Task orchestration methods, devices, storage media, and computer program products based on large language models.
By using a large language model to identify and verify the intent of natural language adjustment instructions, generating adjustment intent declaration files and performing modification analysis, the problem of large language models being unable to handle large-scale task orchestration plans is solved, thereby improving the reliability and efficiency of task orchestration.
Patent Information
- Application Number
- CN202611592922.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-14
- Publication Date
- 2026-11-17
AI Technical Summary
The context window of a large language model is limited by hardware resources and cannot effectively handle large-scale full-task orchestration plans, making it difficult to reliably complete the adjustment operations of large-scale structured data in multi-organization and multi-task scenarios such as enterprise financial month-end closing.
By using a large language model to identify the intent of natural language adjustment instructions, an adjustment intent declaration file is generated. Based on this file, the initial task orchestration plan is modified. The credibility of the modification is verified by combining the modification analysis report to ensure the legality and interpretability of the modification, thus avoiding the need to load the full data for verification.
It improves the reliability and efficiency of task orchestration without loading the full task orchestration plan, ensures the legality and interpretability of modification results, and avoids processing difficulties caused by data size exceeding the upper limit of the context window.
Smart Images

Figure CN123115358A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer data processing technology, and in particular to a task orchestration method, device, readable storage medium, and computer program product based on a large language model. Background Technology
[0002] With the development of enterprise information technology, task orchestration technology has emerged. This technology organizes multiple tasks in a business process according to their execution order, dependencies, and hierarchical affiliation, forming a structured task orchestration plan to guide the orderly execution of the business process. Taking the enterprise financial month-end closing scenario as an example, the financial month-end closing is a core business process that is executed cyclically every month. It is necessary to orchestrate the various tasks in the closing process, clarifying the execution time, prerequisites, execution objects, and hierarchical affiliation of each task, so as to ensure that the month-end closing process can proceed in an orderly manner and be completed on time.
[0003] In related technologies, adjustments to task orchestration plans can be done manually. After the initial orchestration is completed, users need to adjust the task orchestration plan according to business changes, such as adjusting task execution times, changing task execution targets, adding or deleting tasks, etc.
[0004] With the development of large language model technology, methods have emerged that utilize large language models to assist in task orchestration and adjustment. This method typically loads the entire task orchestration plan into the context of the large language model, which then understands and modifies the task data based on the user's natural language adjustment instructions, and writes the modified data back.
[0005] The above-mentioned method of using large language models to assist in task orchestration adjustment has the following technical problems: the context window of the large language model is limited by hardware resources and has a physical upper limit, while the data scale of the full task orchestration plan is often large. Especially in multi-organization and multi-task scenarios such as enterprise financial month-end closing, the data scale of the full task orchestration plan may exceed the upper limit of the context window of the large language model, which makes it impossible for the large language model to hold the full data in the context and makes it difficult to reliably complete the adjustment operation of large-scale structured data. Summary of the Invention
[0006] Therefore, it is necessary to provide a task orchestration method, device, storage medium, and computer program product based on a large language model that can improve orchestration efficiency and reliability in response to the above-mentioned technical problems.
[0007] Firstly, this application provides a task orchestration method based on a large language model, the method comprising:
[0008] Obtain natural language adjustment instructions for the initial task orchestration plan of the task to be orchestrated;
[0009] The intent of the natural language adjustment instructions is identified by a large language model to obtain an adjustment intent declaration file for the initial task orchestration plan;
[0010] Based on the adjustment intent statement file, the initial task orchestration plan is modified to obtain the modified orchestration plan;
[0011] Based on the adjustment intent statement document and the initial task orchestration plan, the modifications to the revised orchestration plan compared to the initial task orchestration plan are analyzed to obtain a modification analysis report; the modification status includes the legality and explainability of the modification;
[0012] The credibility of the modified orchestration plan is verified using the large language model based on the modification analysis report.
[0013] If the credibility verification of the modified orchestration plan passes, the modification analysis report and the modified orchestration plan are returned.
[0014] Secondly, this application also provides a task orchestration apparatus, the apparatus comprising:
[0015] The acquisition module is used to acquire natural language adjustment instructions for the initial task orchestration plan of the task to be orchestrated;
[0016] The recognition module is used to perform intent recognition on the natural language adjustment instructions through a large language model to obtain an adjustment intent declaration file for the initial task orchestration plan;
[0017] The modification module is used to modify the initial task orchestration plan based on the adjustment intent declaration file to obtain the modified orchestration plan;
[0018] The analysis module is used to analyze the modifications made to the modified orchestration plan compared to the initial task orchestration plan based on the adjustment intent declaration file and the initial task orchestration plan, and to obtain a modification analysis report; the modification information includes the legality and interpretability of the modification.
[0019] The verification module is used to verify the credibility of the modified orchestration plan based on the modification analysis report using the large language model;
[0020] The return module is used to return the modification analysis report and the modified orchestration plan if the credibility verification of the modified orchestration plan passes.
[0021] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps included in any of the foregoing method embodiments.
[0022] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps included in any of the foregoing method embodiments.
[0023] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps included in any of the foregoing method embodiments.
[0024] The aforementioned task orchestration method, apparatus, device, storage medium, and computer program product based on a large language model (MLM) obtains an adjustment intent declaration file by recognizing the intent of natural language adjustment instructions using the MLM. Based on this file, the initial task orchestration plan is modified. The modified orchestration plan is then analyzed using both the MLM and the initial orchestration plan, resulting in a modification analysis report that includes verification of the modification's legality and interpretability. The MLM then verifies the reliability of the modified orchestration plan based on this report. If verification is successful, the MLM report and the modified orchestration plan are returned. Since the MLM does not require loading the full task orchestration plan throughout the process, its context consumption is independent of data size, thus solving the problem of unreliable large-scale structured data adjustments due to data exceeding the context window limit. Furthermore, because the modification analysis covers both legality and interpretability, it provides a basis for the MLM to perform reliability verification using only a small modification analysis report, ensuring reliable confirmation of the adjustment results and improving the reliability and efficiency of task orchestration adjustments. Attached Figure Description
[0025] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0026] Figure 1 This is a diagram illustrating the application environment of a task orchestration method based on a large language model in one embodiment.
[0027] Figure 2 This is a flowchart illustrating a task orchestration method based on a large language model in one embodiment;
[0028] Figure 3 This is a structural block diagram of a task orchestration device based on a large language model in one embodiment;
[0029] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0031] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.
[0032] The task orchestration method based on a large language model provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.
[0033] In one exemplary embodiment, such as Figure 2 As shown, a task orchestration method based on a large language model is provided. This method can be applied to terminal 102 or server 104. Figure 1Taking server 104 as an example, the following steps are included:
[0034] Step 202: Obtain natural language adjustment instructions for the initial task orchestration plan for the task to be orchestrated.
[0035] In this context, "tasks to be orchestrated" refers to multiple tasks that require orchestration, and these tasks may have interrelationships. Specifically, tasks to be orchestrated can be financial month-end closing tasks, and the initial task orchestration plan is the financial month-end closing task orchestration plan. Financial month-end closing tasks refer to the various tasks included in the monthly closing process executed by the enterprise, such as general ledger closing, accounts receivable closing, accounts payable closing, tax closing, depreciation accrual, and report generation. Understandably, after the initial orchestration, users may need to perform multiple rounds of interactive adjustments, such as "delay general ledger closing by 2 hours," "do not orchestrate the Accounts Payable (AP) module this month," or "move the critical path forward by one day," before final confirmation and database entry.
[0036] The initial task orchestration plan refers to the plan data obtained after orchestrating the tasks to be orchestrated. It may include information such as the task's identifier, name, affiliated organization, business system, start time, end time, execution object, prerequisite dependencies, and hierarchical inclusion relationships between tasks. Natural language adjustment commands refer to the instructions entered by the user in natural language form to adjust the initial task orchestration plan.
[0037] Step 204: Use a large language model to identify the intent of the natural language adjustment instructions and obtain the adjustment intent declaration file corresponding to the initial task orchestration plan.
[0038] The adjustment intent declaration file is a structured file used to declare the intention of this adjustment. The adjustment intent declaration file only declares "what adjustments to make, which data to make, and to what extent", and does not include the full task orchestration plan itself to be adjusted.
[0039] Specifically, the natural language adjustment instructions are parsed using a large language model to identify the various elements of the adjustment intent and convert them into an adjustment intent declaration file consisting of a description field and at least one adjustment item. Each adjustment item may include an adjustment type, an adjustment scope, and adjustment parameters. The adjustment type indicates the type of operation performed on the data; the adjustment scope indicates the subset of data targeted by the operation, such as specifying an organizational list and filtering conditions; and the adjustment parameters indicate the specific value of the operation, such as offset in minutes or the identifier of the target object.
[0040] For example, when a user enters "Delay the closing of the general ledger for all organizations by 2 hours", the large language model recognizes that the adjustment type is time offset, the adjustment range is tasks with the task code prefix "GL_CLOSE" in all organizations, and the adjustment parameter is a delay of 120 minutes, thereby generating the corresponding adjustment intent declaration file.
[0041] It should be noted that in this embodiment, the large language model does not load the full task orchestration plan; it only generates an adjustment intent declaration file based on natural language adjustment instructions and necessary context information. Therefore, the context consumption of the large language model does not increase with the data size of the full task orchestration plan, thus avoiding the problem of being unable to process data due to the data size exceeding the upper limit of the context window.
[0042] Step 206: Modify the initial task orchestration plan based on the adjustment intent statement file to obtain the modified orchestration plan.
[0043] Specifically, based on the adjustment scope and parameters in the adjustment intent declaration file, the task data that needs to be modified is determined, and corresponding modification operations are performed on the determined task data according to the adjustment type. Modification operations may include time adjustments, task additions and deletions, dependency rewriting, and changes to execution objects. After the modifications are completed, the server writes the modified data back, resulting in the modified orchestration plan.
[0044] It should be noted that this modification operation can be performed by a deterministic program, thus ensuring the determinism and repeatability of the modification result. A deterministic program is one whose execution result does not depend on random factors, and the same input will always produce the same output. Compared with directly modifying data line by line by a large language model, the modification operation performed by a deterministic program will not encounter probabilistic errors, and the modification result can be fully and repeatedly verified.
[0045] Step 208: Based on the adjustment intent statement document and the initial task orchestration plan, analyze the changes in the modified orchestration plan compared to the initial task orchestration plan, and obtain a modification analysis report; the modification details include the legality and interpretability of the modifications.
[0046] Modification legitimacy is used to characterize whether the modified orchestration plan itself meets the preset legitimacy requirements. Modification interpretability is used to characterize whether each data modification that occurs in the modified orchestration plan can be reasonably explained, that is, whether the source of the modification can be clearly identified.
[0047] Specifically, the legality of the modification can be obtained by performing a compliance check on the modified orchestration plan, and the interpretability of the modification can be obtained by performing attribution analysis on every data modification that occurs in the modified orchestration plan relative to the initial task orchestration plan.
[0048] Step 210: Verify the credibility of the modified orchestration plan based on the modification analysis report using a large language model.
[0049] The large language model receives modification analysis reports and verifies the credibility of the modified orchestration plan based on the legality and interpretability of the modifications in the reports. If both the legality and interpretability of the modifications meet the preset conditions, the large language model determines that the credibility verification of the modified orchestration plan has passed; if the legality or interpretability of the modifications does not meet the preset conditions, the large language model determines that the credibility verification of the modified orchestration plan has failed.
[0050] Considering that the amount of data in the modified analysis report can be relatively small (e.g., no more than 2KB), the large language model only needs to complete the credibility verification based on the modified analysis report, without having to load the full data of the modified orchestration plan for verification item by item, thereby effectively improving the efficiency of task orchestration processing.
[0051] Step 212: If the credibility verification of the modified orchestration plan passes, return the modification analysis report and the modified orchestration plan.
[0052] If the credibility verification of the modified orchestration plan passes, the returned content may include a natural language description of the modified analysis report and the file path corresponding to the modified orchestration plan, allowing users to understand the specifics of this adjustment. Since the large language model only needs to process the relatively small amount of data in the adjustment intent declaration file and the modified analysis report during each round of adjustments, the context consumption of multiple rounds of adjustments will not cumulatively increase with the number of adjustment rounds and the data scale. Users can confirm the results of this adjustment based on the returned content and decide whether to submit new adjustment instructions.
[0053] Optionally, if the credibility verification of the modified orchestration plan fails, the large language model can take appropriate action based on the problem type indicated in the modification analysis report. For example, if the legality of the modification does not meet the preset conditions, it indicates that the modified orchestration plan itself contains illegal data. The large language model can locate the illegal data based on the modification analysis report and regenerate the adjustment intent declaration file accordingly to trigger a new round of modifications. If the interpretability of the modification does not meet the preset conditions, it indicates that there are modifications in the modified orchestration plan with unexplained sources. The large language model can load the modified orchestration plan to verify the unexplained modifications. Through the above methods, the system can continue to run and gradually converge to a credible result even if the credibility verification fails, avoiding process interruption.
[0054] This application embodiment uses a large language model to identify the intent of natural language adjustment instructions to obtain an adjustment intent declaration file. Based on the adjustment intent declaration file, the initial task orchestration plan is modified. The modified orchestration plan is then analyzed based on both the adjustment intent declaration file and the initial task orchestration plan, resulting in a modification analysis report that includes the legality and interpretability of the modifications. The large language model then verifies the credibility of the modified orchestration plan based on this modification analysis report. If the verification passes, the modification analysis report and the modified orchestration plan are returned. Since the large language model does not need to load the full task orchestration plan throughout the process, its context consumption is independent of the data size, thus solving the problem of unreliable large-scale structured data adjustment due to data size exceeding the context window limit. Simultaneously, because the modification analysis covers both the legality and interpretability of the modifications, it provides a basis for the large language model to complete credibility verification based on a relatively small modification analysis report, thereby ensuring that the large language model can reliably confirm the adjustment results and improving the reliability and efficiency of task orchestration adjustments.
[0055] In some embodiments, the tasks to be orchestrated include multiple tasks with task relationships; the initial task orchestration plan is modified based on the adjustment intent declaration file to obtain a modified orchestration plan, including:
[0056] The initial task orchestration plan is modified according to the adjustment intent declaration file by a preset script program to obtain an intermediate orchestration plan;
[0057] Anomalies in the intermediate orchestration plan are identified by a script program based on constraints corresponding to task relationships, and the identification results are obtained.
[0058] The intermediate orchestration plan is repaired based on the recognition results using a script program, resulting in a modified orchestration plan.
[0059] Task relationships refer to the relationships between tasks to be orchestrated that affect the order of task execution or task attribution. Task relationships can include prerequisite dependencies and hierarchical inclusion relationships. A prerequisite dependency means that one task can only begin execution after another task is completed; for example, the "General Ledger Closing" task can only begin after the "Voucher Posting" task is completed. A hierarchical inclusion relationship refers to the inclusion relationship between a task set and its subtasks; for example, the "Finance Sector Task Set" includes the "General Ledger Closing" task and the "Accounts Receivable Closing" task, and the time range of the parent task set needs to cover the time ranges of all its subtasks.
[0060] A script program is a program composed of deterministic code used to perform data modification operations. An intermediate orchestration plan refers to the plan data after the script program has modified the initial task orchestration plan according to the adjustment intent declaration file, but before anomaly identification and data repair have been performed.
[0061] Specifically, the script reads the initial task orchestration plan, parses each adjustment item in the adjustment intent declaration file, determines the task data that needs to be modified based on the adjustment scope and parameters, and performs corresponding modification operations on the determined task data according to the adjustment type. Modification operations may include, for example: offsetting the start and end times of matching tasks by a specified number of minutes; setting the start and end times of matching tasks to specified values; shortening the end time of matching tasks by a compression ratio; removing matching tasks from the task array; adding new tasks to a specified organization; replacing the execution object identifier of matching tasks; and adding, deleting, or replacing the pre-dependency fields of matching tasks. After performing the above modifications on the initial task orchestration plan, the script obtains the intermediate orchestration plan.
[0062] Because scripts are deterministic, they will always produce the same output for the same input, thus the modification results can be repeatedly verified. Compared to directly modifying data line by line by line using a large language model, scripts do not introduce probabilistic errors and are therefore more reliable.
[0063] The constraints corresponding to task associations refer to the constraints imposed on task data to ensure that the task associations hold true. Specifically, for prerequisite dependencies, constraints can include prerequisite dependency reference validity constraints and timing constraints. Precursor dependency reference validity constraints require that a task's prerequisite references must point to valid, existing tasks, while timing constraints require that the end time of a prerequisite task cannot be greater than the start time of the subsequent task. A prerequisite dependency reference refers to the reference information in the task data used to indicate its prerequisite tasks; for example, it can be represented as a list of prerequisite task identifiers, an array of prerequisite task objects, or a single prerequisite task identifier.
[0064] Correspondingly, for hierarchical inclusion relationships, constraints can include time coverage constraints, which require that the end time of the parent task set must cover the maximum end time of all its child tasks.
[0065] Since scripts may alter the relationships between tasks when performing modification operations—for example, deleting a task may invalidate the prerequisite dependencies of other tasks, or delaying a subtask may cause the time range of the parent task set to no longer cover the subtask—scripts are required to identify and repair inconsistent data resulting from modifications.
[0066] The identification result refers to the result obtained by the script program after identifying abnormal data in the intermediate orchestration plan based on constraints, indicating the abnormal data present in the intermediate orchestration plan. The identification result can include dependency relationship anomalies and task execution time anomalies.
[0067] Specifically, the script iterates through the task data in the intermediate orchestration plan, checking at least one of the following according to the constraints corresponding to the task relationships: whether the task's predecessor dependency references point to valid, whether the end time of the parent task set covers the maximum end time of all its child tasks, and whether the end time of the predecessor task is greater than the start time of the subsequent task. For data that does not meet the constraints, the script identifies it as anomalous data and records it in the identification results. The identification results record the type of anomalous data, such as dependency relationship anomalies or task execution time anomalies, the location of the anomaly (e.g., the task identifier involved), and the specific content of the anomaly.
[0068] The script program performs corresponding data repair operations on the intermediate orchestration plan based on the abnormal data recorded in the identification results. The goal of the data repair operation is to ensure that the repaired plan data once again meets the constraints corresponding to the task relationships. After the repair is completed, the script program uses the repaired plan data as the modified orchestration plan. For example, in a financial month-end closing scenario, after the script program modifies the initial task orchestration plan, abnormal data such as dangling prerequisite references, insufficient time coverage of the parent task set, and timing conflicts may appear in the intermediate orchestration plan. After identifying the above abnormal data, the script program performs the corresponding repair operations to ensure that the modified orchestration plan once again meets the constraints corresponding to the task relationships.
[0069] In some embodiments, task association relationships include prerequisite dependencies and hierarchical inclusion relationships; the identification results include abnormal dependency data and abnormal task execution time data.
[0070] Anomaly identification is performed on the intermediate orchestration plan using a script program based on constraints corresponding to task relationships. The identification results include:
[0071] The script program identifies the references to the predecessors of failed tasks in the intermediate orchestration plan based on the predecessor dependencies, and obtains dependency anomaly data.
[0072] The script program performs time coverage detection on the execution end time of tasks with hierarchical inclusion relationships in the intermediate orchestration plan based on the hierarchical inclusion relationship, and obtains time coverage abnormal data as task execution time abnormal data.
[0073] The script program performs timing conflict detection on the start time of tasks with pre-dependent relationships in the intermediate orchestration plan, and obtains timing conflict anomaly data as task execution time anomaly data.
[0074] In this context, "invalidated tasks" refers to tasks that have been deleted, disabled, or repeatedly cleared in the intermediate orchestration plan. Since modification operations may include deleting, disabling, or repeatedly clearing tasks, the set of valid tasks in the intermediate orchestration plan may change compared to the initial task orchestration plan. Therefore, when identifying abnormal data in the intermediate orchestration plan, the script can first determine the set of valid tasks in the current intermediate orchestration plan. The script then iterates through all the preceding dependency references of tasks in the intermediate orchestration plan, checking each reference to see if the task it points to exists in the set of valid tasks. If a preceding dependency reference points to a task that does not exist in the set of valid tasks, it indicates that the preceding dependency reference is a dangling reference. The script identifies this as abnormal dependency data and records it in the identification results.
[0075] Precursor dependency references in intermediate orchestration plans may have multiple field formats, such as array format (containing complete predecessor task objects), identifier array format (containing only predecessor task identifiers), and single-value format (containing only a single predecessor task identifier). When identifying them, the script can first extract the various formats of predecessor dependency references into a unified identifier list, and then compare them one by one with the valid task set, thus ensuring compatibility with different field formats.
[0076] Hierarchical containment refers to the containment relationship between a task set and its subtasks. The end time of the parent task set must cover the maximum end time of all its subtasks. Time coverage aberrations occur when the end time of the parent task set does not cover the maximum end time of its subtasks.
[0077] Specifically, the script iterates through all parent-child task sets in the intermediate orchestration plan. For each parent task set, it determines the maximum end time among all its child tasks and compares this maximum end time with the end time of the parent task set itself. If the end time of the parent task set is less than the maximum end time of its child tasks, it indicates that the time range of the parent task set does not cover its child tasks. The script identifies this situation as time coverage anomaly data and records it in the identification results.
[0078] It should be noted that hierarchical containment relationships can be multi-level nested, meaning a parent task set can contain child task sets, and child task sets can contain further child tasks. In multi-level nested scenarios, the script can perform time coverage checks level by level. Since time coverage anomalies in the parent task set may propagate further upwards as time adjustments are made to its child task sets, the script can employ a multi-round scanning approach until no new time coverage anomalies appear in the entire task tree.
[0079] A timing conflict occurs when the completion time of a preceding task is later than the start time of a subsequent task. Timing conflict detection involves checking whether tasks with pre-existing dependencies meet timing requirements. A timing conflict exception occurs when the completion time of a preceding task is later than the start time of a subsequent task.
[0080] Specifically, the script iterates through all task pairs with prerequisite dependencies in the intermediate orchestration plan. For each pair of tasks with prerequisite dependencies, it compares the end time of the prerequisite task with the start time of the subsequent task. If the end time of the prerequisite task is greater than the start time of the subsequent task, it indicates that there is a timing conflict between the two. The script identifies this situation as a timing conflict exception and records it in the identification results.
[0081] Optionally, time-series conflict anomalies can be categorized according to the severity of the conflict. For example, conflicts with a time difference exceeding a preset threshold can be marked as severe conflicts, while those with a time difference below the preset threshold can be marked as minor conflicts. In subsequent summary output, the script can highlight severe conflicts as key issues, allowing users to quickly identify critical problems.
[0082] In some embodiments, a script program performs data repair on the intermediate orchestration plan based on the identification results to obtain a modified orchestration plan, including:
[0083] Based on dependency anomaly data, remove task data that references failed tasks as prerequisites in the intermediate orchestration plan using scripts;
[0084] Based on time coverage anomaly data, the script program extends the task end time of the parent task set in the intermediate orchestration plan, so that the end time of the parent task set covers the maximum end time of the child tasks.
[0085] The script program eliminates timing conflicts in the intermediate orchestration plan where the end time of the preceding task is later than the start time of the subsequent task, based on timing conflict anomaly data.
[0086] In this context, abnormal dependency data refers to task data whose preceding dependency references point to expired tasks. Task data that uses expired tasks as preceding dependency references refers to data whose preceding dependency references contain the corresponding portion of the data identified as an expired task.
[0087] Specifically, based on the abnormal dependency data recorded in the identification results, the specific preceding dependency reference is located and removed from the corresponding task data. The removal operation only targets the preceding dependency reference pointing to the invalid task and does not affect other preceding dependency references in the task data, nor does it affect other task data.
[0088] Considering that preceding dependency references in the intermediate orchestration plan may have various field formats, the script removes the corresponding preceding task object from the array for array-formatted preceding dependency references (containing the complete preceding task object); removes the corresponding preceding task identifier from the array for identifier-array-formatted preceding dependency references; and clears the entire preceding dependency reference if the identifier points to an invalid task. By handling compatibility issues with different field formats, the script can correctly remove dangling references in various forms.
[0089] Time coverage anomaly data refers to task data where the end time of the parent task set is less than the maximum end time of its child tasks. Based on the time coverage anomaly data recorded in the identification results, the specific parent task set can be located, and its end time can be extended to the maximum end time of all its child tasks. The extension operation only modifies the end time of the parent task set, without changing its start time or any time information of the child tasks.
[0090] Because hierarchical relationships can be multi-level nested, extending the end time of a parent task set might cause time coverage anomalies in its parent task sets as well. Therefore, the script can perform the extension operation using a multi-round scan approach: in each round, it iterates through all parent task sets, extending those whose end times do not cover the maximum end times of their child tasks; if at least one extension operation occurs in this round, it proceeds to the next round; if no extension operation occurs in this round, it indicates that the entire task tree has reached a stable state (i.e., a fixed point), and the script terminates the scan. Through multi-round scans, the script can handle associated extensions in multi-level nested structures, ensuring that the time coverage of the entire task tree is repaired.
[0091] For example, in a financial month-end scenario, the end time of a subtask T2 is adjusted from "06-28 18:00" to "06-28 20:00". The end time of its parent task set P1 remains "06-28 18:00", and the end time of P1's parent task set P2 remains "06-28 19:00". In the first scan, the script extends the end time of P1 to "06-28 20:00". In the second scan, it finds that the end time of P2, "06-28 19:00", is less than the new end time of P1, "06-28 20:00", so it extends the end time of P2 to "06-28 20:00". In the third scan, no further extension operations occur, and the script terminates the scan. After these multiple scans, the end times of P1 and P2 both cover the maximum end time of their subtasks.
[0092] Timing conflict anomaly data refers to task data where the end time of the preceding task is later than the start time of the subsequent task. Based on the timing conflict anomaly data recorded in the identification results, the specific preceding and subsequent task pairs are located, and the timing conflict is eliminated. The method for eliminating timing conflicts can be set according to actual business needs. For example, the script can postpone the start time of the subsequent task to the end time of the preceding task, thereby eliminating the timing conflict; it can also advance the end time of the preceding task to the start time of the subsequent task, thereby eliminating the timing conflict; or it can output the timing conflict for the user to decide, allowing the user to choose the subsequent processing method.
[0093] Considering that the method of resolving timing conflicts may affect other relationships between subsequent or preceding tasks—for example, delaying the start time of a subsequent task may cause its end time to be delayed, potentially leading to time coverage anomalies in its parent task set—the script can re-execute the aforementioned time coverage check after resolving the timing conflicts until no new abnormal data appears in the entire task tree.
[0094] In some embodiments, based on the adjusted intent statement file and the initial task orchestration plan, the modifications to the revised orchestration plan compared to the initial task orchestration plan are analyzed to obtain a modification analysis report, including:
[0095] The legality of the modification is determined by verifying the compliance of the modified orchestration plan's task structure and the task execution objects through a preset execution engine.
[0096] The execution engine determines the fields to be modified based on the adjustment intent declaration file, and determines the type of reason for modification as the interpretability of the modification data based on the matching of the modified data in the modified orchestration plan with the fields to be modified.
[0097] The execution engine, independent of the script program, is a program module used to validate and attribute the modified orchestration plan. The execution engine reads the modified orchestration plan and performs compliance checks on the task structure and the task execution object. Task structure compliance checks may include: whether the task code is unique, whether the task's affiliated organization is valid, whether the task set is not empty, and whether the task time is within the billing cycle. Task execution object compliance checks may include: whether the execution object exists, whether the execution object has the corresponding permissions, and whether the execution object has been repeatedly assigned. Based on the results of these checks, the execution engine determines the legality of the modification.
[0098] Optionally, the verification results of task structure compliance checks and task execution object compliance checks can be categorized according to the violation level. For example, violation levels can be divided into blocking level, pattern level, and alert level. Blocking level violations refer to violations that will prevent the modified orchestration plan from executing normally, such as duplicate task code or invalid task-assigned organization; pattern level violations refer to violations that belong to a known pattern, do not block execution but need to be logged; alert level violations refer to violations that require user attention but do not affect execution. When determining the legality of a modification, the execution engine can determine that the modification meets the requirements if there is no blocking level violation; otherwise, it can determine that the modification does not meet the requirements.
[0099] The field to be modified refers to the field that needs to be modified, as directly specified in the adjustment intent declaration file. The modification reason type refers to the source type of the modified data, used to characterize the reason for the modification. Modification reason types can include at least one of direct modification, related modification, and accidental modification.
[0100] Specifically, the modification reason type can include at least one of direct modification, related modification, and accidental modification. Direct modification refers to a modification directly specified by the adjustment intent declaration file; related modification refers to a modification not directly specified by the adjustment intent declaration file but triggered by a direct modification; accidental modification refers to a modification that is neither a direct modification nor a related modification. The execution engine determines the interpretability of the modification based on the above matching conditions.
[0101] For example, in a financial month-end closing scenario, the intention statement document instructs to "delay the general ledger closing for all organizations by 2 hours." The corresponding instruction modification fields are the start and end times of the general ledger closing task. The execution engine matches the modified data in the revised orchestration plan with the instruction modification fields: if the start and end times of the general ledger closing task are modified, these fields belong to the instruction modification fields, and the modification reason type is determined to be direct modification; if the end time of the parent task set is extended because the parent task set needs to cover the maximum end time of the child tasks, these fields do not belong to the instruction modification fields, but are related to the time adjustment of the general ledger closing task, and the modification reason type is determined to be related modification; if there are other field modifications unrelated to this adjustment, the modification reason type is determined to be unexpected modification.
[0102] In some embodiments, the adjustment intent declaration file includes the adjustment type and adjustment scope; the execution engine determines the indicated modification field based on the adjustment intent declaration file, and determines the modification reason type of the modified data based on the matching between the modified data in the modified orchestration plan and the indicated modification field, including:
[0103] The execution engine determines the instruction modification fields specified in the adjustment intent declaration file based on the adjustment type and adjustment scope.
[0104] The execution engine determines the modification reason type of the modified data in the modified orchestration plan that matches the indicated modification field as a direct modification.
[0105] The execution engine determines the modification reason type of modified data that is not directly modified but has a relationship with the field indicating modification as a related modification;
[0106] The execution engine identifies unexpected modifications as those that are not directly modified or related to other modifications.
[0107] The adjustment type indicates the type of operation performed on the data. Adjustment types can include time adjustment, task addition / deletion, dependency rewrite, and object change. The adjustment scope indicates the subset of data targeted by the operation and can include organizational lists and filtering conditions. The field to be modified refers to the field directly specified in the adjustment intent declaration file that needs to be modified.
[0108] The execution engine parses each adjustment item in the adjustment intent declaration file, determines the set of fields corresponding to the adjustment type based on the adjustment type, and determines the subset of tasks targeted by the adjustment type based on the adjustment scope. The intersection of the two is the instruction field to be modified for the adjustment specified by the adjustment item.
[0109] It should be noted that the field sets corresponding to different adjustment types can be pre-mapped. For example, the field set corresponding to the time offset type can include start and end times; the field set corresponding to the time setting type can include start and end times; the field set corresponding to the time compression type can include end time; the field set corresponding to the execution object change type can include the execution object identifier; and the field set corresponding to the dependency rewrite type can include the preceding dependency identifier. The execution engine can determine the field set corresponding to each adjustment type by querying this mapping relationship.
[0110] Direct modification refers to modifications directly specified by the adjustment intent declaration file. Modified data matching the indicated modification field means that the field containing the modified data belongs to the indicated modification field, and the task containing the modified data belongs to a subset of the tasks corresponding to the indicated modification field.
[0111] The execution engine iterates through the modified data in the revised orchestration plan. For each modified data entry, it determines whether the field it contains belongs to the indicated modification field and whether the task it contains belongs to the subset of tasks corresponding to the indicated modification field. If both conditions are met, the modification reason type for that modified data is determined to be direct modification.
[0112] Related modifications refer to modifications that are not directly specified in the adjustment intent declaration document but are triggered by direct modifications. A relationship exists between modified data and the field indicating modification: although the modified data is not the field indicating modification itself, its cause is related to the modification of the field indicating modification.
[0113] Specifically, the association may include at least one of the following: First, the field containing the modified data is a prerequisite dependency field, and the modification operation specified in the adjustment intent declaration file includes task deletion or task disabling; Second, the field containing the modified data is a name field, and the corresponding identifier field in the modification operation specified in the adjustment intent declaration file is modified; Third, the field containing the modified data is an end time field, and the task containing the modified data does not match any adjustment range in the adjustment intent declaration file.
[0114] For the first type of relationship, when a modification operation deletes a task, the prerequisite dependencies of other tasks on that task are cleared, thus triggering a modification of the prerequisite dependency fields. This modification is not directly specified in the adjustment intent declaration file, but it is triggered by the direct modification of deleting the task, therefore it is a related modification.
[0115] For the second type of relationship, when a modification operation changes the execution object identifier of a task, the execution object name field will also be updated synchronously to ensure data consistency. This modification of the name field is not directly specified in the adjustment intent declaration file, but it is triggered by the modification of the execution object identifier, and therefore constitutes a related modification.
[0116] For the third type of relationship, when a modification operation postpones the end time of a subtask, the end time of the parent task set will be extended to ensure the time coverage of the parent task set. This modification of the parent task set's end time is not directly specified in the adjustment intent declaration file (because the parent task set may not be within the scope of the adjustment), but it is triggered by the modification of the subtask's end time, and therefore belongs to a related modification.
[0117] When determining whether modified data is a related modification, the execution engine checks the aforementioned relationships sequentially. If the modified data satisfies any of these relationships, its modification reason type is determined to be a related modification.
[0118] Unexpected modifications refer to data changes that are neither direct nor related modifications. Unexpected modifications can mean that the modified data cannot be explained by the adjustment intent declaration file or known relationships, and may indicate data anomalies or modification errors. Specifically, the execution engine iterates through the modified data in the post-modification orchestration plan, and for modified data that is neither a direct nor a related modification, its modification reason type is determined to be unexpected modification.
[0119] It should be noted that when the number of unexpected modifications is zero, it means that every modification in the revised orchestration plan can be reasonably explained, and the interpretability of the modifications meets the requirements. When the number of unexpected modifications is not zero, it means that there are modifications that cannot be reasonably explained, and the interpretability of the modifications does not meet the requirements. The execution engine can record the information of unexpected modifications in the modification analysis report so that the large language model can determine the credibility of the revised orchestration plan based on this information.
[0120] In this embodiment, the indication modification field is determined based on the adjustment type and adjustment scope. The modification reason type for data in the modified orchestration plan that matches the indication modification field is determined as direct modification; the modification reason type for data that is not direct modification but is related to the indication modification field is determined as associated modification; and the modification reason type for data that is neither direct nor associated modification is determined as unexpected modification. This allows for a complete and mutually exclusive attribution of each modified data point in the modified orchestration plan. Since each modification reason type corresponds to a different source, the interpretability of the modification can fully characterize the source of the modified data. When the number of unexpected modifications is zero, each modified data point in the modified orchestration plan can be reasonably explained, thus providing a basis for the large language model to complete credibility verification based solely on the modification analysis report.
[0121] In some embodiments, modifying the initial task orchestration plan based on the adjusted intent statement file includes:
[0122] Based on the adjustment intent declaration file, the modification scope of the initial task orchestration plan is matched to obtain a preview result of the modification scope;
[0123] The preview results of the modification scope are confirmed using a large language model. If the preview results of the modification scope are confirmed to be acceptable, the initial task orchestration plan is modified based on the adjustment intent declaration file.
[0124] Specifically, "modification scope matching" refers to the operation of determining the task data that needs to be modified in the initial task orchestration plan based on the adjustment scope in the adjustment intent declaration file. "Modification scope preview result" refers to the preview of the tasks matched by the modification scope before the actual modification is executed.
[0125] The execution engine reads the adjustment intent declaration file and parses the adjustment scope of each adjustment item. The adjustment scope can include an organization list and filtering conditions. The organization list can be any of all organizations or a list of specified organization identifiers. Filtering conditions can support combinations of one or more of the following dimensions: exact match of business system codes, exact match of task codes, task code prefix match, fuzzy match of task names, exact match of task identifier lists, and hierarchical match. Based on the above adjustment scope, the execution engine performs multi-dimensional filtering on the task data in the initial task orchestration plan, determines the tasks matched by each adjustment item, and generates a preview result of the modification scope. The preview result of the modification scope can include the number of tasks matched by each adjustment item and information about the top N matched tasks, such as task identifier, task name, and affiliated organization.
[0126] Confirming the preview results of the modification scope using the large language model refers to the process by which the large language model determines whether the scope of the modification aligns with the user's intent based on the preview results. Passing confirmation means the large language model determines that the modification scope matches the user's intent. Figure 1 To.
[0127] Specifically, the system receives the preview results of the modification range through a large language model, and determines whether the modification range matches the user's intention based on the user's natural language adjustment instructions and the hit task information in the preview results. Figure 1 If they match, the process is confirmed, and the large language model instructs the execution engine to modify the initial task orchestration plan based on the adjustment intent declaration file. If they do not match, the large language model can prompt the user that the scope of modification is inconsistent with the intent, and the user can decide whether to continue or adjust the adjustment intent declaration file before continuing.
[0128] It's important to note that by performing modification range matching and having the large language model confirm the modification range preview results before execution, deviations in the modification range can be detected before the actual modification is implemented, avoiding large-scale erroneous modifications due to incorrect modification ranges. Furthermore, since the modification range preview results have a relatively small data volume, the large language model only needs to use this preview result for confirmation, without needing to load the full task orchestration plan. This ensures that the context consumption of the large language model is independent of the data size.
[0129] In some embodiments, the credibility verification of the modified orchestration plan based on the modification analysis report is performed using a large language model, including:
[0130] The legality and interpretability of the modification are comprehensively analyzed by a large language model. If both the legality and interpretability of the modification meet the preset conditions, the credibility verification of the modified orchestration plan is deemed to have passed.
[0131] The preset conditions refer to the pre-defined conditions used to determine whether the legality and interpretability of the modification meet the requirements. Preset conditions for the legality of the modification may include: no blocking-level violations exist; preset conditions for the interpretability of the modification may include: the number of unexpected modifications is zero.
[0132] The large language model extracts the legality and interpretability of modifications from the modification analysis report to determine whether the legality and interpretability of the modifications meet preset conditions. If both the legality and interpretability of the modifications meet the preset conditions, the large language model determines that the credibility verification of the modified orchestration plan has passed; if either the legality or interpretability of the modifications fails to meet the preset conditions, the large language model determines that the credibility verification of the modified orchestration plan has failed.
[0133] It should be noted that the legality and interpretability of modifications characterize the reliability of the modified orchestration plan from different dimensions. Meeting the preset conditions for legality indicates that the modified orchestration plan itself does not contain illegal data, while meeting the preset conditions for interpretability indicates that every modified data point in the modified orchestration plan has a clear source. Both conditions must be met simultaneously to demonstrate that the modified orchestration plan contains neither illegal data nor unexplainable modifications, thus confirming that the reliability verification of the modified orchestration plan has passed. By comprehensively analyzing the legality and interpretability of modifications, the need to load the entire dataset into the large language model for item-by-item verification is avoided, thereby ensuring that the context consumption of the large language model is independent of the data size.
[0134] In some embodiments, after verifying the credibility of the modified orchestration plan based on the modification analysis report using a large language model, the method further includes:
[0135] If the legality of the modification does not meet the preset conditions, the illegal modification data is identified by the large language model based on the modification analysis report, and the adjustment intent declaration file is adjusted based on the illegal modification data.
[0136] Illegally modified data refers to task data in the modified scheduling plan that does not meet the legality requirements. For example, illegally modified data may include task data with duplicate task codes, task data with invalid task-assigned organizations, task data with non-existent execution objects, and task data with execution objects that have been repeatedly assigned.
[0137] If the modification's legality does not meet the preset conditions, the large language model locates the illegal modification data based on the modification analysis report. The large language model then determines the corresponding adjustment item in the adjustment intent declaration file based on the illegal modification data and adjusts that item. Adjustments may include: modifying the adjustment scope to narrow or expand the affected task subset, modifying adjustment parameters to ensure the modification result meets legality requirements, or adding new adjustment items to handle illegal modification data. The adjusted adjustment intent declaration file can serve as input for the next round of adjustments. The script program re-executes the modifications based on the adjusted adjustment intent declaration file, gradually converging to a modified orchestration plan that meets legality requirements.
[0138] For example, in a financial month-end closing scenario, a user inputs "Delay general ledger closing for all organizations by 2 hours." After the script program executes the modification, the execution target of a certain general ledger closing task in the modified orchestration plan is judged as illegally modified data because it does not exist in the original organization. The large language model locates this illegally modified data based on the modification analysis report, determines its corresponding adjustment item, and adds an adjustment item to the adjustment intent declaration file, changing the execution target of the aforementioned general ledger closing task to a valid execution target. The script program re-executes the modification based on the adjusted adjustment intent declaration file, thereby obtaining a modified orchestration plan that meets the legality requirements.
[0139] In some embodiments, after verifying the credibility of the modified orchestration plan based on the modification analysis report using a large language model, the method further includes:
[0140] If the interpretability of the modification does not meet the preset conditions, the unexpected modification data is identified by the large language model based on the modification analysis report, and the unexpected modification is verified and processed according to the modified orchestration plan.
[0141] In this context, "modification interpretability not meeting preset conditions" refers to the presence of unexplained modified data in the revised orchestration plan, i.e., unexpected modified data. If modification interpretability does not meet preset conditions, the unexpected modified data is located using the large language model based on the modification analysis report. The large language model then verifies the unexpected modified data based on the revised orchestration plan. Verification methods may include: determining the source of the unexpected modified data, judging whether the unexpected modified data belongs to normal business logic, and deciding whether to accept the unexpected modified data based on the judgment result.
[0142] It should be noted that since the failure of modification interpretability to meet the preset conditions means that there is modified data that cannot be interpreted by the adjustment intent declaration file or known relationships, the large language model needs to load the modified orchestration plan to verify the unexpectedly modified data. Unlike the large language model which loads all data for verification, this embodiment only loads the modified orchestration plan when unexpected data is found, and the purpose of loading is to verify the unexpectedly modified data, not to verify all modified data one by one. Therefore, even when the modification interpretability fails to meet the preset conditions, the context consumption of the large language model is still far lower than the scheme that loads all data.
[0143] In one implementation, the verification process may include: the large language model reading the task data involved in the unexpectedly modified data in the revised orchestration plan, and combining the adjustment intent declaration file and the modification analysis report to determine whether the unexpectedly modified data was caused by an implicit adjustment in the user's intent. For example, if the unexpectedly modified data was caused by an adjustment that was not explicitly expressed by the user in the natural language adjustment instruction but is reasonable from a business perspective, the large language model can confirm that the unexpectedly modified data is a reasonable modification and add it to the adjustment intent declaration file so that it can be reasonably explained in subsequent adjustments; if the unexpectedly modified data cannot be reasonably explained, the large language model can prompt the user to process the unexpectedly modified data or roll back the revised orchestration plan.
[0144] For example, in a financial month-end closing scenario, a user inputs "Delay all organizational general ledger closing by 2 hours." After the script executes the modification, a pre-dependency reference for a task in the modified orchestration plan is cleaned up. The large language model, based on the modification analysis report, determines that this modification to the pre-dependency reference is unexpected. The large language model reads the relevant data for that task in the modified orchestration plan and, combined with the adjustment intent declaration file, determines that the reason the pre-dependency reference was cleaned up is that its corresponding pre-task was deleted in this modification. Since this modification is a reasonable change triggered by an implicit adjustment in the user's intent, the large language model confirms that this unexpected modification is reasonable and adds its corresponding adjustment to the adjustment intent declaration file so that it can be reasonably explained in subsequent adjustments.
[0145] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.
[0146] Based on the same inventive concept, this application also provides a task orchestration apparatus for implementing the task orchestration method based on a large language model as described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations of one or more embodiments of the task orchestration apparatus based on a large language model provided below can be found in the limitations of the task orchestration method based on a large language model described above, and will not be repeated here.
[0147] In one exemplary embodiment, such as Figure 3 As shown, a task orchestration device 300 based on a large language model is provided, comprising:
[0148] The acquisition module 302 is used to acquire natural language adjustment instructions for the initial task orchestration plan of the task to be orchestrated;
[0149] The recognition module 304 is used to perform intent recognition on the natural language adjustment instructions through a large language model to obtain an adjustment intent declaration file for the initial task orchestration plan;
[0150] Modification module 306 is used to modify the initial task orchestration plan based on the adjustment intent declaration file to obtain the modified orchestration plan;
[0151] Analysis module 308 is used to analyze the modifications of the modified orchestration plan compared to the initial task orchestration plan based on the adjustment intent declaration file and the initial task orchestration plan, and to obtain a modification analysis report; the modification status includes the legality of the modification and the interpretability of the modification;
[0152] Verification module 310 is used to verify the credibility of the modified orchestration plan based on the modification analysis report using the large language model;
[0153] Return module 312 is used to return the modification analysis report and the modified orchestration plan if the credibility verification of the modified orchestration plan passes.
[0154] Each module in the aforementioned task orchestration device 300 can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.
[0155] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a task orchestration method based on a large language model. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.
[0156] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0157] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps included in any of the foregoing method embodiments.
[0158] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps included in any of the foregoing method embodiments.
[0159] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps included in any of the foregoing method embodiments.
[0160] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0161] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.
[0162] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0163] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A task orchestration method based on a large language model, characterized in that, The method includes: Obtain natural language adjustment instructions for the initial task orchestration plan of the task to be orchestrated; The intent of the natural language adjustment instructions is identified by a large language model to obtain an adjustment intent declaration file for the initial task orchestration plan; Based on the adjustment intent statement file, the initial task orchestration plan is modified to obtain the modified orchestration plan; Based on the adjustment intent statement document and the initial task orchestration plan, the modifications to the revised orchestration plan compared to the initial task orchestration plan are analyzed to obtain a modification analysis report; the modification status includes the legality and explainability of the modification; The credibility of the modified orchestration plan is verified using the large language model based on the modification analysis report. If the credibility verification of the modified orchestration plan passes, the modification analysis report and the modified orchestration plan are returned.
2. The method according to claim 1, characterized in that, The tasks to be orchestrated include multiple tasks that have task relationships; The modification of the initial task orchestration plan based on the adjustment intent declaration file to obtain the modified orchestration plan includes: An intermediate orchestration plan is obtained by modifying the initial task orchestration plan according to the adjustment intent declaration file using a preset script program. The script program identifies abnormal data in the intermediate orchestration plan based on the constraints corresponding to the task association relationship, and obtains the identification results. The script program repairs the intermediate orchestration plan based on the identification results to obtain the modified orchestration plan.
3. The method according to claim 2, characterized in that, The task association relationships include prerequisite dependencies and hierarchical inclusion relationships; the identification results include abnormal dependency data and abnormal task execution time data. The step of identifying abnormal data in the intermediate orchestration plan based on the constraints corresponding to the task association relationship through the script program, and obtaining the identification result, includes: Based on the aforementioned prerequisite dependencies, the preceding dependency references to failed tasks in the intermediate orchestration plan are identified to obtain the dependency anomaly data. Based on the hierarchical inclusion relationship, time coverage detection is performed on the execution end time of tasks with hierarchical inclusion relationships in the intermediate orchestration plan, and time coverage abnormal data is used as task execution time abnormal data. Timing conflict detection is performed on the start time of tasks with the aforementioned prerequisite dependencies in the intermediate orchestration plan, and the resulting timing conflict anomaly data is used as the task execution time anomaly data.
4. The method according to claim 3, characterized in that, The step of performing data repair on the intermediate orchestration plan based on the identification results using the script program to obtain the modified orchestration plan includes: The script program removes task data that references invalidated tasks as prerequisites in the intermediate orchestration plan based on the dependency anomaly data. Based on the time coverage anomaly data, the script program extends the task end time of the parent task set in the intermediate orchestration plan, so that the end time of the parent task set covers the maximum end time of the subtasks. The script program eliminates timing conflicts in the intermediate orchestration plan where the end time of a preceding task is greater than the start time of a subsequent task, based on the timing conflict anomaly data.
5. The method according to claim 1, characterized in that, Based on the adjustment intent statement file and the initial task orchestration plan, the modification of the revised orchestration plan compared to the initial task orchestration plan is analyzed to obtain a modification analysis report, including: The legality of the modification is determined by verifying the compliance of the modified orchestration plan's task structure and the task execution object through a preset execution engine. The execution engine determines the indicated modification field based on the adjustment intent declaration file, and determines the modification reason type of the modified data based on the matching of the modified data in the modified orchestration plan with the indicated modification field, thereby obtaining the interpretability of the modification.
6. The method according to claim 5, characterized in that, The adjustment intent declaration file includes the adjustment type and adjustment scope; the execution engine determines the indication modification field based on the adjustment intent declaration file, and determines the modification reason type of the modification data based on the matching between the modified data in the modified orchestration plan and the indication modification field, thereby obtaining the interpretability of the modification, including: The execution engine determines the instruction modification field specified in the adjustment intent declaration file based on the adjustment type and adjustment scope. The execution engine determines the modification reason type of the modified data in the modified orchestration plan that matches the indicated modification field as a direct modification. The execution engine determines the modification reason type of modification data that does not belong to the direct modification but is related to the indicated modification field as associated modification; The execution engine determines the modification reason of the modified data that does not belong to the direct modification or the associated modification as an unexpected modification.
7. The method according to claim 1, characterized in that, The modification of the initial task orchestration plan based on the adjustment intent declaration file includes: Based on the adjustment intent declaration file, the modification scope of the initial task orchestration plan is matched to obtain a preview result of the modification scope; The preview results of the modification scope are confirmed using the large language model. If the preview results of the modification scope are confirmed to be successful, the initial task orchestration plan is modified based on the adjustment intent declaration file.
8. The method according to claim 1, characterized in that, The credibility verification of the modified orchestration plan based on the modification analysis report using the large language model includes: The legality and interpretability of the modification are comprehensively analyzed using the large language model. If both the legality and interpretability of the modification meet the preset conditions, the credibility verification of the modified orchestration plan is deemed successful.
9. The method according to claim 8, characterized in that, After verifying the credibility of the modified orchestration plan based on the modification analysis report using the large language model, the method further includes: if the legality of the modification does not meet the preset conditions, determining illegal modification data using the large language model based on the modification analysis report, and adjusting the adjustment intent declaration file based on the illegal modification data.
10. The method according to claim 8, characterized in that, After verifying the credibility of the modified orchestration plan based on the modification analysis report using the large language model, the method further includes: if the interpretability of the modification does not meet the preset conditions, determining unexpected modification data using the large language model based on the modification analysis report, and verifying the unexpected modification based on the modified orchestration plan.
11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 10.
12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 10.
13. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 10.
