Automatic operation method and related device

Abstract

The invention discloses an automatic operation method and a related device, and relates to the field of information processing, and the method comprises the steps: receiving a target task; determining target description information according to the target task; according to the target description information, a muscle memory library is inquired to obtain a muscle memory list, the muscle memory library comprises a plurality of muscle memories, the muscle memories are generated when a large language model successfully executes a corresponding task, and the muscle memory list comprises at least one muscle memory; controlling a large language model to determine a matching result with the target task according to each muscle memory in the muscle memory list; and executing the target task according to the matching result to obtain an execution result.

Description

Technical Field

[0001] The present application relates to the field of information processing, and in particular to an automated operation method and related devices. Background Art

[0002] With the development of intelligent agents, they can now automate tasks on electronic devices, such as operating Windows computers, Android phones and even games.

[0003] For example, the intelligent agent uses multimodal large language models (MLLMs) to control keyboard and mouse input to complete automated task operations.

[0004] The process of the agent performing the task includes: 1. Screen capture; 2. Labeling the operable parts on the screenshot; 3. Submitting the labeled screenshot, the type of operations available, and the operation records before the current step to the MLLMs. The MLLMs determines whether the task is completed. If completed, execute 4; if not completed, obtain the item and operation type of the next step; 4. After receiving feedback from the MLLMs, the agent calls the code to execute the next step.

[0005] When the agent performs a task, it needs to use MLLMs once for each step. Since it takes a long time to call MLLMs once, when the task contains multiple steps, it will take too long to execute the task. Summary of the invention

[0006] The first aspect of the present application provides an automated operation method, comprising:

[0007] Receive target tasks;

[0008] Determining target description information according to the target task;

[0009] According to the target description information, query a muscle memory library to obtain a muscle memory list, wherein the muscle memory library includes a plurality of muscle memories, wherein the muscle memories are generated when the large language model successfully performs a corresponding task, and the muscle memory list includes at least one muscle memory;

[0010] Controlling the large language model to determine a matching result with the target task according to each muscle memory in the muscle memory list;

[0011] The target task is executed according to the matching result to obtain an execution result.

[0012] In a possible implementation, determining target description information according to the target task includes:

[0013] Inputting at least text information included in the target task into a large language model, wherein the large language model is capable of generating target description information including at least the text information;

[0014] Correspondingly, according to the target description information, the muscle memory library is queried to obtain a muscle memory list, including:

[0015] A muscle memory list consisting of at least one muscle memory is obtained by screening according to the target description information and the description information of each muscle memory in the muscle memory library.

[0016] In a possible implementation, executing the target task according to the matching result to obtain an execution result includes:

[0017] According to the matching result, it is indicated that there is a target muscle memory matching the target task in the muscle memory list, and the target task is executed according to the target muscle memory to obtain a first execution result;

[0018] According to the matching result indicating that the target muscle memory does not exist in the muscle memory list, the large language model is controlled to execute the target task to obtain a second execution result.

[0019] In a possible implementation, executing the target task according to the matching result, after obtaining the execution result, further includes:

[0020] According to the execution result, judging whether the target task is executed successfully and obtaining a judgment result;

[0021] The target task is successfully executed based on the judgment result, and the muscle memory bank is updated according to the execution result.

[0022] In a possible implementation, judging whether the target task is successfully executed based on the execution result to obtain the judgment result includes:

[0023] receiving feedback information, wherein the feedback information is information input by an operator in response to the execution result;

[0024] Based on the feedback information, it is determined whether the target task is successfully executed.

[0025] In a possible implementation, updating the muscle memory bank according to the execution result includes:

[0026] According to the second execution result, controlling the large language model to generate code to complete the target task;

[0027] Controlling the large language model to determine parameterizable items in the code of the target task to obtain parameterized code;

[0028] Muscle memory corresponding to the target task is generated based on the target description information and parameterized code.

[0029] In a possible implementation, the step of indicating that the target task is successfully executed based on the judgment result and updating the muscle memory base according to the execution result includes:

[0030] According to the first execution result, determining that the target task is not successfully executed, controlling the large language model to execute the target task, and obtaining a third execution result;

[0031] According to the third execution result, judging whether the target task is executed successfully, and obtaining a third judgment result;

[0032] The target task is successfully executed based on the third judgment result, and the muscle memory bank is updated according to the third execution result.

[0033] In a possible implementation, updating the muscle memory bank according to the third execution result includes:

[0034] Based on the third execution result, controlling the large language model to generate code to complete the target task;

[0035] According to the code of the target task, the code in the target muscle memory is updated.

[0036] In a possible implementation, updating the code in the target muscle memory according to the code of the target task includes:

[0037] Controlling the large language model to compare the code of the target task with the code in the target muscle memory to obtain the code difference;

[0038] Based on the code difference representation that the operation logic has not changed and the code change rule, the large language model is controlled to modify the parameters of the code in the target muscle memory;

[0039] Based on the code difference characterizing that the operation logic has not changed and the code changes irregularly, the large language model is controlled to determine the parameter selection small model, the code in the target muscle memory is modified by using the parameter selection small model, and the parameter selection small model is added to the target muscle memory;

[0040] Based on the code distinction representing the change in operation logic, the large language model is controlled to replace the code in the target muscle memory with the code of the target task.

[0041] A second aspect of the present application provides an automated operating device, comprising:

[0042] A memory, used to store a muscle memory library, wherein the muscle memory library includes a plurality of muscle memories, and the muscle memories are generated when the large language model successfully performs a corresponding task;

[0043] Interface, used to receive target tasks;

[0044] A processor is used to determine target description information based on the target task; query a muscle memory library based on the target description information to obtain a muscle memory list, wherein the muscle memory list includes at least one muscle memory; control a large language model to determine a matching result with the target task based on each muscle memory in the muscle memory list; and execute the target task based on the matching result to obtain an execution result.

[0045] A third aspect of the present application provides a computer program product, comprising computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the automated operation method of the first aspect or any implementation of the first aspect.

[0046] A fourth aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0047] The memory is used to store computer programs;

[0048] The processor is used to execute the computer program so that the electronic device can implement the automated operation method of the above-mentioned first aspect or any implementation manner of the first aspect.

[0049] A fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can realize the automated operation method of the above-mentioned first aspect or any implementation method of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0051] Figure 1 It is a flowchart of an automated operation method provided in an embodiment of the present application;

[0052] Figure 2 It is a flowchart of executing the target task according to the matching result and obtaining the execution result provided by the embodiment of the present application;

[0053] Figure 3 It is a schematic diagram of a process for updating a muscle memory bank provided in an embodiment of the present application;

[0054] Figure 4It is a flowchart of determining whether the target task is successfully executed based on the execution result and obtaining the determination result provided by the embodiment of the present application;

[0055] Figure 5 It is a schematic diagram of a process of updating a muscle memory bank according to an execution result provided in an embodiment of the present application;

[0056] Figure 6 It is a flowchart of updating the muscle memory bank according to the execution result based on the judgment result indicating the successful execution of the target task provided by the embodiment of the present application;

[0057] Figure 7 is a schematic diagram of a process for updating a muscle memory bank according to a third execution result provided by an embodiment of the present application;

[0058] Figure 8 It is a flowchart of updating the code in the target muscle memory according to the code of the target task provided by an embodiment of the present application;

[0059] Fig. 9 It is a structural schematic diagram of an automated operating device provided in an embodiment of the present application;

[0060] Fig.10 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method section of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0062] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0063] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, which is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0064] The inventors have found that human muscle memory has the following characteristics:

[0065] 1. The action space is limited and the action mode is relatively fixed: human limb activities must follow ergonomic principles, which makes the range of selectable actions limited;

[0066] 2. Quick and efficient response: No need to go through a complex cognitive decision-making process, and can respond quickly;

[0067] 3. Adaptability: It can automatically adjust the action parameters according to the slight changes in the environment, showing the characteristics of flexibility rather than mechanical rigidity;

[0068] 4. Persistent and stable: Once formed, the behavior pattern will be deeply solidified and not easy to disappear;

[0069] 5. Repeated practice is required: Forming stable muscle memory requires a lot of repeated training.

[0070] Analogous to human behavior patterns, when performing skilled sports and operational actions, although supervision of the cognitive system is required, more reliance is placed on the already formed muscle memory to achieve fast and smooth execution.

[0071] Therefore, in this application, the concept of "muscle memory" is introduced into the field of intelligent agent operation. Muscle memory adopts the intelligent agent architecture of "parameterized code and parameter selection small model". When the large language model successfully executes any task, the corresponding muscle memory is generated according to the execution process. The action is executed with reference to human muscle memory, and the generated muscle memory is applied to subsequent task responses. The execution speed is fast and a certain adaptability is maintained.

[0072] Figure 1 is a flow chart of an automated operation method provided in an embodiment of the present application, such as Figure 1 As shown, an automated operation method provided by an embodiment of the present application is applied to a functional module of an electronic device, the functional module being capable of calling a large language model, and may include steps 101 to 105, which are described in detail below.

[0073] 101. Receive target tasks;

[0074] The target task is a task that needs to be performed automatically, and the target task can be any task in any scenario.

[0075] For example, for a mailbox application, the target task may be to read unread emails therein; for a drawing application, the target task may be to draw a specific image in an input area, etc. The specific content of the target task is not limited in this application.

[0076] The operator may input a task on an input device, and the electronic device executing the automated operation method in the present application receives the target task and executes subsequent steps in response.

[0077] 102. Determine target description information according to the target task;

[0078] The target description information is information used to describe the target task, which may include various information related to the target task, such as the application involved in the target task, specific execution-related information, etc., but is certainly not limited thereto.

[0079] The target task is a task to be performed by the target. When the operator inputs the target task, he only inputs simple text information, which may not contain all the information. It is necessary to analyze the target task to obtain its target description information.

[0080] In a possible implementation, the target description information may be obtained by expanding the text information contained in the target task.

[0081] In a possible implementation, at least text information included in the target task is input into a large language model, and the large language model can generate target description information including at least the text information.

[0082] Specifically, the large language model can convert the text information into a vector form, and expand it using the text information in the vector form to obtain target description information containing more information content.

[0083] Among them, the user information of the target task, the information of the adopted equipment, etc. and the text information contained in the target task can be input into the large language model together. The large language model can combine the information of multiple dimensions for analysis to generate the target description information.

[0084] As an example, the target task is "open unread emails", the user information of the target task can be the login account "User A" of the current device, and the device used can be a mobile phone. Accordingly, the large language model uses the aforementioned content, and the target description information obtained can be "open the unread emails of User A on the mobile phone".

[0085] 103. According to the target description information, query the muscle memory library to obtain a muscle memory list, wherein the muscle memory library includes a plurality of muscle memories, and the muscle memory is generated when the large language model successfully executes the corresponding task, and the muscle memory list includes at least one muscle memory;

[0086] Among them, a muscle memory library is preset, which contains a number of muscle memories, and the muscle memory is generated when the large language model successfully executes the corresponding task.

[0087] Among them, when the large language model successfully executes any task, it generates corresponding muscle memory according to the execution process, and the muscle memory can be executed the next time the same / similar task needs to be executed.

[0088] Among them, after determining the target description information, the muscle memory library can be queried to obtain a muscle memory list.

[0089] The muscle memory list may be a plurality of muscle memories that are close in matching degree to the target description information, or it may be said that the muscle memories in the muscle memory library are sorted.

[0090] In one possible implementation, the muscle memory may include description information, code, and a small model.

[0091] The description information of the muscle memory is used to describe the scope of application.

[0092] The code is the complete operation logic of the muscle memory to execute the corresponding task, and the changing parts of the code are abstracted as parameters. The main structure and process of the code are fixed, and some details that change according to the needs of different scenarios are designed as parameters, so that the code can accept different parameter values ​​when executed, so as to adapt to various situations.

[0093] Among them, the small model is the processing model used when the task steps are executed. The data processing capability of the small model is smaller than that of the large language model. The small model can predict the required parameter values ​​according to the current environmental characteristics.

[0094] The current environment feature may be a feature of the environment involved when the corresponding step is executed, such as a screenshot, a UI (User Interface) element position, operable list data in an application, and the like.

[0095] In a possible implementation, a muscle memory list consisting of at least one muscle memory is obtained by screening according to the target description information and the description information of each muscle memory in the muscle memory library.

[0096] The target description information is compared with the description information of each muscle memory in the muscle memory library for similarity, and the muscle memories are sorted in descending order of similarity, from which a set number of muscle memories are determined to form a muscle memory list.

[0097] As an example, if there are 10 muscle memories in the muscle memory list, the top 10 muscle memories in similarity ranking are selected.

[0098] 104. Control the large language model to determine a matching result with the target task according to each muscle memory in the muscle memory list;

[0099] The muscle memory list is input into the large language model, and the large language model is controlled to analyze whether each muscle memory in the muscle memory list matches the target task, so as to obtain a matching result.

[0100] The matching result may be that there is available muscle memory in the muscle memory list, or that there is no available muscle memory in the muscle memory list.

[0101] Among them, the muscle memory is available, which means that the muscle memory is able to perform the target task.

[0102] 105. Execute the target task according to the matching result to obtain an execution result.

[0103] Among them, since the matching result includes two situations, the target task is executed according to different situations to obtain corresponding execution results.

[0104] Among them, different methods can be used to perform target tasks according to different situations.

[0105] In a possible implementation, when the matching result indicates that there is available muscle memory in the muscle memory list, the available muscle memory can be used to perform the target task. When there is no available muscle memory, the large language model is used to perform the target task. Figure 2 This process is described in detail in .

[0106] In a possible implementation, after obtaining the execution result, it is necessary to determine whether the target task is successfully executed and update the muscle memory library. Figure 3 The process of updating the muscle memory bank is described in detail in the .

[0107] Among them, the large language model is used to decide which method to use to perform the target task. The specific execution can be achieved by the code in the muscle memory to execute the target task. The muscle memory can execute commonly used, fixed-mode actions (tasks). Compared with the large model, it is lighter in size and takes less time to call.

[0108] In this embodiment, a target task is received; target description information is determined according to the target task; according to the target description information, a muscle memory library is queried to obtain a muscle memory list, the muscle memory library includes a plurality of muscle memories, muscle memories are generated when the large language model successfully executes the corresponding task, and the muscle memory list includes at least one muscle memory; the large language model is controlled to determine the matching result with the target task according to each muscle memory in the muscle memory list; the target task is executed according to the matching result to obtain the execution result. Through the target description information of the target task, muscle memories are screened in the muscle memory library to obtain a muscle memory list, and the matching result of each muscle memory in the muscle memory list and the target task is determined by using the large model, and the target task is executed according to the matching result to obtain the execution result. During the execution process, the action of executing the target task using muscle memory does not require the large language model to make high-level decisions, and does not require the large language model to execute step by step for each action step of each task. Therefore, it is not necessary to call the large language model once for each step operation, which reduces the time required for task execution and improves the efficiency of task execution.

[0109] Figure 2 It is a flowchart of executing the target task according to the matching result and obtaining the execution result provided by the embodiment of the present application, which may include steps 201 to 202, and these steps are described in detail below.

[0110] 201. Indicating that there is a target muscle memory matching the target task in the muscle memory list according to the matching result, executing the target task according to the target muscle memory to obtain a first execution result;

[0111] If there is a target muscle memory matching the target task in the muscle memory list, it indicates that the target task or a similar task has been successfully executed and the target muscle memory has been retained.

[0112] Correspondingly, the target task is executed according to the target muscle memory to obtain a first execution result.

[0113] In a possible implementation, there is code in the target muscle memory, and the target task is executed by running the muscle memory code and calling the corresponding parameter selection small model in the process.

[0114] As an example, the target task is to open unread emails. The code in the muscle memory includes multiple sections, each of which executes a step. Specifically, it can be: run the code to open the email client; run the code to open the inbox; run the code to read the email list data in the inbox; run the code to call the small model, input the email list, and get the unread email id (Message-ID, the unique identifier of the email); run the code to open the email through the unread email id.

[0115] 202. According to the matching result, it is indicated that the target muscle memory does not exist in the muscle memory list, and the large language model is controlled to execute the target task to obtain a second execution result.

[0116] If there is no target muscle memory matching the target task in the muscle memory list, it indicates that the target task or a similar task has never been executed and the target muscle memory has not been retained.

[0117] Accordingly, the large language model can be controlled to execute the target task, specifically starting from the first step of the target task, and the large language model gradually completes the task to obtain a second execution result.

[0118] When the large language model executes the target task, the large language model is called once for each execution step until the execution result is finally obtained.

[0119] In a possible implementation, after executing the target task, it can be further determined whether the target task is successfully executed. If successful, the muscle memory bank can be updated with the execution result to increase the scope of application of the muscle memory bank. Figure 3 This process is described in detail in .

[0120] In this embodiment, based on the matching result characterizing the existence of target muscle memory matching the target task in the muscle memory list, the target task is executed based on the target muscle memory to obtain a first execution result; based on the matching result characterizing the absence of the target muscle memory in the muscle memory list, the large language model is controlled to execute the target task to obtain a second execution result, thereby achieving the goal of directly using the target muscle memory to execute the target task when the target muscle memory exists, without the need for the large model to execute the task step by step, thereby improving the task execution efficiency; if the target muscle memory does not exist, the large model is still used to execute the target task to ensure that the target task can be executed.

[0121] Figure 3 It is a flowchart of updating the muscle memory bank provided in an embodiment of the present application, which may include steps 301 to 302. These steps are described in detail below.

[0122] 301. According to the execution result, determine whether the target task is successfully executed and obtain a determination result;

[0123] Among them, after executing the target task according to the matching result and obtaining the execution result, it is judged whether the execution of the target task is successful based on the execution result to obtain the judgment result.

[0124] In a possible implementation, a screenshot of the interface corresponding to the execution result can be taken, and the target description information can be linked to the screenshot. Figure 1The message is sent to the large language model, which determines whether the target task is successfully executed.

[0125] In a possible implementation, whether the target task is successfully executed may be determined based on whether all steps in the target task are executed.

[0126] Among them, the completed steps can be recorded in the execution results. By using the completed steps recorded in the execution results, it is determined whether all the steps required to execute the target task have been executed. If all of them have been executed, it can be judged that the target task has been successfully executed and the subsequent step 302 can be executed. Otherwise, the execution of the task has failed.

[0127] As an example, the target task can be executed in 5 steps. When executing the target task, you need to execute the 5 steps in the target task in the order of execution to get the execution result. When a step cannot be executed, stop the execution process and get an execution result. When all 5 steps are completed, get the execution result. Based on whether the execution steps recorded in the execution result are 5, determine whether the target task is successfully executed.

[0128] In a possible implementation, whether the target task is successfully executed may be determined based on whether the last step of execution matches the target task.

[0129] The completed steps can be recorded in the execution result. By checking whether the last execution step recorded in the execution result matches the target task, if they match, it can be determined that the target task is executed successfully and the subsequent step 302 can be executed. Otherwise, the execution of the task fails.

[0130] As an example, the target task is "open unread mail task". When executing this target task, the following steps are performed: open the mail client; open the inbox; read the mail list data in the inbox; call the small model, input the mail list, get the unread mail id; open the mail through the unread mail id. The last step in the execution process is "open the mail through the unread mail id", which matches the target task "open unread mail task", and it is determined that the target task is executed successfully.

[0131] In a possible implementation, the execution result indicates that the target task has failed to execute, and the target task can be re-executed. During the execution, the large language model can be controlled to execute the target task to improve the accuracy of successful task execution.

[0132] In a possible implementation, the target task can be judged whether it is executed successfully based on the feedback information of the operation body on the execution result. Figure 4 The following provides detailed instructions for determining whether the target task is successfully executed.

[0133] 302. Based on the judgment result indicating that the target task is successfully executed, the muscle memory bank is updated according to the execution result.

[0134] Among them, if the judgment result indicates that the target task is successfully executed and the process of the target task executed this time is determined to be correct, the muscle memory bank can be updated according to the execution result.

[0135] In a possible implementation, the updating of the muscle memory bank may be performed in any of the following ways: adding new muscle memory to the muscle memory bank, or updating existing muscle memory in the muscle memory bank.

[0136] In a possible implementation, muscle memory can be updated in different ways for different execution results. Figure 5 and Figure 6-Figure 7 , the process of updating the muscle memory bank for the two execution results is explained in detail.

[0137] In this embodiment, after obtaining the execution result, it is determined whether the execution of the target task is successful based on the execution result to obtain a judgment result; based on the judgment result, it is characterized that the execution of the target task is successful, and the muscle memory bank is updated based on the execution result, so that the muscle memory bank can be compatible with the target task executed this time, thereby improving the application scope of the muscle memory bank.

[0138] Figure 4 It is a flowchart of determining whether the target task is successfully executed based on the execution result and obtaining the judgment result provided by the embodiment of the present application, which may include steps 301 to 302. These steps are described in detail below.

[0139] 401. Receive feedback information, where the feedback information is information input by the operator regarding the execution result;

[0140] Among them, if the target task is executed, the execution result is output.

[0141] The user may provide feedback on the execution result by inputting feedback information through an input device. The feedback information includes information on whether the execution result meets the user's expectations.

[0142] In a specific implementation, the execution result can be output through an output device. The user can judge whether the execution result is what he wants based on the content output by the output device, and perform input operations based on the output content. The user's operations are received through the input device, and the operation information can be used as feedback information.

[0143] The input operation may include inputting other tasks, re-inputting the target task, or returning to an interface before the target task is executed.

[0144] 402. Determine whether the target task is successfully executed based on the feedback information.

[0145] The feedback information is analyzed to determine whether the feedback information is negative feedback on the target task execution result. If it is negative feedback, the target task execution is determined to have failed; otherwise, the target task execution is determined to have succeeded.

[0146] Among them, negative feedback means not approving the execution result of the target task, and positive feedback means approving the execution result of the target task.

[0147] Among them, negative feedback is a return operation for the execution result of the target task, re-inputting the target task to re-execute the target task, etc.

[0148] The positive feedback may be to continue inputting other tasks.

[0149] As an example, the target task is "open unread emails". After executing the target task and obtaining the execution result, the execution result is output and the user's feedback is received. For example, the user clicks to return to the previous interface and reopens the unread email. This operation can be used as negative feedback, indicating that the user does not approve of the execution result. Accordingly, it can be determined that the target task has failed. For example, the user inputs a new task "exit email and open conference software". This operation can be used as positive feedback for the execution result of the target task, indicating that the user approves of the execution result and continues other tasks. Accordingly, it can be determined that the target task has been successfully executed.

[0150] In a possible implementation, after the target task is completed and the execution result is obtained, it can be determined first whether all steps of the target task are executed, and then further combined with user feedback to determine whether the target task is successfully executed.

[0151] In a possible implementation, a screenshot of the interface corresponding to the execution result may be sent to the large language model together with the feedback information, and the large language model determines whether the target task is successfully executed.

[0152] In this embodiment, feedback information is received, and the feedback information is information input by the operator regarding the execution result; based on the feedback information, it is determined whether the execution of the target task is successful, and the execution result after the target task is automatically executed is achieved. It can be determined whether the execution of the target task is successful in combination with the feedback of the operator, so as to determine whether the execution result of the target task meets the expectations of the operator, and to update the muscle memory bank according to the successful execution result, so that when the task is subsequently executed, it will be more in line with the expectations of the operator and the accuracy of the execution will be improved.

[0153] Figure 5It is a flowchart of updating the muscle memory bank according to the execution results provided by an embodiment of the present application, which may include steps 501 to 503, and these steps are described in detail below.

[0154] 501. According to the second execution result, control the large language model to generate code for completing the target task;

[0155] Among them, when the target muscle memory does not exist in the muscle memory list, the large language model is controlled to execute the target task to obtain a second execution result, which is the execution result finally obtained by the large language model participating in the execution step by step.

[0156] During the execution process, each step of the execution needs to call the large language model, and the large language model participates in each step. Accordingly, the large language model can generate the code for the target task.

[0157] In a possible implementation, based on the second execution result characterizing that the target task is successfully executed, the large language model generates complete code for executing the target task according to the successful steps. The code can be divided into multiple sections according to the steps, and each section corresponds to one step.

[0158] 502. Control the large language model to determine parameterizable items in the code of the target task, and obtain parameterized code;

[0159] Among them, the large language model has the function of analyzing code.

[0160] Correspondingly, the large language model can consider the code of the target task and whether the code can be parameterized to adapt to more situations.

[0161] Among them, common patterns are extracted from the code of the target task, and the changing parts are abstracted as parameters to obtain parameterizable items, which are used as parameter adjustment codes to obtain parameterized code.

[0162] The code of the target task is parameterized to obtain parameterized code, which is compatible with tasks with different parameter values, thereby improving the application scope of the code.

[0163] The parameterized code contains the complete operation logic, in which the changing parts are abstracted as parameters. The main structure and process of the code are fixed, and some details may change according to different scenarios or requirements. These details can be involved as parameters, so that the code can accept different parameter values ​​when executed, thereby adapting to various situations.

[0164] Among them, the use of large models to generate code not only utilizes the powerful processing capabilities of the large model, but also strengthens the system of the large model, realizing the paradigm of enhancing the large model itself.

[0165] 503. Generate muscle memory corresponding to the target task based on the target description information and parameterized code.

[0166] Among them, the target description information generated in the above steps is used as the description information in the muscle memory to describe the application scope of the muscle memory.

[0167] Specifically, according to the combination of the components in the muscle memory, the target description information and the parameterized code are stored correspondingly to obtain the muscle memory corresponding to the target task.

[0168] Among them, since the code is parameterized, when there are other tasks in the future, when the only difference between the two is the parameter values, the code generated this time can also be compatible with the task, ensuring that the task can also be successfully executed through the code in the muscle memory corresponding to the target task, without the need for a large language model to gradually execute the task, thereby reducing the time required to execute the task.

[0169] In this embodiment, based on the second execution result, a code for completing the target task is obtained; the large language model is controlled to determine the parameterizable items in the code of the target task to obtain the parameterized code; and the muscle memory corresponding to the target task is generated based on the target description information and the parameterized code. After the large language model executes the target task, it needs to generate parameterized code, and use the code to generate muscle memory, so that muscle memory is added to the muscle memory library. The target task can be executed later using muscle memory, and the large language model does not need to execute the target task every time, which reduces the time required for task execution and improves task execution efficiency.

[0170] Figure 6 The embodiment of the present application provides a flowchart of updating the muscle memory bank based on the judgment result indicating successful execution of the target task, which may include steps 601 to 603. These steps are described in detail below.

[0171] 601. According to the first execution result, it is determined that the target task is not successfully executed, and the large language model is controlled to execute the target task, thereby obtaining a third execution result;

[0172] Among them, when there is a target muscle memory in the muscle memory list, the target task is executed according to the target muscle memory to obtain a first execution result, and the first execution result is the final execution result of executing the target task according to the target muscle memory.

[0173] Among them, when the small model or code in the target muscle memory is incomplete, small model processing errors or problems may occur when executing the code, and the target task cannot be successfully performed. Therefore, the target muscle needs to be updated.

[0174] Among them, the imperfection is due to its incompatibility with new situations arising in the target task, and the code or parameterized small model in the target muscle memory does not match the target task.

[0175] If it is determined based on the first execution result that the target task has not been successfully executed, it means that the code or parameterized small model in the target muscle memory does not match the target task.

[0176] Accordingly, the large language model is controlled to execute the target task. Specifically, the large language model participates in the execution step by step to obtain a third execution result.

[0177] In the process of the large language model executing the target task, each step of execution calls the large language model, and the large language model participates in each step.

[0178] Among them, if the first execution result indicates that the target task is successfully executed, indicating that the target muscle memory can successfully execute the target task, no processing is performed and the muscle memory library is not updated.

[0179] 602. Judging whether the target task is successfully executed according to the third execution result, and obtaining a third judgment result;

[0180] Among them, the process of judging whether the target task is executed and obtaining the third judgment result is similar to the aforementioned step 301. For specific explanations, please refer to the explanations in the aforementioned step 301, which will not be repeated in this embodiment.

[0181] 603. Based on the third judgment result indicating that the target task is successfully executed, the muscle memory bank is updated according to the third execution result.

[0182] If the target task is executed successfully, the muscle memory bank is updated according to the third execution result.

[0183] Among them, the target task was not successfully executed by the target muscle memory, but was successfully executed by the large language model. The muscle memory bank was updated using the third execution result of the successful execution of the large language model to ensure that the muscle memory bank is compatible with the target task. When the target task is received again in the future, the muscle memory in the muscle memory bank can be used to successfully execute the target task without calling the large language model again.

[0184] In one possible implementation, after completing the target task, the large language model generates the code for the target task and uses the code to update the target muscle memory code. Figure 7 The process of updating the muscle memory bank based on the third execution result is described in detail.

[0185] In this embodiment, based on the first execution result, it is determined that the target task has not been successfully executed, and the large language model is controlled to execute the target task to obtain a third execution result; based on the third execution result, it is determined that the target task has been successfully executed, and the muscle memory bank is updated based on the third execution result. When the target muscle memory cannot successfully execute the target task, the large language model is called, and the target task is executed by the large language model. The muscle memory bank is updated using the third execution result successfully executed by the large language model, thereby ensuring that the muscle memory bank is compatible with the target task, thereby reducing the duration of task execution as a whole.

[0186] Figure 7 It is a flowchart of updating the muscle memory bank according to the third execution result provided by an embodiment of the present application, which may include steps 701 to 702, and these steps are described in detail below.

[0187] 701. Based on the third execution result, control the large language model to generate code to complete the target task;

[0188] Among them, when the target muscle memory cannot successfully execute the target task, the large language model is called to gradually execute the target task to obtain a third execution result.

[0189] In the process of the large language model executing the target task, each step of execution calls the large language model, and the large language model participates in each step. Accordingly, based on the third execution result representing the execution success, the large language model generates code to complete the target task.

[0190] In the present application, the large language model can be used as a code generator to generate code. After each task is executed by the large language model, the code for executing the task is generated.

[0191] Correspondingly, the large language model can determine whether muscle memory can be used to execute the target task, and determine whether the target task is successfully executed based on the execution results. It acts as a code generator and supervisor, and muscle memory plays the main role in executing the target task. In the task processing architecture, it reduces the burden of the large language model, can reduce the number of times the large model is called during task processing, and improves overall processing efficiency.

[0192] 702. Update the code in the target muscle memory according to the code of the target task.

[0193] Among them, according to the code of the target task generated by the large language model, the code in the target muscle memory in the muscle memory library is updated.

[0194] The update may be a code change caused by adding parameters, adding parameters to select a small model, or replacing all codes.

[0195] In one possible implementation, the code in the target muscle memory can be updated in different ways for different situations. Figure 8 The process of updating the code in the target muscle memory based on the code of the target task is described in detail.

[0196] In this embodiment, based on the third execution result, the large language model is controlled to generate the code to complete the target task; according to the code of the target task, the code in the target muscle memory is updated. When the target muscle memory cannot successfully execute the target task, the large language model is called to gradually execute the target task to obtain the third execution result, the code of the target task generated by the large language model is controlled, and the code in the target muscle memory is updated, so as to realize the update of the muscle memory library and ensure that the muscle memory library is compatible with the target task.

[0197] Figure 8 It is a flowchart of updating the code in the target muscle memory based on the code of the target task provided by an embodiment of the present application, which may include steps 801 to 804, and these steps are described in detail below.

[0198] 801. Control the large language model to compare the code of the target task with the code in the target muscle memory to obtain a code difference;

[0199] The code of the target task generated by the large language model is compared with the code in the target muscle memory to obtain the difference between the two codes.

[0200] Among them, the large language model has the function of analyzing code.

[0201] In one possible implementation, the large language model compares the code of the target task with the code in the target muscle memory, and combines it with pre-trained code modification rules to make the following considerations, such as whether to add code functions, whether to extract new parameters, whether a new parameter selection model is needed, etc.

[0202] In one possible implementation, the code in the target muscle memory is processed and updated, and the description information in the target muscle memory is also adjusted. The original description information of the target muscle memory is combined with the target description information of the target task to obtain new description information. The new description information contains the original description information and the target description information to match the task to which the updated code in the target muscle memory applies.

[0203] Among them, the large language model can judge whether the operation logic has changed, whether it is a regular change, etc. based on the code difference, and consider the operation logic and whether parameters are involved.

[0204] If the operation logic has not changed, the difference between the two may only be due to parameter influence, and the parameter influence may be regular parameter changes or irregular changes. Different code updates are performed for different situations, and steps 802-803 are executed.

[0205] If the operation logic changes, the code representing the target task is very different from the code in the target muscle memory, and a major change needs to be made to the code in the target muscle memory, and step 804 is executed.

[0206] 802. Based on the code difference representation that the operation logic has not changed and the code change rule, control the large language model to modify the parameters of the code in the target muscle memory;

[0207] Among them, if the code distinction represents that the operation logic has not changed and the code change rules, it represents that there are corresponding content changes in the code in the target muscle memory and the code of the target task. The corresponding changed content can be summarized as parameters, and the code can be modified by adding parameters. The added parameters can be used to achieve compatibility between the code of the target task and the code in the target muscle memory.

[0208] In a possible implementation, the code is reconstructed based on the code of the target task and the code of the target muscle memory, a parameter is added to the reconstructed code, and a step of obtaining a value corresponding to the parameter is added to the reconstructed code.

[0209] As an example, the code distinction is the position change and size change of the window, and in the target muscle memory the code increases the position parameter and size parameter of the window.

[0210] As an example, when refactoring the code, add two parameters: window position and window size; add a step in the code: get the window position and size.

[0211] The code after the parameter addition is used to adjust the code in the target muscle memory so that the code in the target muscle memory can include the code after the parameter addition.

[0212] 803. Based on the code distinction characterizing that the operation logic has not changed and the code changes irregularly, the large language model is controlled to determine a parameter selection small model, the parameter selection small model is used to modify the code in the target muscle memory, and the parameter selection small model is added to the target muscle memory;

[0213] Among them, if the code difference represents that the operation logic has not changed, but the change is irregular, a small parameter selection model can be introduced to execute the irregularly changing part.

[0214] In a possible implementation, a basic small model is first loaded, and the basic small model is trained through the execution process of the target task and the execution process of each historical task corresponding to the target muscle memory to obtain a parameter selection small model.

[0215] In the specific implementation, after the small model is determined, each time the task is executed, as long as it involves calling the small model, the corresponding scene characteristics of the task are recorded. When the number of records reaches the set threshold, the small model is trained using the recorded content to obtain the parameter selection small model, so that the parameter selection small model can be called before each specific step is executed.

[0216] Among them, the scene feature can be scene-related features such as the target description information of the task, which are input into the small model and the specific parameter values ​​required for the parameterized code are output. After multiple trainings, a parameter selection small model can be obtained.

[0217] Accordingly, after the parameter selection small model is determined, the corresponding part of the code in the target muscle memory code is modified to call the parameter selection small model when executing the part of the code.

[0218] As an example, the small model is used as a model in executing "open unread mails", and its input can be a list of mails read from the UI, and the small model returns the unread mail ids.

[0219] Among them, a muscle memory may include one or more parameter selection small models, which can call the small model to perform processing tasks, achieving fast processing with high accuracy.

[0220] As an example, the target task is to select the nearest holiday. Since the position of the holiday icon represented by clicking in the interface is not fixed, it is impossible to determine the position rule of the click operation. Therefore, a parameter selection model can be set to select the clicked icon, and the code in the target muscle memory is also modified accordingly. The parameter selection model will be called before the click operation to obtain the position or type of the target icon. After the parameter selection model is run, the subsequent code will be executed in combination with the running result of the parameter selection model.

[0221] 804. Based on the code distinguishing the change in the operation logic, the large language model is controlled to replace the code in the target muscle memory with the code of the target task.

[0222] Among them, when the operating logic of the code changes, the problem can no longer be solved by simply adding parameters, adding parameters to select small models, etc.

[0223] Among them, if the code of the target task is completely different from the code operation logic in the target muscle memory, it means that the code in the target muscle memory cannot be reused for the target task. It is necessary to abandon the code in the target muscle memory and directly adopt a new operation logic code. The newly generated code of the large model (the code of the target task) is used as the basis to replace the code in the target muscle memory to achieve the purpose of updating the muscle memory.

[0224] Correspondingly, the target task is executed by muscle memory, which is an intelligent agent architecture that includes parameterized code and a small model. The large language model can determine whether muscle memory can be used to execute the target task and whether the target task is successfully executed based on the execution result. It acts as a code generator and supervisor and does not need to participate in every step of each task.

[0225] In a possible implementation, if the target task is a repeatable task, after updating the target muscle memory, the target task can be executed again using the updated target muscle memory to verify the code in the updated target muscle memory.

[0226] Among them, it is determined whether the updated target muscle memory successfully executes the target task to verify whether the update is successful. If successful, the verification result indicates that the update is successful, and the updated target muscle memory is retained; if unsuccessful, the verification result indicates that the update failed, and the large language model can be called to re-execute the target task, and the code in the target muscle memory is updated based on the code obtained by the successful execution.

[0227] In a possible implementation, if the target task is a non-repeatable task, after updating the target muscle memory, the subsequent received task is executed. When the target muscle memory is used to execute the task, whether the update is successful is verified based on the execution result. If the task is executed successfully, the representation verification result is determined to be the success of the update, and the updated target muscle memory is retained; if it is unsuccessful, the representation verification result is the failure of the update, and the large language model can be called to re-execute the target task, and the code in the target muscle memory is updated based on the code obtained by the successful execution.

[0228] In this embodiment, the large language model is controlled to compare the code of the target task with the code in the target muscle memory to obtain a code difference; based on the code difference characterizing that the operation logic has not changed and the code changes regularly, the large language model is controlled to modify the parameters of the code in the target muscle memory; based on the code difference characterizing that the operation logic has not changed and the code changes irregularly, the large language model is controlled to determine the parameter selection small model, the parameter selection small model is used to modify the code in the target muscle memory, and the parameter selection small model is added to the target muscle memory; based on the code difference characterizing the operation logic change, the large language model is controlled to replace the code in the target muscle memory with the code of the target task, so that different methods are used to update the target muscle memory for different code changes. Moreover, since the present solution is to perform progressive optimization on the code in the muscle memory, the operation mode of each muscle memory is solidified, ensuring that the operation mode is stable when it can execute tasks, and through parameterized design, the balance between flexibility and efficiency is guaranteed, so that when using muscle memory to execute tasks, it can be compatible with more tasks of the same type and can also obtain stable operation results.

[0229] An automated operation method provided by an embodiment of the present application is introduced above, and a device for executing the automated operation method is introduced below.

[0230] See also Fig. 9 , Fig. 9 This is a schematic diagram of the structure of an automated operating device provided in an embodiment of the present application. Fig. 9 As shown, the automatic operation device 900 comprises:

[0231] Memory 901, used to store a muscle memory library, which includes a plurality of muscle memories, and the muscle memories are generated when the large language model successfully performs a corresponding task;

[0232] Interface 902, used to receive a target task;

[0233] Processor 903 is used to determine target description information based on the target task; query the muscle memory library based on the target description information to obtain a muscle memory list, where the muscle memory list includes at least one muscle memory; control the large language model to determine a matching result with the target task based on each muscle memory in the muscle memory list; execute the target task based on the matching result to obtain an execution result.

[0234] In one possible implementation, the processor includes:

[0235] A determination module, used to determine target description information according to the target task;

[0236] A query module, configured to query a muscle memory library according to the target description information to obtain a muscle memory list, wherein the muscle memory list includes at least one muscle memory;

[0237] A control module, used for controlling the large language model to determine a matching result with the target task according to each muscle memory in the muscle memory list;

[0238] The execution module is used to execute the target task according to the matching result to obtain an execution result.

[0239] In a possible implementation, a module is determined, specifically for:

[0240] Inputting at least text information included in the target task into a large language model, and the large language model is capable of generating target description information including at least the text information;

[0241] Correspondingly, the query module is specifically used for:

[0242] A muscle memory list consisting of at least one muscle memory is obtained by screening according to the target description information and the description information of each muscle memory in the muscle memory library.

[0243] In a possible implementation, the execution module includes:

[0244] A first execution unit is used to indicate, according to the matching result, that there is a target muscle memory matching the target task in the muscle memory list, and to execute the target task according to the target muscle memory to obtain a first execution result;

[0245] The second execution unit is used to control the large language model to execute the target task according to the matching result indicating that the target muscle memory does not exist in the muscle memory list, so as to obtain a second execution result.

[0246] In a possible implementation, the method further includes:

[0247] A judgment module is used to judge whether the target task is successfully executed according to the execution result and obtain a judgment result;

[0248] The updating module is used to indicate that the target task is successfully executed based on the judgment result, and to update the muscle memory bank according to the execution result.

[0249] In a possible implementation, the judging module includes:

[0250] A receiving unit, used to receive feedback information, where the feedback information is information input by the operator in response to the execution result;

[0251] The first judgment unit is used to judge whether the target task is successfully executed according to the feedback information.

[0252] In a possible implementation, the updating module includes:

[0253] A first control unit, configured to control the large language model to generate code to complete the target task according to the second execution result;

[0254] A second control unit is used to control the large language model to determine parameterizable items in the code of the target task to obtain parameterized code;

[0255] The generation unit is used to generate muscle memory corresponding to the target task according to the target description information and parameterized code.

[0256] In a possible implementation, the updating module includes:

[0257] A third control unit is used to determine, based on the first execution result, that the target task has not been successfully executed, and control the large language model to execute the target task to obtain a third execution result;

[0258] A second judgment unit is used to judge whether the target task is successfully executed according to the third execution result, and obtain a third judgment result;

[0259] The updating unit is used to indicate that the target task is successfully executed based on the third judgment result, and to update the muscle memory bank according to the third execution result.

[0260] In a possible implementation, the updating unit includes:

[0261] A control subunit, configured to control the large language model to generate code to complete the target task based on the third execution result;

[0262] The updating subunit is used to update the code in the target muscle memory according to the code of the target task.

[0263] In a possible implementation, the updating subunit is specifically used for:

[0264] Control the large language model to compare the code of the target task with the code in the target muscle memory to obtain the code difference;

[0265] Based on the difference in the code representation that the operation logic has not changed and the code change rule, the large language model is controlled to modify the parameters of the code in the target muscle memory;

[0266] Based on the code difference characterizing that the operation logic has not changed and the code changes irregularly, the large language model is controlled to determine the parameter selection small model, the parameter selection small model is used to modify the code in the target muscle memory, and the parameter selection small model is added to the target muscle memory;

[0267] Based on the code distinction representing the change in operation logic, the large language model is controlled to replace the code in the target muscle memory with the code of the target task.

[0268] It should be noted that for the functional explanation of each component structure in an automated operation device provided in an embodiment of the present application, please refer to the explanation in the aforementioned method embodiment, which will not be repeated in this embodiment.

[0269] In this embodiment, it includes: a memory for storing a muscle memory library, the muscle memory library includes a number of muscle memories, and the muscle memories are generated when the large language model successfully executes the corresponding task; a first interface for receiving a target task; a processor for determining target description information according to the target task; querying the muscle memory library according to the target description information to obtain a muscle memory list, the muscle memory list including at least one muscle memory; controlling the large language model to determine the matching result with the target task according to each muscle memory in the muscle memory list; executing the target task according to the matching result to obtain an execution result. In the execution process, the action of executing the target task using muscle memory does not require the large language model to make high-level decisions, and does not require the large language model to execute step by step for each action step of each task. Therefore, it is not necessary to call the large language model once for each step operation, which reduces the time required for task execution and improves the efficiency of task execution.

[0270] The present application also provides an electronic device in an embodiment. Fig.10 As shown, it shows a schematic diagram of the structure of an electronic device suitable for implementing the automated operation method in the embodiment of the present application. The electronic device in the embodiment of the present application may include but is not limited to fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Fig.10 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0271] like Fig.10 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1008 to a random access memory (RAM) 1003. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 1003. The processing device 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0272] Typically, the following devices may be connected to the I / O interface 1005: an input device 1006 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 1007 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1008 including, for example, a memory card, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Fig.10 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0273] An embodiment of the present application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the automated operation methods provided in the embodiments of the present application.

[0274] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any automated operation method provided in the embodiment of the present application.

[0275] It should also be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0276] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0277] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0278] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a training device, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

Claims

1. An automated operation method, comprising: Receive target tasks; Determining target description information according to the target task; According to the target description information, query a muscle memory library to obtain a muscle memory list, wherein the muscle memory library includes a plurality of muscle memories, wherein the muscle memories are generated when the large language model successfully performs a corresponding task, and the muscle memory list includes at least one muscle memory; Controlling the large language model to determine a matching result with the target task according to each muscle memory in the muscle memory list; The target task is executed according to the matching result to obtain an execution result.

2. The automated operation method according to claim 1, wherein determining target description information according to the target task comprises: Inputting at least text information included in the target task into a large language model, wherein the large language model is capable of generating target description information including at least the text information; Correspondingly, according to the target description information, the muscle memory library is queried to obtain a muscle memory list, including: A muscle memory list consisting of at least one muscle memory is obtained by screening according to the target description information and the description information of each muscle memory in the muscle memory library.

3. The automated operation method according to claim 1, wherein executing the target task according to the matching result to obtain the execution result comprises: According to the matching result, it is indicated that there is a target muscle memory matching the target task in the muscle memory list, and the target task is executed according to the target muscle memory to obtain a first execution result; According to the matching result indicating that the target muscle memory does not exist in the muscle memory list, the large language model is controlled to execute the target task to obtain a second execution result.

4. The automated operation method according to claim 3, wherein after executing the target task according to the matching result and obtaining the execution result, the method further comprises: According to the execution result, judging whether the target task is executed successfully and obtaining a judgment result; The target task is successfully executed based on the judgment result, and the muscle memory bank is updated according to the execution result.

5. The automated operation method according to claim 4, wherein judging whether the target task is successfully executed based on the execution result to obtain the judgment result comprises: receiving feedback information, wherein the feedback information is information input by an operator in response to the execution result; Based on the feedback information, it is determined whether the target task is successfully executed.

6. The automated operation method according to claim 4, wherein the step of updating the muscle memory according to the execution result comprises: According to the second execution result, controlling the large language model to generate code to complete the target task; Controlling the large language model to determine parameterizable items in the code of the target task to obtain parameterized code; Muscle memory corresponding to the target task is generated based on the target description information and parameterized code.

7. The automated operation method according to claim 4, wherein the step of indicating that the target task is successfully executed based on the judgment result and updating the muscle memory base according to the execution result comprises: According to the first execution result, determining that the target task is not successfully executed, controlling the large language model to execute the target task, and obtaining a third execution result; According to the third execution result, judging whether the target task is executed successfully, and obtaining a third judgment result; The target task is successfully executed based on the third judgment result, and the muscle memory bank is updated according to the third execution result.

8. The automated operation method according to claim 7, wherein updating the muscle memory bank according to the third execution result comprises: Based on the third execution result, controlling the large language model to generate code to complete the target task; According to the code of the target task, the code in the target muscle memory is updated.

9. The automated operation method according to claim 8, wherein updating the code in the target muscle memory according to the code of the target task comprises: Controlling the large language model to compare the code of the target task with the code in the target muscle memory to obtain the code difference; Based on the code difference representation that the operation logic has not changed and the code change rule, the large language model is controlled to modify the parameters of the code in the target muscle memory; Based on the code difference characterizing that the operation logic has not changed and the code changes irregularly, the large language model is controlled to determine the parameter selection small model, the code in the target muscle memory is modified by using the parameter selection small model, and the parameter selection small model is added to the target muscle memory; Based on the code distinction representing the change in operation logic, the large language model is controlled to replace the code in the target muscle memory with the code of the target task.

10. An automated operating device, comprising: A memory, used to store a muscle memory library, wherein the muscle memory library includes a plurality of muscle memories, and the muscle memories are generated when the large language model successfully performs a corresponding task; Interface, used to receive target tasks; A processor, configured to determine target description information according to the target task; According to the target description information, a muscle memory library is queried to obtain a muscle memory list, wherein the muscle memory list includes at least one muscle memory; a large language model is controlled to determine a matching result with the target task according to each muscle memory in the muscle memory list; and the target task is executed according to the matching result to obtain an execution result.

Images