the multiple running results.

2. The method according to claim 1, wherein Determine a target task type according to the input target task description, including: Obtain multiple task type-task description pairs, where each task type-task description pair includes a task description and a task type; According to the multiple task type-task description pairs and the target task description, determine a target task type-task description pair, where the target task description is the task description in the target task type-task description pair; Determine the target task type according to the target task type-task description pair.

3. The method according to claim 1, wherein Determine a target subtask sequence according to the target task and the target task description, including: Determine multiple target subtasks and the running order of the multiple target subtasks according to the target task type and the target task description; Determine the target subtask sequence according to the multiple target subtasks and the running order.

4. The method according to claim 3, wherein Run the multiple segments of target code according to the target subtask sequence, including: Determine the running order corresponding to the multiple target subtasks according to the target subtask sequence; Run the multiple segments of target code according to the running order.

5. The method according to claim 1, wherein Generate a target task result according to the multiple running results, including: Verify whether the multiple running results are correct; When the multiple running results are all correct, generate the target task result according to the multiple running results.

6. The method according to claim 5, wherein The method further includes: When the target running result among the multiple running results is incorrect, determine the abnormal target subtask corresponding to the target running result; where the abnormal target subtask is the task that generates the target running result; Generate update code according to the abnormal target subtask; Run the update code to obtain the update result corresponding to the abnormal target subtask; Generate the target task result according to the multiple running results verified to be correct and the update result.

7. The method according to claim 1, wherein The target task result includes a task analysis report; generating a target task result according to the multiple running results includes: Determine a target report writing habit according to the target task type; Generate the task analysis report according to the target report writing habit and the multiple running results.

8. A task processing device, characterized in that, The device includes: A first determination module for determining a target task type according to the input target task description; A second determination module for splitting the target task type into multiple target subtasks according to the target task description, and determining a target subtask sequence, where the target subtask sequence includes multiple target subtasks; A first generation module, configured to generate multiple segments of target code according to the multiple target subtasks, where one target subtask corresponds to one segment of the target code; A running module, configured to run the multiple segments of target code according to the target subtask sequence to obtain multiple running results; A second generation module, configured to generate a target task result according to the multiple running results.

9. A computer device, characterized in that, Comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein: The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Computer instructions are stored in the computer-readable storage medium, and when the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

12. An agent based on a large language model, characterized in that, Comprising: An identifier, configured to determine a target task type according to an input target task description; A planner, configured to determine a target subtask sequence according to the target task type and the target task description, where the target subtask sequence includes multiple target subtasks; And generate multiple segments of target code according to the multiple target subtasks, where one target subtask corresponds to one segment of the target code; An executor, configured to run the multiple segments of target code according to the target subtask sequence to obtain multiple running results; An interpreter, configured to generate a target task result according to the multiple running results.

Images