A task execution method based on a large language model and related equipment

Abstract

This application relates to the field of artificial intelligence technology, and provides a task execution method and related equipment based on a large language model. The method includes: decomposing a complex task into multiple subtasks using a task decomposition model, and constructing an execution model for each subtask; planning all subtasks using a task planning model to obtain the logical execution order of all subtasks; calibrating each subtask to obtain multiple final subtasks, and then combining all final subtasks according to the execution order to obtain a combined task; and executing the combined task using all execution models to obtain the final execution result of the complex task. The method of this application can improve the versatility of task execution methods.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a task execution method and related equipment based on a large language model. Background Technology

[0002] Earth science research and geographical knowledge discovery are highly complex and multidimensional tasks. Solving related problems requires processing massive amounts of data and performing statistical information mining on complex issues. Large Language Models (LLMs) encode vast amounts of human language, including Earth science and geographical knowledge, thereby enabling human-like understanding and reasoning abilities for complex tasks. Simultaneously, AI agents driven by LLMs exhibit more automated task understanding and reasoning. Therefore, automated Earth science research and geographical knowledge discovery are possible through LLM-driven AI agents. However, due to the complexity of Earth science research and geographical knowledge discovery, and the high degree of complexity in most related tasks, current LLM-driven AI agents suffer from poor generalizability in their task execution methods when applied to geographical science exploration and discovery tasks. Summary of the Invention

[0003] This application provides a task execution method and related equipment based on a large language model, which can solve the problem of poor versatility of task execution methods.

[0004] In a first aspect, embodiments of this application provide a task execution method based on a large language model, the task execution method comprising:

[0005] The task decomposition model is used to break down a complex task into multiple subtasks and to build an execution model for each subtask; the task decomposition model is a large language model capable of task decomposition.

[0006] The task planning model is used to plan all subtasks and obtain the execution order of all subtasks; the task planning model is a large language model capable of subtask planning.

[0007] Each subtask is calibrated to obtain multiple final subtasks. Then, all final subtasks are combined in the order of execution to obtain a combined task. The combined task includes all planned final subtasks as well as the combination reasoning rules and combination reasoning methods between all final subtasks. The combination reasoning rules are used to describe the order in which all final subtasks are executed, and the combination reasoning methods are methods to optimize the combination reasoning rules.

[0008] By utilizing all execution models to perform combined tasks, the final execution result of complex tasks can be obtained.

[0009] Optionally, a task decomposition model can be used to decompose a complex task into multiple subtasks, including:

[0010] Through the formula:

[0011] LLDM(I)={t1,t2,...,t n}

[0012] Obtain multiple subtasks {t1, t2, ..., t n};

[0013] Where LLDM() represents the task decomposition model, I represents the complex task, t1 represents the first subtask, t2 represents the second subtask, and t n This represents the nth subtask, where n is the total number of subtasks.

[0014] Optionally, an execution model is built for each subtask, including:

[0015] Through the formula:

[0016] LLMM={(t1,M1),(t2,M2),...,(t n M n )}

[0017] Obtain the combination of each subtask and the execution model {(t1,M1),(t2,M2),...,(t n M n )};

[0018] Where LLMM represents the modeling model, M1 represents the execution model of the first subtask, M2 represents the execution model of the second subtask, and M... n This represents the execution model for the nth subtask.

[0019] Optionally, a task planning model can be used to plan all subtasks to obtain the execution order of all subtasks, including:

[0020] Through the formula:

[0021] LLPM(M,E,G)={d1,d2,...,d n}

[0022] Obtain the execution order {d1,d2,...,d n};

[0023] Where LLPM() represents the planning model, M represents all execution models, E represents the task working environment of the complex task, G represents the task objective of the complex task, d1 represents the first executed subtask, d2 represents the second executed subtask, and d n This indicates the third subtask to be executed.

[0024] Optionally, each subtask can be calibrated to obtain multiple final subtasks, including:

[0025] Through the formula:

[0026] LLCM(D,R C E)={c1,c2,...,c n}

[0027] Obtain the set of final subtasks {c1, c2, ..., c n};

[0028] Where LLCM() represents the calibration model, D represents the calibration data, and R C This represents the task calibration rule, E represents the task objective error and deviation constraints, c1 represents the final subtask corresponding to the first subtask, c2 represents the final subtask corresponding to the second subtask, and c... n This represents the final subtask corresponding to the nth subtask.

[0029] Optionally, a combined task can be performed using all execution models to obtain the final execution result of the complex task, including:

[0030] Through the formula:

[0031] LLEM(D,R e ,R m )={d1,d2,...,d n}

[0032] Get the final execution result {d1,d2,...,d n};

[0033] Where LLEM() represents the combined model, D represents the input data, and R represents the input data. e Represents the rules of combinatorial reasoning, R m Let d1 represent the execution result of the first final subtask executed in the combined task, and d2 represent the execution result of the second final subtask executed in the combined task. n This represents the execution result of the nth final subtask in a combined task.

[0034] Secondly, embodiments of this application provide a task execution device based on a large language model, comprising:

[0035] The decomposition module uses a task decomposition model to break down complex tasks into multiple subtasks and builds an execution model for each subtask; the task decomposition model is a large language model capable of task decomposition.

[0036] The planning module uses a task planning model to plan all subtasks and obtain the execution order of all subtasks; the task planning model is a large language model capable of subtask planning.

[0037] The calibration module calibrates each subtask to obtain multiple final subtasks, and then combines all the final subtasks in the execution order to obtain a combined task. The combined task includes all the planned final subtasks and the combination reasoning rules and combination reasoning methods between all the final subtasks. The combination reasoning rules are used to describe the execution order of all the final subtasks, and the combination reasoning method is a method to optimize the combination reasoning rules.

[0038] The execution module utilizes all execution models to perform combined tasks and obtain the final execution result of complex tasks.

[0039] Thirdly, embodiments of this application provide a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned task execution method based on a large language model.

[0040] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned task execution method based on a large language model.

[0041] The above-mentioned solution in this application has the following beneficial effects:

[0042] In the embodiments of this application, a complex task to be executed is decomposed into multiple subtasks using a task decomposition model, and an execution model is constructed for each subtask. A task planning model is then used to plan the execution order of all subtasks. Each subtask is then calibrated to obtain multiple final subtasks. These final subtasks are then combined according to their execution order to obtain a combined task. Finally, the combined task is executed using all execution models to obtain the final execution result of the complex task. Decomposing a complex task into multiple subtasks transforms a complex task into multiple simpler subtasks, effectively reducing the difficulty of executing the complex task. Planning all subtasks improves the logic and rationality of their execution. Combining and executing all final subtasks with lower execution difficulty based on a logically sound execution order enhances the rationality and versatility of the task execution method.

[0043] Other beneficial effects of this application will be described in detail in the following detailed description section. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A flowchart illustrating a task execution method based on a large language model provided in an embodiment of this application;

[0046] Figure 2 This is a schematic diagram of the structure of a task execution system based on a large language model provided in an embodiment of this application;

[0047] Figure 3 A schematic diagram illustrating a first final execution result provided in an embodiment of this application;

[0048] Figure 4 A schematic diagram illustrating a second final execution result provided in an embodiment of this application;

[0049] Figure 5 A schematic diagram illustrating a third final execution result provided in an embodiment of this application;

[0050] Figure 6 This is a schematic diagram of the structure of a task execution device based on a large language model provided in an embodiment of this application;

[0051] Figure 7 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0052] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0053] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0054] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0055] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0056] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0057] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0058] To address the poor versatility of existing task execution methods, this application provides a task execution method based on a large language model. This method decomposes a complex task into multiple subtasks using a task decomposition model, constructs an execution model for each subtask, and then plans the execution order of all subtasks using a task planning model. Each subtask is then calibrated to obtain multiple final subtasks. These final subtasks are then combined according to their execution order to form a combined task. Finally, the combined task is executed using all execution models to obtain the final execution result of the complex task. Decomposing the complex task into multiple subtasks effectively reduces the difficulty of executing the complex task. Planning all subtasks improves the logic and rationality of their execution. Combining and executing all final subtasks with lower execution difficulty based on a logically sound execution order enhances the rationality and versatility of the task execution method.

[0059] The following is an illustrative example of the task execution method based on a large language model provided in this application.

[0060] like Figure 1 As shown, the task execution method based on a large language model provided in this application includes the following steps:

[0061] Step 11: Use the task decomposition model to decompose the complex task to be executed into multiple subtasks, and build an execution model for each subtask.

[0062] The task decomposition model described above is a large language model capable of task decomposition. The complex task described above is a user-input task that needs to be performed, such as drawing a distribution map of the elderly population.

[0063] The first step is to use a task decomposition model to break down the complex task to be executed into multiple subtasks.

[0064] Specifically, through the formula:

[0065] LLDM(I)={t1,t2,...,t n}

[0066] Obtain multiple subtasks {t1, t2, ..., t n}

[0067] Where LLDM() represents the task decomposition model, I represents the complex task, t1 represents the first subtask, t2 represents the second subtask, and t n This represents the nth subtask, where n is the total number of subtasks.

[0068] For example, a large language model, through natural language processing technology, can understand complex task descriptions input by users and break them down into multiple specific, executable subtasks. Inputting a complex task into a task decomposition model yields multiple subtasks. For instance, for the task of mapping the distribution of the elderly population mentioned above, inputting this task into the task decomposition model results in the following subtasks: "Subtask 1: Data Collection + Collecting geospatial data on the distribution of the elderly population (over 65 years old), including elderly population data results," "Subtask 2: Data Preprocessing + Cleaning the collected data, including removing missing or outlier data points," "Subtask 3: Spatial Distribution Analysis + Analyzing the geospatial distribution of sampling points, including calculating point density and identifying hotspot areas," and "Subtask 4: Map Drawing + Using geographic information system software or libraries to draw a distribution map of sampling points, marking different densities and hotspot areas."

[0069] The second step is to build an execution model for each subtask.

[0070] Specifically, through the formula:

[0071] LLMM={(t1,M1),(t2,M2),...,(t n M n )}

[0072] Obtain the combination of each subtask and the execution model {(t1,M1),(t2,M2),...,(t n M n )}.

[0073] Where LLMM represents the modeling model, M1 represents the execution model of the first subtask, M2 represents the execution model of the second subtask, and M... n This represents the execution model for the nth subtask.

[0074] For example, the modeling model is a large language model capable of model building. Based on the natural language understanding ability of the large language model for task modeling, contrastive learning is designed to enable the large language model to understand tools or execute model descriptions, thereby stimulating the task modeling ability of the large model. Each sub-task is input into the modeling model, which can match the sub-task's task objectives and descriptions in a model space that stores various models and tools (such as object detection models, drawing tools, and recognition models). The matched models and tools are then used as the execution models for the sub-tasks. The execution model is a model or tool that can achieve the task objectives of a subtask. For example, for "Subtask 1: Data Acquisition + Collecting geospatial data on the distribution of the elderly population (over 65 years old), including elderly population data results," the modeling model performs model matching in the model space based on the task objectives and description of Subtask 1. The matched execution model for Subtask 1 is a data acquisition tool (such as an online database). For "Subtask 2: Data Preprocessing + Cleaning the collected data, including removing missing or outlier data points," the modeling model performs model matching in the model space based on the task objectives and description of Subtask 2. The matched execution model for Subtask 2 is a data acquisition tool (such as an online database). According to the cleaning tools (such as data processing scripts); for "Subtask 3: Spatial Distribution Analysis + Analyze the geospatial distribution of sampling points, including calculating point density and identifying hotspot areas", the modeling model performs model matching in the model space according to the task objectives and description of Subtask 3, and the matching results in the execution model of Subtask 3, which is the spatial analysis model (such as geostatistical methods); for "Subtask 4: Map Drawing + Use Geographic Information System (GIS) software or libraries to draw a distribution map of sampling points, marking different densities and hotspot areas", the modeling model performs model matching in the model space according to the task objectives and description of Subtask 4, and the matching results in the execution model of Subtask 4, which is the GIS.

[0075] In some embodiments of this application, since it is necessary to instruct a large language model, complex tasks input by the user can exist in the form of task instructions in an instruction space used to present the task objectives and content of the complex task. In the instruction space, the user's task instructions are in the form of natural language data. For example, for the task of drawing a distribution map of the elderly population, the task instructions in the instruction space are: "There is a task of drawing a distribution map of the elderly population that needs to be performed. In this task, the distribution of geospatial sampling points is calculated and drawn. Then, the task decomposition, task modeling, task planning, task calibration and task combination are carried out. Finally, the distribution map of the elderly population is drawn by combining drawing tools. The result is described as + <interpret this distribution map>. Please analyze this specific task step by step according to the analysis paradigm."

[0076] It is worth mentioning that breaking down complex tasks into multiple subtasks can transform complex tasks into multiple simple subtasks, effectively reducing the difficulty of performing complex tasks.

[0077] Step 12: Use the task planning model to plan all subtasks and obtain the execution order of all subtasks.

[0078] The task planning model described above is a large language model capable of subtask planning, such as the LLM+P model.

[0079] Specifically, through the formula:

[0080] LLPM(M,E,G)={d1,d2,...,d n}

[0081] Obtain the execution order {d1,d2,...,d n}

[0082] Where LLPM() represents the planning model, M represents all execution models, E represents the task working environment of the complex task, G represents the task objective of the complex task, d1 represents the first executed subtask, d2 represents the second executed subtask, and d n This indicates the third subtask to be executed.

[0083] It should be noted that by inputting the execution models of all subtasks, the task working environment of complex tasks, and task objectives into the planning model, the execution order of each execution model is obtained. Based on this order, the subtasks corresponding to each execution model are planned to obtain the execution sequence. The aforementioned task working environment refers to the constraints that the user comprehensively considers during task execution. These constraints may include resource limitations, time limitations, geographical location, external interference factors, etc., all of which are environmental variables that the user needs to consider during task planning. For example, in a distribution map drawing task, the task working environment may include the map area corresponding to the distribution map pre-set by the user, the time corresponding to the relevant data obtained, etc.

[0084] The task objective is the result of executing a complex task, such as drawing a distribution map as described above, where the task objective is the distribution map itself.

[0085] For example, a large language model automatically understands the task objectives and the working environment to adaptively establish the relationships between subtasks, forming the execution order of the subtasks. For example, for subtask 1, subtask 2, and subtask 3, the large language model believes that subtask 2 should be executed before subtask 1 and subtask 3 should be executed last. Therefore, the first subtask to be executed is subtask 2, the second subtask to be executed is subtask 1, and the third subtask to be executed is subtask 3.

[0086] In some embodiments of this application, after obtaining the execution sequence, the execution sequence can be sent to the user's device. After receiving the confirmation instruction returned by the user's device, the next step is performed. If no confirmation instruction is received from the user's device, it means that the execution sequence has not met the user's expectations. If the user believes that the execution sequence is unreasonable, the planning model adjusts the order of subtasks in the execution sequence.

[0087] It is worth mentioning that planning all subtasks improves the logic and rationality of executing them.

[0088] Step 13: calibrate each subtask to obtain multiple final subtasks, and then combine all the final subtasks in the execution order to obtain a combined task.

[0089] The aforementioned combined tasks include all planned final subtasks and the combined reasoning rules and methods between them. The combined reasoning rules describe the execution order of all final subtasks, determining their order, dependencies, priorities, etc., thus avoiding conflicts and improving execution efficiency. The combined reasoning methods are those that optimize the combined reasoning rules, such as logical reasoning, heuristic algorithms, and user-interaction-dependent scheduling methods, optimizing the combined tasks and resource allocation to ensure they are completed in an orderly, efficient, and correct manner.

[0090] The first step is to calibrate each subtask to obtain multiple final subtasks.

[0091] Specifically, through the formula:

[0092] LLCM(D,R C E)={c1,c2,...,c n}

[0093] Obtain the set of final subtasks {c1, c2, ..., c n}

[0094] Where LLCM() represents the calibration model, D represents the calibration data, and R C This represents the task calibration rule, E represents the task objective error and deviation constraints, c1 represents the final subtask corresponding to the first subtask, c2 represents the final subtask corresponding to the second subtask, and c... n This represents the final subtask corresponding to the nth subtask.

[0095] It should be noted that the calibration data mentioned above is pre-stored data for calibration. Calibration data can be stored in storage devices such as data space. Task calibration will establish a monitoring mechanism for target errors and deviations in task modeling and planning based on task judgment rules and the rationality of task judgment, effectively eliminating the illusion problem caused by large language models. Each sub-task is input into the calibration model, and the calibration model calibrates the sub-tasks to obtain the final sub-task. Calibration data is used to adjust and correct the standard dataset or benchmark information output by the task model. This data is used to compare and calibrate the model's output, ensuring the accuracy and consistency of task results. By comparing the task results with the calibration data, errors generated by the model are identified and corrected. This ensures consistency of results at different stages or during different task executions. Task calibration rules refer to specific methods or guidelines for calibrating the task execution process according to specific standards and requirements. These rules guide how to use calibration data to correct tasks and ensure that tasks meet expected goals after calibration. They provide clear processes and standards for the application of calibration data, ensuring the effectiveness of calibration. Through clear rules, the influence of subjective factors in the calibration process is reduced, ensuring more reliable results. Task target deviation refers to the difference or deviation between the actual result and the expected target during task execution. It is a core element of task calibration, determining the accuracy of task execution by identifying and measuring deviations. By calculating the target deviation, the gap between the task result and the expectation can be assessed, thus understanding the accuracy of task execution. After identifying the deviation, the required calibration or adjustment can be determined based on its magnitude and nature. Deviation constraints refer to the limitations set during task calibration to ensure that the final task result is within an acceptable range. These conditions define the acceptable range of task target deviation; adjustments are required when this range is exceeded. By setting constraints, it is ensured that the error of the task result does not exceed the acceptable range. Providing clear objectives for the calibration process makes the calibration work more targeted and effective.

[0096] For example, in a task to map the distribution of the elderly population, suppose we need to plot the locations of the elderly population within a specific region to optimize sampling efficiency and resource allocation. Below are examples of specific calibration data, task calibration rules, task objective bias, and bias constraints:

[0097] Calibration data includes actual geographic information data, historical sampling point location data, and regional population density data. These are used to compare the sampling point distribution depicted in the model with the actual situation, correcting any discrepancies.

[0098] Task calibration rules: such as calibrating the spatial location of sampling points based on actual geographic information to ensure that sampling points are not located in unreasonable places (such as water bodies, mountains, etc.), and calibrating the distribution density of sampling points based on population density.

[0099] Task target deviation: The difference between the distance between the locations of elderly population distribution points plotted by the model and their actual reasonable locations, or the difference between the number of elderly population distribution points set by the model and the regional needs. The smaller the deviation, the more reasonable the distribution points of the elderly population.

[0100] Deviation constraints: The actual location deviation of each elderly population distribution point must not exceed 50 meters, or the distribution density deviation must not exceed 5%, to ensure that the final elderly population distribution points are reasonable and feasible. If these limits are exceeded, the model needs to be recalibrated.

[0101] Through these specific data, rules, biases, and constraints, the distribution map drawing task can be accurately calibrated to ensure that the final result meets actual needs and reduces unnecessary errors and waste of resources.

[0102] Task calibration can also include human-computer interactive calibration. Human calibration involves human experts or users reviewing, adjusting, and correcting the results generated by the model. It combines human experience and judgment, compensating for the shortcomings of automated machine calibration, and ensuring that the task results are not only technically reasonable but also meet practical needs. Human calibration includes expert review, user feedback, and manual correction. For example, in distribution mapping, experts can review and adjust the locations of elderly population distribution points, and user feedback can optimize the distribution of points, thereby ensuring more accurate and practical results.

[0103] For example, the calibration model described above is a large language model that can perform task calibration, such as a large language model with a self-refine algorithm.

[0104] The second step is to combine all the final subtasks in the order of execution to obtain the combined task.

[0105] Specifically, based on the planning of all subtasks in the execution order and the correspondence between subtasks and the final subtask, the final subtasks are combined sequentially according to the execution order. The combination model provides combination reasoning rules and methods. Specifically, the combination model is a large language model capable of task combination, such as a large language model with algorithms like Tree of Thought (ToT). The execution order and all final subtasks are input into the combination model. The combination model plans the final subtasks based on the execution order and, by analyzing the task objectives and descriptions of the final subtasks, provides combination reasoning rules and methods.

[0106] For example, if the execution sequence is: subtask 2, subtask 1, subtask 3, then the order of the final subtasks in the combined task is final subtask 2, final subtask 1, final subtask 3. In the above task of drawing a nucleic acid sampling point distribution map, the combined reasoning rules and combined reasoning methods are as follows.

[0107] In the above task of drawing the distribution map of nucleic acid sampling points, the combinatorial reasoning rule is the order in which subtasks 1, 2, 3, and 4 are executed after calibration. If the task is executed in the order of subtasks 1, 2, 3, and 4 after calibration, the combinatorial reasoning method can be a user-interactive scheduling method. The combinatorial reasoning rule is optimized by interacting with the user so that it meets the user's expectations.

[0108] It is worth mentioning that by calibrating subtasks, the accuracy and rationality of subtasks can be improved, and the deviation of subtasks can be reduced.

[0109] Step 14: Execute the combined task using all execution models to obtain the final execution result of the complex task.

[0110] Specifically, through the formula:

[0111] LLEM(D,R e ,R m )={d1,d2,...,d n}

[0112] Get the final execution result {d1,d2,...,d n}

[0113] Where LLEM() represents the combined model, D represents the input data, and R represents the input data. e Represents the rules of combinatorial reasoning, R m Let d1 represent the execution result of the first final subtask executed in the combined task, and d2 represent the execution result of the second final subtask executed in the combined task. n This represents the execution result of the nth final subtask in a combined task.

[0114] It should be noted that the input data mentioned above can be data related to complex tasks obtained using data acquisition tools (such as online databases), or it can be data provided by the user. When obtaining the final execution result, each final subtask is executed sequentially according to the final subtask in the combined task, as well as the combined inference rules and methods, using the execution model corresponding to the final subtask.

[0115] During user-system interaction, users can manually assess the system's rationality. The large language model is invoked by matching the description of the task tool, and users issue commands to the model to execute it via natural language. Combinatorial reasoning rules and methods are input into the system through user interaction, and the system automatically understands these rules and methods.

[0116] For example, the final execution result is the result of executing a complex task. After obtaining the final execution result, it can be sent to the user's device. For instance, if the complex task is to draw a distribution map of the elderly population in a certain region, the final execution result would be as follows: Figure 3 The image shown is a map of elderly population density in a certain region based on simulated data, along with its interpretation. Figure 3 The explanation provided is: This is a scatter plot of the elderly population generated based on simulated data, showing the latitude and longitude and the density of the elderly population in different regions. The gray level and size of each scatter point reflect the density of the elderly population at that location. Figure 3 The horizontal axis represents longitude, the vertical axis represents latitude, and the gray level of the scatter plots represents the percentage of the elderly population density in the region. Through this map, we can observe the distribution pattern of the elderly population in the region. This visualization tool is of great value for understanding the geographical distribution of the elderly population and guiding public health decisions and resource allocation.

[0117] It is worth mentioning that breaking down complex tasks into multiple subtasks can transform complex tasks into simpler ones, effectively reducing the difficulty of executing complex tasks. Planning all subtasks improves the logic and rationality of their execution. Combining and executing all final subtasks with low execution difficulty based on a logically ordered sequence can enhance the rationality and versatility of the task execution method.

[0118] In some embodiments of this application, the above steps can be integrated into the same system to obtain a task execution system based on a large language model, such as... Figure 2As shown, the system includes an instruction space, a task space, a model space, and a data space. The instruction space receives complex tasks and instructions sent by the user, containing task instructions described in natural language. This space aims to present task requirements in an intuitive and easy-to-understand natural language format, enabling users and the system to clearly understand the task objectives and content. The task space executes steps 11 to 14. The model space stores various models and tools. The data space stores and manages open and private data, responsible for collecting, storing, processing, and managing various data required for task execution, ensuring data security and availability, and providing necessary data support for task execution. The specific structure of the task space includes a task decomposition module, a task modeling module, a task planning module, a task calibration module, and a task combination module. The task decomposition module decomposes complex tasks into multiple subtasks. The task modeling module builds execution models for subtasks. The task planning module plans all subtasks. The task calibration model calibrates each subtask. The task combination module combines all final subtasks.

[0119] For the aforementioned task execution system, since the models all employ large language models, system prompts can be set to enhance the system's ability to partially understand tasks. For example, a system prompt could be: "You want to generalize all tasks that users ask you to perform. The key steps of generalization include task decomposition, task modeling, task planning, task calibration, and task combination. Your name is Geo-Jarvis, and you are an assistant for Earth science research and geographical knowledge discovery. Complex tasks can be decomposed into several subtasks. The description of a subtask is: <Subtask Name> + <Description of Subtask>. Please note the conciseness of the language. Furthermore, when a user asks you to modify a subtask, you can modify the subtask itself." First, you need to proactively ask the user if the subtask decomposition is reasonable. Then, when the user asks you to model the subtasks, you need to model all subtasks, describing the modeling as "<subtask x> call + <submodel x>". When the user asks you to plan tasks, you need to consider the execution order of the subtasks and provide that order. When the user asks you to calibrate the execution of subtasks, you need to reflect on the execution order and ask the user to confirm whether the subtask planning and calibration are correct. Finally, when the user asks you to combine tasks to obtain the execution result of a complex task, you need to provide your understanding of the execution, describing the result as "<execution result> + <reason>".

[0120] The application of the above task execution system will be illustrated below with specific examples.

[0121] Users are required to perform a spatial distribution pattern analysis task for the elderly population. The instructions in the instruction space are: "Assuming you are a data expert with all knowledge of geospatial distribution pattern analysis, you are required to perform a spatial distribution pattern analysis task for the elderly population. In this task, you need to perform spatial distribution pattern analysis based on the type of spatial distribution pattern, and then proceed with task decomposition, task modeling, task planning, task calibration, and task combination. Finally, you need to use drawing tools to draw the results of the spatial distribution pattern analysis of the elderly population. The results are described as <Distribution Pattern Analysis> + <Interpretation of Distribution Pattern Analysis>. Please analyze this specific task step by step according to the analysis paradigm."

[0122] The task execution system breaks down this complex task into several sub-tasks: "Sub-task 1: Data Collection: Collect geospatial data related to the elderly population, including elderly population distribution and population density," "Sub-task 2: Data Preprocessing: Clean and integrate data to ensure the data format is suitable for spatial analysis," "Sub-task 3: Distribution Pattern Identification: Clean and integrate data to ensure the data format is suitable for spatial analysis," "Sub-task 4: Distribution Pattern Analysis: Use statistical and geographic information system tools to analyze the identified distribution patterns," "Sub-task 5: Visualization: Use plotting tools to create visual representations of the distribution patterns," and "Sub-task 6: Results Interpretation: Based on the distribution patterns and the output of the analysis tools, interpret the potential causes and impacts of the elderly population distribution patterns."

[0123] Then, an execution model is built for each subtask, resulting in: Subtask 1 calls the data acquisition model, Subtask 2 calls the data cleaning model, Subtask 3 calls the pattern recognition algorithm, Subtask 4 calls the spatial analysis model, Subtask 5 calls the GIS drawing tool, and Subtask 6 calls the result interpretation model.

[0124] Next, plan all subtasks and execute them in the following order: subtask 1, subtask 2, subtask 3, subtask 4, subtask 5, subtask 6.

[0125] Then, the subtasks are calibrated by asking the user, sending the user a message that asks, "Please confirm whether the above decomposition and planning meet your expectations, or whether there are any parts that need adjustment."

[0126] After the user confirms that the decomposition and planning are correct, the system receives the relevant data on the elderly population sent by the user and executes all subtasks in the order of execution, resulting in the following: Figure 4 The final task result shown is as follows: Figure 4 The diagram is a heat map of the distribution of the elderly population. The horizontal and vertical axes in the diagram are area numbers, and the darker the gray, the more elderly people there are in that area.

[0127] If a user needs to perform a remote sensing image detection task for maritime vessel types, the instruction would be: "Assuming you are an expert with all the knowledge of analyzing remote sensing image data of maritime vessel targets, you are required to perform a remote sensing image detection task for maritime vessel types. In this task, you need to disassemble the hull into key sub-components based on the different types of vessels, and then continue with task decomposition, task modeling, task planning, task calibration, and task combination. Finally, you need to give the type of the unknown vessel target, and the result description is <Execution Result> + <Reason>. Please analyze this specific task step by step according to your analysis paradigm."

[0128] The task is broken down into the following sub-tasks: "Sub-task 1: Image Preprocessing: Cleaning, standardizing, and enhancing remote sensing image data to improve the identifiability of ship targets"; "Sub-task 2: Hull Disassembly into Key Sub-component Identification: The process of decomposing the hull graphic into its key sub-components (such as hull structure, deck facilities, funnel features, etc.)"; "Sub-task 3: Feature Extraction: Extracting key features (such as shape, size, texture, etc.) from each sub-component"; "Sub-task 4: Classification Model Selection and Training: Selecting a suitable machine learning or deep learning model for remote sensing image analysis and training it with labeled data"; "Sub-task 5: Ship Type Prediction: Inputting the extracted features into the trained model to predict the ship type"; and "Sub-task 6: Result Analysis and Interpretation: Analyzing the prediction results and providing a detailed explanation of the identified ship types."

[0129] The execution model is constructed as follows: Subtask 1 calls the image processing sub-model, Subtask 2 calls the object detection sub-model, Subtask 3 calls the feature extraction sub-model, Subtask 4 calls the machine learning or deep learning training sub-model, Subtask 5 calls the classification prediction sub-model, and Subtask 6 calls the data analysis sub-model.

[0130] The execution order is: subtask 1, subtask 2, subtask 3, subtask 4, subtask 5, subtask 6.

[0131] Then, task calibration is performed. The system believes that during execution, the task may need to be adjusted based on the quality of the preprocessed image, the accuracy of sub-part recognition, and the effectiveness of feature extraction. For example, if a sub-part is found to be difficult to recognize, it may be necessary to return to the image preprocessing stage for more specialized image enhancement.

[0132] The system receives ship images sent by the user, combines and executes tasks, and the final result is a prediction of the ship type and rationale based on extracted features and model data. These features include: a long and wide hull, numerous containers stacked on the deck, specific cranes and loading machinery in the structure, and a hull and deck layout indicating a purpose for large-scale cargo transport. These characteristics match those of a typical container ship; therefore, this ship is a container ship because its long and wide hull, container-stacking deck structure, and the presence of loading machinery are all prominent features of container ships.

[0133] If a user needs to perform a task to discover unknown spatial relationships among the elderly population, the instructions are as follows: Assuming you are a data expert with all the knowledge of discovering unknown spatial relationships in geospatial data, you are given a task to perform. In this task, you need to discover the types of spatial relationships and the unknown spatial relationships, and then proceed with task decomposition, task modeling, task planning, task calibration, and task combination. Finally, you need to use drawing tools to draw the analysis results of the discovery of unknown spatial relationships among the elderly population. The result description is <Unknown Spatial Relationships> + <Interpretation of Unknown Spatial Relationship Discoveries>. Please analyze this specific task step by step according to your analysis paradigm.

[0134] The task is broken down into the following sub-tasks: Sub-task 1: Data Collection: Collect relevant data on the distribution of the elderly population, population density, etc.; Sub-task 2: Data Preprocessing: Clean the data, handle missing values, and standardize the data format; Sub-task 3: Spatial Relationship Type Analysis: Determine the key spatial relationship types in the study, such as distance, adjacency, and connectivity; Sub-task 4: Spatial Model Exploration: Use spatial statistical methods to explore the spatial distribution patterns of the elderly population; Sub-task 5: Unknown Spatial Relationship Mining: Use machine learning and pattern recognition techniques to mine unknown spatial relationships; Sub-task 6: Results Visualization: Use GIS tools to draw spatial relationship maps and highlight important findings.

[0135] The task model is as follows: Subtask 1 calls the data acquisition interface, Subtask 2 calls the data cleaning algorithm, Subtask 3 calls the spatial analysis library, Subtask 4 calls the spatial statistics tool, Subtask 5 calls the machine learning model, and Subtask 6 calls the GIS drawing tool.

[0136] The task plan is as follows: Subtask 1, Subtask 2, Subtask 3, Subtask 4, Subtask 5, Subtask 6.

[0137] Task calibration: Send a message to the user requesting confirmation that the breakdown and execution order of the above subtasks are reasonable. We need to ensure the data processing logic is correct and that the analysis methods are suitable for detecting the spatial relationships of the elderly population.

[0138] Receive relevant data sent by the user, execute all subtasks in the order of execution, and combine the outputs of all subtasks to form a result such as... Figure 5 The bar chart shown below shows the elderly population of several provinces. Figure 5 The horizontal axis of the chart represents provinces, and the vertical axis represents the number of elderly people, in units of 10,000 people. The explanation is as follows: This is a bar chart of the number of elderly people in each province. It can clearly show the spatial relationship of the elderly population in each province. This kind of chart can help us intuitively understand the geographical distribution pattern of the elderly population and provide a basis for further spatial relationship analysis.

[0139] It can be seen that the task execution method of this application can successfully complete the execution of a variety of complex tasks and has good versatility.

[0140] The task execution device based on a large language model provided in this application is described below by way of example.

[0141] like Figure 6 As shown, this application provides a task execution device based on a large language model, the task execution device 600 based on a large language model includes:

[0142] The decomposition module 601 uses a task decomposition model to decompose a complex task to be executed into multiple subtasks and builds an execution model for each subtask; the task decomposition model is a large language model capable of task decomposition.

[0143] Planning module 602 uses a task planning model to plan all subtasks and obtain the execution order of all subtasks; the task planning model is a large language model capable of subtask planning.

[0144] The calibration module 603 calibrates each subtask to obtain multiple final subtasks, and then combines all the final subtasks in the execution order to obtain a combined task. The combined task includes all the planned final subtasks as well as the combination reasoning rules and combination reasoning methods between all the final subtasks. The combination reasoning rules are used to describe the execution order of all the final subtasks, and the combination reasoning methods are methods to optimize the combination reasoning rules.

[0145] Execution module 604 utilizes all execution models to execute combined tasks and obtain the final execution result of complex tasks.

[0146] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0147] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0148] like Figure 7 As shown, an embodiment of this application provides a terminal device, wherein the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 7 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0149] Specifically, when the processor D100 executes the computer program D102, it decomposes the complex task to be executed into multiple subtasks using a task decomposition model, constructs an execution model for each subtask, and then plans all subtasks using a task planning model to obtain the execution order of all subtasks. Each subtask is then calibrated to obtain multiple final subtasks. These final subtasks are then combined according to the execution order to obtain a combined task. Finally, the combined task is executed using all execution models to obtain the final execution result of the complex task. Decomposing the complex task into multiple subtasks transforms a complex task into multiple simpler subtasks, effectively reducing the difficulty of executing the complex task. Planning all subtasks improves the logic and rationality of their execution. Combining and executing all final subtasks with lower execution difficulty based on a logically sound execution order improves the rationality and versatility of the task execution method.

[0150] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0151] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0152] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0153] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0154] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a task execution method apparatus / terminal device based on a large language model, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0155] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0156] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0157] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A large language model-based task execution method, characterized by, The method comprises the following steps: a task decomposition model is used to decompose a complex task to be executed into a plurality of subtasks, and an execution model is constructed for each of the subtasks; the task decomposition model is a large language model capable of task decomposition, the complex task is an old population distribution mapping task, and the plurality of subtasks comprise data acquisition and collection of geographic spatial data of old population distribution, data preprocessing and cleaning of collected data, spatial distribution analysis and analysis of geographic spatial distribution of sampling points, and map drawing and drawing of sampling point distribution map using geographic information system software or library; a task planning model is used to plan all the subtasks to obtain an execution sequence of all the subtasks; the task planning model is a large language model capable of subtask planning; each of the subtasks is calibrated to obtain a plurality of final subtasks, and all the final subtasks are combined according to the execution sequence to obtain a combined task; the combined task comprises all the final subtasks and combined reasoning rules and combined reasoning methods between the final subtasks, the combined reasoning rules are used to describe the sequence of execution of all the final subtasks, and the combined reasoning methods are methods for optimizing the combined reasoning rules; all the execution models are used to execute the combined task to obtain a final execution result of the complex task; the final execution result is an old population density map and corresponding explanation.

2. The task execution method according to claim 1, wherein The complex task is decomposed into a plurality of subtasks by using the task decomposition model, which comprises: by formula: ； Acquiring a plurality of subtasks ; wherein, represents a task decomposition model, represents the complex task, represents a 1st subtask, represents a 2nd subtask, represents a 3rd subtask, represents a 4th subtask.

3. The task execution method according to claim 2, wherein An execution model is constructed for each of the subtasks, which comprises: by formula: ； Acquiring a combination of each subtask and execution model ; wherein, denotes a modeling model, denotes an execution model of the first subtask, denotes an execution model of the second subtask, denotes an execution model of the first subtask, is the total number of subtasks.

4. The task execution method according to claim 1, wherein All the subtasks are planned by using the task planning model to obtain an execution sequence of all the subtasks, which comprises: by formula: ； Acquisition execution order ; wherein, represents a planning model, represents all execution models, represents a task work environment of a complex task, represents a task goal of a complex task, represents a first executed subtask, represents a second executed subtask, represents a third executed subtask.

5. The task execution method according to claim 1, wherein Each of the subtasks is calibrated to obtain a plurality of final subtasks, which comprises: by formula: ； acquiring a set of final sub-tasks ; wherein, denotes a calibration model, denotes calibration data, denotes a task calibration rule, denotes a task target error and bias constraint condition, denotes a final subtask corresponding to a 1st subtask, denotes a final subtask corresponding to a 2nd subtask, denotes a final subtask corresponding to a 3rd subtask, denotes a final subtask corresponding to a 4th subtask.

6. The task execution method according to claim 1, wherein All the execution models are used to execute the combined task to obtain a final execution result of the complex task, which comprises: by formula: ； Obtaining final execution result ; wherein, represents a combination model, represents input data, represents a combination inference rule, represents a combination inference method, represents an execution result of a final subtask executed first in the combination task, represents an execution result of a final subtask executed second in the combination task, represents an execution result of a final subtask executed third in the combination task, represents an execution result of a final subtask executed fourth in the combination task.

7. A large language model-based task execution apparatus, characterized by comprising: The method comprises the following steps: a task decomposition module is used to decompose a complex task to be executed into a plurality of subtasks, and an execution model is constructed for each of the subtasks; the task decomposition model is a large language model capable of task decomposition, the complex task is an old population distribution mapping task, and the plurality of subtasks comprise data acquisition and collection of geographic spatial data of old population distribution, data preprocessing and cleaning of collected data, spatial distribution analysis and analysis of geographic spatial distribution of sampling points, and map drawing and drawing of sampling point distribution map using geographic information system software or library; a planning module is used to plan all the subtasks by using a task planning model to obtain an execution sequence of all the subtasks; the task planning model is a large language model capable of subtask planning; and each of the subtasks is calibrated to obtain a plurality of final subtasks, and all the final subtasks are combined according to the execution sequence to obtain a combined task; the combined task comprises all the final subtasks and combined reasoning rules and combined reasoning methods between the final subtasks, the combined reasoning rules are used to describe the sequence of execution of all the final subtasks, and the combined reasoning methods are methods for optimizing the combined reasoning rules. The calibration module calibrates each of the sub-tasks to obtain a plurality of final sub-tasks, and then combines all the final sub-tasks according to the execution sequence to obtain a combined task; the combined task includes all the final sub-tasks, and a combined reasoning rule and a combined reasoning method between the final sub-tasks, the combined reasoning rule is used to describe the sequence of execution of all the final sub-tasks, and the combined reasoning method is an optimization method for the combined reasoning rule; The execution module executes the combined task by using all the execution models to obtain a final execution result of the complex task; the final execution result is an old population density map and a corresponding explanation.

8. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the task execution method based on the large language model according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program is executed by the processor to implement the task execution method based on the large language model according to any one of claims 1 to 6.

Images