A task scheduling method, device and medium based on language large model

Abstract

The embodiments of this specification disclose a task scheduling method, device and medium based on a language big model, which relate to the field of artificial intelligence technology. The method includes: obtaining task information to be scheduled and a pre-set service library of the task to be scheduled, the service library including multiple services and service information of each service; generating a task description text of the task to be scheduled based on the task information to be scheduled, the task description text including task input and output information; determining a service set description text corresponding to the service library based on the service information of each service in the service library, the service set description text including multiple service descriptions, each service description including service input and output information; determining a task scheduling result of the task to be scheduled based on the task description text of the task to be scheduled and the service set description text corresponding to the service library through the language big model, the task scheduling result including multiple required services and the service scheduling order between the multiple required services.

Description

Technical Field

[0001] This specification relates to the field of artificial intelligence technology, and in particular to a task scheduling method, device, and medium based on a large language model. Background Art

[0002] As economies develop, enterprise teams gradually expand. Smaller teams often begin by performing manual administrative tasks, such as cleaning data, training machine learning models, tracking results, and deploying models to production servers. As the team and solution grow, the number of repetitive steps increases. Therefore, ensuring reliable execution of tasks becomes increasingly important.

[0003] Currently, many tools offer task orchestration services, such as MLFlow, kubeflow, and Luigi. These tools orchestrate tasks using a directed acyclic graph (DAG), greatly simplifying the task orchestration of common business processes. However, these tools require manual orchestration, rather than automated process orchestration. This means that manual orchestration requires the orchestrator to possess a certain level of professional background and a thorough understanding of the task and existing services. Limited by the orchestrator's professional background, this can lead to duplicate service development due to ignorance of existing services, resulting in low utilization of existing services. Therefore, traditional task orchestration methods require manual orchestration by the orchestrator and are limited by their professional background, resulting in underutilization of existing services and low orchestration efficiency. Summary of the Invention

[0004] One or more embodiments of this specification provide a task orchestration method, device, and medium based on a language large model to solve the following technical problems: Traditional task orchestration methods require manual orchestration by the orchestration user and are limited by the professional background of the orchestration user, resulting in the inability to fully utilize existing services and low orchestration efficiency.

[0005] One or more embodiments of this specification adopt the following technical solutions:

[0006] One or more embodiments of the present specification provide a task orchestration method based on a language big model, the method comprising: obtaining task information to be orchestrated and a pre-set service library of a task to be orchestrated, wherein the service library comprises multiple services and service information of each of the services; describing the task to be orchestrated using the task information to be orchestrated, and generating a task description text for the task to be orchestrated, wherein the task description text comprises task input and output information; describing the multiple services according to the service information of each of the services in the service library, and determining a service set description text corresponding to the service library, wherein the service set description text comprises multiple service descriptions, and each of the service descriptions comprises service input and output information; determining a task orchestration result of the task to be orchestrated based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library, through the language big model, wherein the task orchestration result comprises multiple required services and a service orchestration order between the multiple required services.

[0007] Furthermore, after determining the task orchestration result of the task to be orchestrated through a large language model based on the task description text of the task to be orchestrated and the service set description corresponding to the service library, the method also includes: generating a feasibility verification description sample of the task to be orchestrated based on the task orchestration result, the task description text of the task to be orchestrated and the service set description text; inputting the feasibility verification description sample into the large language model to determine the feasibility verification result of the task orchestration result, so as to perform feasibility verification on the task orchestration result through the feasibility verification result.

[0008] Furthermore, before determining the task scheduling result of the task to be scheduled through a large language model based on the task description text of the task to be scheduled and the service set description text corresponding to the service library, the method also includes: obtaining a small sample example task information set, wherein the small sample example task information set includes multiple small sample example tasks and sample task scheduling information of each of the small sample example tasks; based on each of the small sample example task information, describing the small sample example, determining each of the small sample example task description texts to generate a small sample example set, wherein the small sample example set includes the task description text of each of the small sample example tasks and multiple sample task scheduling description texts corresponding to each of the small sample example tasks.

[0009] Furthermore, based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library, the task orchestration result of the task to be orchestrated is determined through a large language model, specifically including: merging the task description text of the task to be orchestrated, the service set description text corresponding to the service library, and the small sample example set to generate an orchestration description sample of the task to be orchestrated; and generating the task orchestration result of the task to be orchestrated based on the task description sample and the large language model.

[0010] Furthermore, before determining the task orchestration result of the task to be orchestrated through a large language model based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library, the method further includes: splitting the service set corresponding to the service library according to the task information to be orchestrated of the task to be orchestrated and the service information of each of the services in the service library to determine a designated service subset corresponding to the task to be orchestrated; and matching the task to be orchestrated with the designated service subset based on the task information to be orchestrated of the task to be orchestrated and the designated service subset to determine whether the task to be orchestrated meets preset orchestration requirements.

[0011] Furthermore, based on the task information to be scheduled and the designated service subset of the task to be scheduled, the task to be scheduled is matched with the designated service subset to determine whether the task to be scheduled meets a preset scheduling requirement. This specifically includes: splitting the task to be scheduled in multiple ways according to the task information to be scheduled, and determining a set of split subtasks of the task to be scheduled corresponding to each split, wherein the split subtask set includes multiple scheduling subtasks and subtask input and output information of each scheduling subtask; based on the designated service information of each designated service in the designated service subset, matching and analyzing the multiple split subtask sets to determine whether there is a current designated split subtask set in the multiple split subtask sets that meets the preset requirement, wherein the preset requirement is that the subtask input and output information of each current designated split subtask in the current designated split subtask set matches the service input and output information of multiple current designated services in the designated service information; when a designated split subtask set that meets the preset requirement exists in the multiple split subtask sets, it is determined that the task to be scheduled meets the scheduling requirement.

[0012] Furthermore, based on the task information to be scheduled of the task to be scheduled and the service information of each of the services in the service library, the service set corresponding to the service library is split to determine the designated service subset corresponding to the task to be scheduled, specifically including: performing cluster analysis on the multiple services based on the service information of each of the services in the service library to obtain multiple service subsets, and determining the service type of each of the service subsets; matching the task information to be scheduled of the task to be scheduled with the service type of each of the service subsets to determine the designated service subset corresponding to the task to be scheduled.

[0013] Furthermore, after determining whether the task to be scheduled meets the preset scheduling requirements, the method also includes: when the task to be scheduled does not meet the preset scheduling requirements, matching each split subtask in the split subtask set with multiple specified services in the specified service subset to determine at least one preset split subtask, wherein the subtask input and output information of the preset split subtask does not match the service input and output information of the multiple specified services; obtaining the subtask input information and subtask output information of each of the preset split subtasks, generating service development requirement information through the subtask input information and subtask output information of each of the preset split subtasks, and sending it to the user end.

[0014] One or more embodiments of this specification provide a task scheduling device based on a large language model, including:

[0015] at least one processor; and,

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0018] Obtaining task information to be orchestrated and a pre-set service library of the task to be orchestrated, wherein the service library includes multiple services and service information of each of the services; describing the task to be orchestrated using the task information to be orchestrated, and generating a task description text for the task to be orchestrated, wherein the task description text includes task input and output information; describing the multiple services based on the service information of each of the services in the service library, and determining a service set description text corresponding to the service library, wherein the service set description text includes multiple service descriptions, and each of the service descriptions includes service input and output information; determining a task orchestration result of the task to be orchestrated based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library through a large language model, wherein the task orchestration result includes multiple required services and a service orchestration order between the multiple required services.

[0019] One or more embodiments of this specification provide a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to:

[0020] Obtaining task information to be orchestrated and a pre-set service library of the task to be orchestrated, wherein the service library includes multiple services and service information of each of the services; describing the task to be orchestrated using the task information to be orchestrated, and generating a task description text for the task to be orchestrated, wherein the task description text includes task input and output information; describing the multiple services based on the service information of each of the services in the service library, and determining a service set description text corresponding to the service library, wherein the service set description text includes multiple service descriptions, and each of the service descriptions includes service input and output information; determining a task orchestration result of the task to be orchestrated based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library through a large language model, wherein the task orchestration result includes multiple required services and a service orchestration order between the multiple required services.

[0021] At least one of the above-mentioned technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects: through the above-mentioned technical solution, through the information of the task to be orchestrated and the service information of the pre-set service library, the restrictions of the professional requirements of the orchestration user on each service are avoided, and all services within the enterprise or team are made available, which further improves the utilization rate of existing services, ensures information synchronization between different internal departments, and reduces the restrictions of the orchestration user on the professional background of the service; according to the task description text of the task to be orchestrated and the service set description text corresponding to the service library, the task orchestration result of the task to be orchestrated is generated through the language big model, which avoids the low orchestration efficiency of manual orchestration, realizes the full utilization of existing services, and effectively improves the orchestration efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some of the embodiments described in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without inventive work. In the drawings:

[0023] Figure 1 A flowchart of a task scheduling method based on a large language model provided in an embodiment of this specification;

[0024] Figure 2 A schematic diagram of an application scenario of another task scheduling method based on a large language model provided in an embodiment of this specification;

[0025] Figure 3 This is a structural diagram of a task scheduling device based on a language large model provided in an embodiment of this specification. DETAILED DESCRIPTION

[0026] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this specification without creative work should fall within the scope of protection of this specification.

[0027] The embodiments of this specification provide a task scheduling method based on a large language model. It should be noted that the execution subject in the embodiments of this specification can be a server or any device with data processing capabilities. Figure 1A flowchart of a task scheduling method based on a language large model is provided in the embodiment of this specification, such as Figure 1 As shown, it mainly includes the following steps:

[0028] Step S101: Acquire task information to be scheduled and a preset service library of the task to be scheduled.

[0029] In one embodiment of the specification, information about a task to be orchestrated is obtained. This information includes the task name, input type, and output type. The input is determined by the actual application scenario, and the output is determined by the goal. For example, the task name is pedestrian tracking, the input is a video (a series of images), and the output is a pedestrian trajectory. A pre-set service library is obtained. This service library is constantly updated. When a new service is developed within an enterprise or team, the newly developed service is updated in the service library, ensuring information synchronization between different internal departments and reducing the restrictions on the professional background of the orchestration user on the service. To effectively utilize the services, when generating the service library, service information for each service must be included, such as the service name, input type, and output type. It should be noted that the input and output of a service are determined by the inherent properties of the service when it was developed. For example, the service name is weather recognition, the input is a 2D image, and the output is the weather type. By setting the service information, the restrictions on the professional requirements of the orchestration user for each service are avoided, ensuring that all services within the enterprise or team are available, and further improving the utilization rate of existing services.

[0030] Step S102 : describing the task to be scheduled based on the task information of the task to be scheduled, and generating a task description text of the task to be scheduled.

[0031] In one embodiment of the present specification, the task to be scheduled is described through the task information to be scheduled of the task to be scheduled, and a task description text of the task to be scheduled is generated. The task description text here includes the task input and output information. It should be noted that the generation of the task description text can be in the form of a description template. The task description template is pre-set, and the task description template includes multiple task description items, each task description item corresponds to an actual description value. The task description items include task name, task input and task output. The corresponding task description values ​​are automatically filled in according to the task information to be scheduled to generate the task description text of the scheduled task. The task description text is: the task input of [task name] is [a description of the input type], the task output is [a description of the output type], and some services in the existing services are sorted to achieve [task name].

[0032] Step S103: Describe multiple services according to the service information of each service in the service library, and determine the service set description text corresponding to the service library.

[0033] In one embodiment of the present specification, multiple services are described based on the service information of each service in the service library, and the service set description text corresponding to the service library is determined, wherein the service set description includes multiple service descriptions, and each service description includes service input and output information. Similarly, the service set description text here can also be in the form of a description template. A service description template is pre-set, and the service description template includes multiple service description items. Each service description item corresponds to an actual service description value, and the service description value is generated based on the service information of each service. The service description item includes the service name, input and output, and the corresponding service description value is automatically filled in according to the service information to generate the service description text for each service. The service description texts of multiple services are combined to generate the service set description text corresponding to the service library. For example, the service set description text is: [Service Name 1] (Input: [Input Type], Output: [Output Type]); [Service Name 2] (Input: [Input Type], Output: [Output Type]); ...; [Service Name N] (Input: [Input Type], Output: [Output Type]). The service set description text of a specific case is: [Service Name: 2D Pedestrian Detection, Input: 2D Image, Output: Pedestrian Detection Box; Service Name: 2D Vehicle Detection, Input: 2D Image, Output: Vehicle Detection Box; Service Name: General Object Recognition, Input: 2D Image, Output: General Object Recognition Type; Service Name: Weather Recognition, Input: 2D Image, Output: Weather Type; Service Name: Image Matching, Input: Detection Box Set, Output: Pairing Relationship; Service Name: Trajectory Tracking, Input: Detection Box Set, Output: Trajectory].

[0034] Before determining a task orchestration result for the task to be orchestrated based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library through a language large model, the method further includes: splitting the service set corresponding to the service library according to the task information to be orchestrated of the task to be orchestrated and the service information of each service in the service library to determine a designated service subset corresponding to the task to be orchestrated; and matching the task to be orchestrated with the designated service subset based on the task information to be orchestrated of the task to be orchestrated and the designated service subset to determine whether the task to be orchestrated meets preset orchestration requirements.

[0035] In actual application scenarios, not all tasks can be orchestrated through existing services. Therefore, before performing task orchestration, it is necessary to judge the task based on the task information to be orchestrated and the service information of each service in the service library to determine whether it is suitable to use existing services for task orchestration.

[0036] In one embodiment of the present specification, the service set corresponding to the service library is split based on the task information to be scheduled and the service information of each service in the service library to determine the designated service subset corresponding to the task to be scheduled. It should be noted that due to the expansion of the enterprise scale, the number of existing services within the enterprise or team is large. In order to reduce the workload of the task scheduling judgment process, the designated service subset corresponding to the task to be scheduled is determined by splitting the service set. The number of designated service subsets here is much smaller than the number of service sets, which can greatly reduce the workload when matching services.

[0037] The service set corresponding to the service library is split according to the task information to be scheduled and the service information of each service in the service library to determine the designated service subset corresponding to the task to be scheduled. Specifically, the splitting includes: performing cluster analysis on the multiple services based on the service information of each service in the service library to obtain multiple service subsets, and determining the service type of each service subset; and matching the task information to be scheduled with the service type of each service subset to determine the designated service subset corresponding to the task to be scheduled.

[0038] In one embodiment of the present specification, a cluster analysis can be performed on multiple services using the service information of each service in the service library. The services in the service library can be classified, and services belonging to the same category can be grouped as a service subset. A service type can also be set for each service subset. When performing classification, classification can be performed based on service function. For example, the service type can be set to identification service or detection service. It can also be classified according to the service subject, such as person classification or animal classification. This is not specifically limited here, and the setting can be selected according to actual needs. Based on the task information to be scheduled and the service type of each service subset, the task to be scheduled is matched with the service type of each service subset to determine the designated service subset corresponding to the task to be scheduled. For example, if the task type of the task to be scheduled is a pedestrian recognition task, the service subset corresponding to the identification service can be determined as the designated service subset in the identification service and detection service. By analyzing the task to be scheduled, the service subset corresponding to the task to be scheduled is matched in the service set of the service library, reducing the number of subsequent operations and further improving the scheduling efficiency.

[0039] Based on the to-be-orchestrated task information of the to-be-orchestrated task and the designated service subset, matching the to-be-orchestrated task with the designated service subset to determine whether the to-be-orchestrated task meets a preset orchestration requirement, specifically comprising: performing multiple splits on the to-be-orchestrated task according to the to-be-orchestrated task information, and determining a split subtask set of the to-be-orchestrated task corresponding to each split, wherein the split subtask set includes multiple orchestration subtasks and subtask input / output information of each orchestration subtask; performing matching analysis with multiple split subtask sets based on the designated service information of each designated service in the designated service subset, and determining whether there is a current designated split subtask set in the multiple split subtask sets that meets the preset requirement, wherein the preset requirement is that the subtask input / output information of each current designated split subtask in the current designated split subtask set matches the service input / output information of multiple current designated services in the designated service information; and determining that the to-be-orchestrated task meets the orchestration requirement when a designated split subtask set that meets the preset requirement exists in the multiple split subtask sets.

[0040] In one embodiment of the present specification, the task to be scheduled is split into multiple parts according to the task information to be scheduled, and the subtask set of the task to be scheduled corresponding to each split is determined. There are multiple ways to split the same task. For example, when the task is a pedestrian tracking task, its corresponding subtasks include a pedestrian detection subtask and a trajectory tracking subtask. Its corresponding subtasks may also include a pedestrian recognition subtask, a pedestrian positioning subtask, and a position trajectory subtask. The task splitting here can be achieved through a model. Each subtask set includes multiple scheduling subtasks and subtask input and output information of each of the scheduling subtasks.

[0041] Assume that the task to be scheduled Where I and O represent task input and task output respectively; the current service set is the set Task splitting This means completing the task Split into a series of subtasks in Based on the specified service information of each specified service in the specified service subset, a matching analysis is performed with multiple split subtask sets to determine whether there is a current specified split subtask set that meets the preset requirements in the multiple split subtask sets. Make Each subtask in Satisfy I t =I j ,O t =O j , then it is called right It is choreographic. That is, the preset requirement here is that the subtask input and output information of each currently specified split subtask in the currently specified split subtask set matches the service input and output information of multiple currently specified services in the specified service information. The match here means that the subtask input information is the same as the service input information of the currently specified service, and the subtask output information is the same as the service output information of the currently specified service. When there is a specified split subtask set that meets the preset requirements among the multiple split subtask sets, it is determined that the task to be choreographed meets the choreography requirements. If the task to be choreographed meets the choreography requirements, step S104 is executed.

[0042] After determining whether the task to be scheduled meets the preset scheduling requirements, the method further includes: when the task to be scheduled does not meet the preset scheduling requirements, matching each split subtask in the split subtask set with multiple specified services in the specified service subset to determine at least one preset split subtask, wherein the subtask input and output information of the preset split subtask does not match the service input and output information of the multiple specified services; obtaining the subtask input information and subtask output information of each of the preset split subtasks, generating service development requirement information based on the subtask input information and subtask output information of each of the preset split subtasks, and sending the service development requirement information to the user end.

[0043] In one embodiment of the present specification, when the task to be scheduled does not meet the preset scheduling requirements, it means that there is a situation in which the subtask cannot be implemented in the service set corresponding to the existing service library. In this case, in order to ensure the smooth implementation of the scheduling task, corresponding developers are required to develop such services. Match the split subtasks in each split subtask set with multiple specified services in the specified service subset to determine at least one preset split subtask, wherein the subtask input and output information of the preset split subtask does not match the service input and output information of the multiple specified services. Obtain the subtask input information and subtask output information of each of the preset split subtasks, generate service development requirement information through the subtask input information and subtask output information of each of the preset split subtasks, and send it to the user end.

[0044] Service development demand information can provide accurate service development requirements while ensuring the utilization rate of current services, provide developers with targeted development suggestions, save development time, and further ensure the efficiency of subsequent task orchestration.

[0045] Step S104 : Based on the task description text of the task to be scheduled and the service set description text corresponding to the service library, a task scheduling result of the task to be scheduled is determined through a language big model.

[0046] The task scheduling result includes multiple required services and a service scheduling sequence between the multiple required services.

[0047] Based on the task description text of the task to be scheduled and the service set description text corresponding to the service library, before determining the task scheduling result of the task to be scheduled through the language large model, the method also includes: obtaining a small sample example task information set, wherein the small sample example task information set includes multiple small sample example tasks and sample task scheduling information of each of the small sample example tasks; based on each of the small sample example task information, describing the small sample example, determining each of the small sample example task description texts, to generate a small sample sample set, wherein the small sample sample set includes the task description text of each of the small sample example tasks and multiple sample task scheduling description texts corresponding to each of the small sample example tasks.

[0048] In one embodiment of the present specification, in order to ensure the accuracy of the model running results of the large language model, it is also possible to provide a small sample example task. A small sample example task information set is obtained, and the small sample example task information set includes multiple small sample example tasks and sample task scheduling information of each small sample example task. Based on each small sample example task information, the small sample example is described, and the description text of each small sample example task is determined to generate a small sample example set, wherein the small sample example set includes the task description text of each small sample example task and multiple sample task scheduling description texts corresponding to each small sample example task. For example, the input of [task name 1] is [description of input type], and the output is [description of output type]: through [service name a], [output type-a] is obtained; through [service name b], [output type-b] is obtained; ...; through [service name n], [output type-1] is obtained. The input of [Task Name 2] is [Description of Input Type], and the output is [Description of Output Type]: through [Service Name A], obtain [Output Type-A]; through [Service Name B], obtain [Output Type-B];...; through [Service Name N], obtain [Output Type-2].

[0049] Based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library, the task orchestration result of the task to be orchestrated is determined through the language big model, specifically including: merging the task description text of the task to be orchestrated, the service set description text corresponding to the service library and the sample task orchestration description text to generate an orchestration description sample of the task to be orchestrated; and generating the task orchestration result of the task to be orchestrated based on the task description sample and the language big model.

[0050] In one embodiment of this specification, Figure 2This is a schematic diagram of an application scenario of another task scheduling method based on a language large model provided in an embodiment of this specification, such as Figure 2 As shown, the task description text, the service set description text corresponding to the service library and the sample task arrangement description text corresponding to the small sample set can be merged to generate the arrangement description sample of the task to be arranged, that is, the corresponding Figure 2 The prompt in the ,inputs the sample orchestration description into the large language model, and outputs the task orchestration result for the task to be orchestrated. It should be noted that if there are no sample examples, the task description text can be directly merged with the service set description text corresponding to the service library, and then input into the large language model to output the task orchestration result. If a small sample set is available, the sample task orchestration description text corresponding to the small sample set is added to help ensure the accuracy of the task orchestration result.

[0051] After determining the task scheduling result of the task to be scheduled based on the task description text of the task to be scheduled and the service set description corresponding to the service library through the language macro model, the method further includes: generating a feasibility verification description sample of the task to be scheduled based on the task scheduling result, the task description text of the task to be scheduled and the service set description; inputting the feasibility verification description sample into the language macro model to determine the feasibility verification result of the task scheduling result, so as to perform feasibility verification on the task scheduling result through the feasibility verification result.

[0052] In one embodiment of the present specification, it is necessary to check whether the orchestration result is valid, that is, it is necessary to perform a feasibility test on the result to be orchestrated. Based on the task orchestration result, the task description text of the task to be orchestrated and the service set description text, a feasibility verification description sample of the task to be orchestrated is generated. Service set description text: 1. [Service name 1] (input: [service input type], output: [service output type]); 2. [Service name 2] (input: [service input type], output: [service output type]); ...; N. [Service name N] (input: [service input type], output: [service output type]). The task description text of the task to be orchestrated is: Input: [task input type]; Output: [task output type]. The corresponding arrangement result description text of the task arrangement result is: Input: [subtask input 1]; Output: [subtask output 1]; Input: [Subtask Input 2]; Output: [Subtask Output 2]... Input: [subtask input T]; Output: [subtask output T]. is the service name of the first service to be called, is the service name of the second called service, and so on.

[0053] When generating a feasibility verification description sample, the content can be filled in in the form of a feasibility verification description template. For example, the feasibility verification description template here can be: the service set description text is [service set description text], the task description text of the orchestration task is [task description text], and through the task orchestration result [orchestration result description text], can [task name] be achieved? The implementation judgment criteria are as follows: the input of any service is the output of the previous service, the service output of the last service is [task output type], and each service exists in the service set. Input the feasibility verification description sample into the language large model to determine the feasibility verification result of the task orchestration result, so as to verify the feasibility of the task orchestration result through the feasibility verification result. By setting up feasibility verification, the accuracy of the task orchestration result is guaranteed.

[0054] In one embodiment of the present specification, step S104 is illustrated by the following case, and the service set description text is: [Service name: 2D pedestrian detection, input: 2D image, output: pedestrian detection frame; Service name: 2D vehicle detection, input: 2D image, output: vehicle detection frame; Service name: General object recognition, input: 2D image, output: general object recognition type; Service name: Weather recognition, input: 2D image, output: weather type; Service name: Image matching, input: detection frame set, output: pairing relationship; Service name: Trajectory tracking, input: detection frame set, output: trajectory], and the task description text is: pedestrian tracking task, input: video (a series of images), output: pedestrian trajectory. Some services in the existing services are sorted, and the required service names are output one by one to realize the pedestrian tracking task. Input the corresponding arrangement description sample into the language model, which outputs the following: To implement the pedestrian tracking task, existing services need to be called in the following order: 2D pedestrian detection - input: 2D image - output: pedestrian detection box; trajectory tracking - input: detection box set - output: trajectory. First, use the 2D pedestrian detection service to detect pedestrians in each frame of the video and obtain pedestrian detection boxes. Then, these detection boxes are input into the trajectory tracking service, which tracks the pedestrian's trajectory based on the detection box's position and time information. Finally, the pedestrian's trajectory information is output.

[0055] Verify the feasibility of the task orchestration generated above by inputting [service set description text] and [pedestrian tracking task, input: video (series of images), output: pedestrian trajectory] into the language model. Can the pedestrian tracking task be achieved using the following service sequence? The achievable standard is that the input of a service must be the output of the previous service, the final output is a pedestrian trajectory, and each service must exist in the service set: [Service Name: 2D Pedestrian Detection, Input: 2D Image, Output: Pedestrian Detection Box; Service Name: Trajectory Tracking, Input: Detection Box Set, Output: Pedestrian Trajectory].

[0056] The verification output result is as follows: Yes, the pedestrian tracking task can be achieved through the following service sequence: 2D pedestrian detection - input: 2D image - output: pedestrian detection frame; trajectory tracking - input: detection frame set - output: trajectory; first, use the 2D pedestrian detection service to detect pedestrians in each frame in the video and obtain pedestrian detection frames; then, pass these detection frames as input to the trajectory tracking service, and track the pedestrian's trajectory based on the position and time information of the detection frame; finally output the pedestrian's trajectory information; the service sequence meets the following requirements: the input of 2D pedestrian detection is a 2D image, which meets the input requirements of the pedestrian tracking task and is the output of the previous service (video converted to 2D image); the input of trajectory tracking is a detection frame set, which can be obtained through the 2D pedestrian detection service; the final output is the pedestrian trajectory, which meets the requirements of the pedestrian tracking task, and each service is in the given service set and 2D pedestrian detection and trajectory tracking are both services in the set, which meet the conditions; therefore, through this service sequence, the pedestrian tracking task can be successfully achieved.

[0057] When a task is choreographic, it can output a valid choreography result and pass the feasibility test. Conversely, when a task is not choreographic, although the choreography result is output, it cannot pass the feasibility test. For example, the service set description text is: [Service Name: 2D Pedestrian Detection, Input: 2D Image, Output: Pedestrian Detection Box; Service Name: 2D Vehicle Detection, Input: 2D Image, Output: Vehicle Detection Box; Service Name: General Object Recognition, Input: 2D Image, Output: General Object Recognition Type; Service Name: Weather Recognition, Input: 2D Image, Output: Weather Type; Service Name: Image Matching, Input: Detection Box Set, Output: Pairing Relationship; Service Name: Trajectory Tracking, Input: Detection Box Set, Output: Trajectory], and the task description text is: Aircraft Tracking Task, Input: Video (a series of images), Output: Aircraft Trajectory. Sort some of the existing services and output the required service names one by one to implement the aircraft tracking task. Inputting the corresponding arrangement description sample into the language model will output the following: To implement the aircraft tracking task, existing services need to be called in the following order: Service Name: Weather Recognition - Input: 2D Image - Output: Weather Type; Service Name: 2D Vehicle Detection - Input: 2D Image - Output: Vehicle Detection Box; Service Name: Image Matching - Input: Detection Box Set - Output: Pairing Relationship; Service Name: Trajectory Tracking - Input: Detection Box Set - Output: Trajectory. Based on the requirements, the services can be orchestrated as follows to track the trajectory of an aircraft. First, the Weather Recognition service is used to determine the weather type of the image in the video, which is important for aircraft detection and tracking. After weather type identification, the 2D Vehicle Detection service is used to detect the aircraft in the video. Next, the Image Matching service is used to pair the aircraft detection boxes in different images. Finally, the Trajectory Tracking service is used to track the aircraft's trajectory. The outputs of these services can be used for further tasks such as target tracking and behavior analysis.

[0058] Verify the feasibility of the task orchestration generated above by inputting [service set description text] and [aircraft tracking task, input: video (series of images), output: aircraft trajectory] into the language model. Can the aircraft tracking task be achieved using the following service sequence? The answer is "yes" or "no." The achievable criteria are that the input of a service must be the output of the previous service, the final output is a pedestrian trajectory, and each service must exist in the service set, [Service Name: Weather Recognition - Input: 2D Image - Output: Weather Type; Service Name: 2D Vehicle Detection - Input: 2D Image - Output: Vehicle Detection Box; Service Name: Image Matching - Input: Detection Box Set - Output: Pairing Relationship; Service Name: Trajectory Tracking - Input: Detection Box Set - Output: Trajectory].

[0059] The verification output is as follows: No. First, the final output should be an aircraft trajectory, not a pedestrian trajectory. Second, the vehicle detection bounding boxes output by the 2D Vehicle Detection service are unrelated to the aircraft, so the aircraft trajectory cannot be obtained using the Image Matching and Trajectory Tracking services. As shown in the above example, when a task is not choreographic, although choreography results are output, it will fail the feasibility check.

[0060] In order to verify the effectiveness of the present invention, 50 tasks (40 of which are schedulable and 10 are not schedulable) were generated by the big model and verified. The experimental results show that the algorithm correctly choreographed 36 of the schedulable tasks with an accuracy rate of 90%; in the pre-verification stage, that is, the stage of identifying the schedulability of tasks before choreography, the big model correctly identified eight of the ten non-schedulable tasks, that is, accurately identified 80% of the non-schedulable tasks; in the post-verification stage, that is, the feasibility verification stage after the completion of the choreography, the big model accurately identified the services in the choreography results that did not exist in the service set. Overall, the choreography success rate of the big model is 92%. It should be noted that the successful choreography here includes correctly choreographing tasks, identifying non-schedulable tasks, and identifying incorrect services in the incorrect choreography.

[0061] Through the above technical solution, through the information of the task to be orchestrated and the service information of the pre-set service library, the restrictions of the professional requirements of the orchestration user on each service are avoided, and all services within the enterprise or team are made available, which further improves the utilization rate of existing services, ensures the synchronization of information between different internal departments, and reduces the restrictions of the orchestration user on the professional background of the service; according to the task description text of the task to be orchestrated and the service set description text corresponding to the service library, the task orchestration result of the task to be orchestrated is generated through the language big model, which avoids the low orchestration efficiency of manual orchestration, realizes the full utilization of existing services, and effectively improves the orchestration efficiency.

[0062] The embodiment of this specification also provides a task scheduling device based on a language large model, such as Figure 3 As shown, the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to:

[0063] Obtaining task information of a task to be orchestrated and a pre-set service library, wherein the service library includes multiple services and service information of each service; describing the task to be orchestrated using the task information of the task to be orchestrated, and generating a task description text of the task to be orchestrated, wherein the task description text includes task input and output information; describing the multiple services based on the service information of each service in the service library, and determining a service set description text corresponding to the service library, wherein the service set description text includes multiple service descriptions, each of which includes service input and output information; determining a task orchestration result of the task to be orchestrated based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library through a large language model, wherein the task orchestration result includes multiple required services and a service orchestration order between the multiple required services.

[0064] The embodiments of this specification also provide a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows:

[0065] Obtaining task information of a task to be orchestrated and a pre-set service library, wherein the service library includes multiple services and service information of each service; describing the task to be orchestrated using the task information of the task to be orchestrated, and generating a task description text of the task to be orchestrated, wherein the task description text includes task input and output information; describing the multiple services based on the service information of each service in the service library, and determining a service set description text corresponding to the service library, wherein the service set description text includes multiple service descriptions, each of which includes service input and output information; determining a task orchestration result of the task to be orchestrated based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library through a large language model, wherein the task orchestration result includes multiple required services and a service orchestration order between the multiple required services.

[0066] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant details, refer to the descriptions of the method embodiments.

[0067] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0068] The devices and media provided in the embodiments of this specification correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0069] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0070] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0071] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0072] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0073] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0074] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0075] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0076] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0077] The foregoing description is merely one or more embodiments of this specification and is not intended to limit this specification. It will be apparent to those skilled in the art that various modifications and variations may be made to one or more embodiments of this specification. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of one or more embodiments of this specification are intended to be within the scope of the claims of this specification.

Claims

1. A task scheduling method based on a large language model, characterized in that: The method comprises: Acquire task information to be scheduled and a pre-set service library of the task to be scheduled, wherein the service library includes multiple services and service information of each of the services; Describing the task to be scheduled using the task information to be scheduled, and generating a task description text for the task to be scheduled, wherein the task description text includes task input and output information; Describing the multiple services according to the service information of each service in the service library, and determining a service set description text corresponding to the service library, wherein the service set description text includes multiple service descriptions, and each service description includes service input and output information; Determining a task orchestration result of the task to be orchestrated based on the task description text of the task to be orchestrated and the service set description text corresponding to the service library through a language large model, wherein the task orchestration result includes multiple required services and a service orchestration order between the multiple required services; After determining the task scheduling result of the task to be scheduled based on the task description text of the task to be scheduled and the service set description corresponding to the service library through the language big model, the method further includes: Generate a feasibility verification description sample of the task to be scheduled according to the task scheduling result, the task description text of the task to be scheduled, and the service set description text; Inputting the feasibility verification description sample into the language macro model, determining a feasibility verification result of the task scheduling result, and performing feasibility verification on the task scheduling result through the feasibility verification result; Before determining a task scheduling result of the task to be scheduled based on the task description text of the task to be scheduled and the service set description text corresponding to the service library through a language large model, the method further includes: Splitting the service set corresponding to the service library according to the task information to be scheduled and the service information of each service in the service library to determine a designated service subset corresponding to the task to be scheduled; Based on the task information to be scheduled and the designated service subset, the task to be scheduled is matched with the designated service subset to determine whether the task to be scheduled meets a preset scheduling requirement; Matching the task to be scheduled with the designated service subset based on the task information to be scheduled and the designated service subset to determine whether the task to be scheduled meets a preset scheduling requirement, specifically including: Splitting the task to be scheduled in multiple ways according to the task information to be scheduled, and determining a set of split subtasks of the task to be scheduled corresponding to each split, wherein the set of split subtasks includes multiple scheduling subtasks and subtask input and output information of each of the scheduling subtasks; Based on the designated service information of each designated service in the designated service subset, a matching analysis is performed with the multiple split subtask sets to determine whether there is a current designated split subtask set in the multiple split subtask sets that meets a preset requirement, wherein the preset requirement is that the subtask input and output information of each current designated split subtask in the current designated split subtask set matches the service input and output information of the multiple current designated services in the designated service information; When there is a designated split subtask set that meets the preset requirements among the multiple split subtask sets, it is determined that the task to be scheduled meets the scheduling requirements.

2. The task scheduling method based on a language large model according to claim 1 is characterized in that: Before determining a task scheduling result of the task to be scheduled based on the task description text of the task to be scheduled and the service set description text corresponding to the service library through a language large model, the method further includes: Acquire a small sample example task information set, wherein the small sample example task information set includes a plurality of small sample example tasks and example task arrangement information of each of the small sample example tasks; Based on the task information of each small sample example, the small sample example is described, and the description text of each small sample example task is determined to generate a small sample example set, wherein the small sample example set includes the task description text of each small sample example task and multiple sample task arrangement description texts corresponding to each small sample example task.

3. The task scheduling method based on a language large model according to claim 2 is characterized in that: Based on the task description text of the task to be scheduled and the service set description text corresponding to the service library, determining the task scheduling result of the task to be scheduled through the language big model, specifically including: Merging the task description text of the task to be scheduled, the service set description text corresponding to the service library, and the small sample example set to generate a scheduling description sample of the task to be scheduled; Based on the task description sample and the language large model, a task scheduling result of the task to be scheduled is generated.

4. The task scheduling method based on a language large model according to claim 1 is characterized in that: Splitting the service set corresponding to the service library according to the task information to be scheduled and the service information of each service in the service library to determine a designated service subset corresponding to the task to be scheduled, specifically including: Performing cluster analysis on the multiple services using the service information of each of the services in the service library to obtain multiple service subsets, and determining a service type for each of the service subsets; The task information to be scheduled of the task to be scheduled is matched with the service type of each of the service subsets to determine a designated service subset corresponding to the task to be scheduled.

5. The task scheduling method based on a language large model according to claim 1 is characterized in that: After determining whether the task to be scheduled meets the preset scheduling requirements, the method further includes: When the task to be scheduled does not meet the preset scheduling requirements, matching each split subtask in the split subtask set with a plurality of specified services in the specified service subset to determine at least one preset split subtask, wherein subtask input and output information of the preset split subtask does not match service input and output information of the plurality of specified services; The subtask input information and subtask output information of each of the preset split subtasks are obtained, and service development requirement information is generated through the subtask input information and subtask output information of each of the preset split subtasks, and sent to the user end.

6. A task scheduling device based on a large language model, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5.

7. A non-volatile computer storage medium storing computer-executable instructions, characterized in that: The computer executable instructions are configured to execute the method according to any one of claims 1 to 5.

Images