Online reasoning method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an online reasoning method and device, electronic equipment and a storage medium, and is used for improving the matching degree of an online reasoning task and a model. The method comprises the following steps: on the basis of task content of an online reasoning task, dividing the online reasoning task to obtain a plurality of sub-tasks; based on the respective weight of each task attribute feature of the subtasks, fusing each task attribute feature to obtain a comprehensive attribute feature; determining a target model matched with the sub-task from the candidate models based on the comprehensive attribute characteristics and the model attribute characteristics of the candidate models, and distributing the sub-task to the target model for processing; and obtaining a processing result of the online reasoning task based on the processing results of the respective target models of the plurality of sub-tasks. According to the method, the online reasoning task is divided, and the matched target model is determined based on the task attribute of each sub-task, so that the task processing efficiency and the resource utilization rate can be improved.

Description

Background Art

[0002] With the rapid development of artificial intelligence technology, online reasoning has become one of the key applications in the field of machine learning and deep learning. Online reasoning refers to the ability of the model to process and output results immediately after receiving online reasoning tasks from the business side. This is crucial for applications that require fast response, such as real-time monitoring, autonomous vehicles, and intelligent customer service.

[0003] At present, when an online reasoning system receives an online reasoning task, it usually assigns the online reasoning task to each model for processing in order. However, due to the different parameter scales of different models, the time and accuracy of processing tasks are also different. If an online reasoning task with high real-time requirements is assigned to a model with a long processing time, the real-time requirements of the task cannot be met. If an online reasoning task with a small amount of data is assigned to a model with a large parameter scale, the computing resources of the model cannot be fully utilized.

[0004] Therefore, how to improve the matching degree between online reasoning tasks and models has become a problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide an online reasoning method, device, electronic device and storage medium to improve the matching degree between online reasoning tasks and models.

[0006] An online reasoning method provided in an embodiment of the present application includes:

[0007] Based on the task content of the online reasoning task, the online reasoning task is divided to obtain multiple subtasks;

[0008] For each subtask, based on the weights of the respective task attribute features of the subtask, the attribute features of the respective tasks are integrated to obtain a comprehensive attribute feature;

[0009] Based on the comprehensive attribute features and the model attribute features of each candidate model, determine a target model matching the subtask from the candidate models, and assign the subtask to the target model for processing;

[0010] Based on the processing results of the target models of the multiple subtasks, the processing result of the online reasoning task is obtained.

[0011] Optionally, the attribute characteristics of each task are obtained in the following manner:

[0012] Performing feature extraction on the subtask based on each preset feature extraction rule to obtain a feature vector of the subtask under each feature extraction rule;

[0013] The obtained feature vectors are mapped respectively to obtain the attribute features of each task.

[0014] Optionally, the task attribute feature includes a task feature value; and the mapping of each obtained feature vector to obtain each task attribute feature includes:

[0015] Mapping each of the characteristic vectors to a preset numerical interval respectively to obtain each task characteristic value, wherein each task characteristic value belongs to the preset numerical interval;

[0016] The step of fusing the task attribute features based on the weights of the task attribute features of the subtasks to obtain a comprehensive attribute feature includes:

[0017] Based on the weights of the respective task attribute features, the task feature values ​​are weighted and summed to obtain the comprehensive attribute feature.

[0018] Optionally, allocating the subtask to the target model for processing includes:

[0019] Determining a resource allocation strategy for the target model based on the current resource usage status and the resource requirements of the target model;

[0020] The cloud native platform is scheduled to execute the resource allocation strategy, and use the allocated computing resources to run the target model to process the subtask.

[0021] Optionally, the weight of each task attribute feature is obtained based on the feature category to which the task attribute feature belongs; the method further includes:

[0022] Acquiring processing status information of the target model in the process of processing the subtask;

[0023] Based on the processing status information, the weights corresponding to the respective feature categories are adjusted.

[0024] Optionally, the comprehensive attribute feature includes a comprehensive feature value, and the model attribute feature includes a model feature value; and determining a target model matching the subtask from each candidate model based on the comprehensive attribute feature and the model attribute features of each candidate model includes:

[0025] Sort the eigenvalues ​​of each model in order of size to obtain the sorted eigenvalues ​​of each model;

[0026] Based on each model eigenvalue except the last model eigenvalue, a plurality of eigenvalue groups are constructed, each eigenvalue group comprising: a model eigenvalue and a model eigenvalue subsequent to the model eigenvalue;

[0027] Determining a target feature value group from the plurality of feature value groups based on the comprehensive feature value, wherein the comprehensive feature value is not greater than a larger model feature value in the target feature value group and not less than a smaller model feature value in the target feature value group;

[0028] The candidate model corresponding to the larger model feature value is used as the target model matching the subtask.

[0029] An online reasoning device provided in an embodiment of the present application includes:

[0030] A division unit, used for dividing the online reasoning task based on the task content of the online reasoning task to obtain a plurality of subtasks;

[0031] A fusion unit, for fusing the attribute features of each subtask based on the weights of the attribute features of each subtask to obtain a comprehensive attribute feature;

[0032] a determination unit, configured to determine a target model matching the subtask from the candidate models based on the comprehensive attribute feature and the model attribute features of each candidate model, and assign the subtask to the target model for processing;

[0033] A summarizing unit is used to obtain the processing result of the online reasoning task based on the processing results of the target models of the multiple subtasks.

[0034] Optionally, the fusion unit is specifically used to obtain the attribute features of each task in the following manner:

[0035] Performing feature extraction on the subtask based on each preset feature extraction rule to obtain a feature vector of the subtask under each feature extraction rule;

[0036] The obtained feature vectors are mapped respectively to obtain the attribute features of each task.

[0037] Optionally, the task attribute feature includes a task feature value; and the fusion unit is specifically used for:

[0038] Mapping each of the characteristic vectors to a preset numerical interval respectively to obtain each task characteristic value, wherein each task characteristic value belongs to the preset numerical interval;

[0039] The step of fusing the task attribute features based on the weights of the task attribute features of the subtasks to obtain a comprehensive attribute feature includes:

[0040] Based on the weights of the respective task attribute features, the task feature values ​​are weighted and summed to obtain the comprehensive attribute feature.

[0041] Optionally, the determining unit is specifically configured to:

[0042] Determining a resource allocation strategy for the target model based on the current resource usage status and the resource requirements of the target model;

[0043] The cloud native platform is scheduled to execute the resource allocation strategy, and use the allocated computing resources to run the target model to process the subtask.

[0044] Optionally, the weight of each task attribute feature is obtained based on the feature category to which the task attribute feature belongs; the device further includes an adjustment unit, configured to:

[0045] Acquiring processing status information of the target model in the process of processing the subtask;

[0046] Based on the processing status information, the weights corresponding to the respective feature categories are adjusted.

[0047] Optionally, the comprehensive attribute feature includes a comprehensive feature value, and the model attribute feature includes a model feature value; and the determining unit is specifically configured to:

[0048] Sort the eigenvalues ​​of each model in order of size to obtain the sorted eigenvalues ​​of each model;

[0049] Based on each model eigenvalue except the last model eigenvalue, a plurality of eigenvalue groups are constructed, each eigenvalue group comprising: a model eigenvalue and a model eigenvalue subsequent to the model eigenvalue;

[0050] Determining a target feature value group from the plurality of feature value groups based on the comprehensive feature value, wherein the comprehensive feature value is not greater than a larger model feature value in the target feature value group and is not less than a smaller model feature value in the target feature value group;

[0051] The candidate model corresponding to the larger model feature value is used as the target model matching the subtask.

[0052] An electronic device provided in an embodiment of the present application includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the above-mentioned online reasoning methods.

[0053] An embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of any one of the above-mentioned online reasoning methods.

[0054] An embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device performs the steps of any one of the above-mentioned online reasoning methods.

[0055] The beneficial effects of this application are as follows:

[0056] The embodiment of the present application provides an online reasoning method, device, electronic device and storage medium, which divides the online reasoning task to obtain multiple subtasks; based on the task attribute characteristics of each subtask and the model attribute characteristics of each candidate model, the target model matching the subtask is determined, and the subtask is assigned to the target model for processing to obtain the processing result of the online reasoning task. Based on the above method, by dividing the online reasoning task, the online reasoning task can be assigned to multiple models for processing, improving the task processing efficiency, and matching the attributes of each subtask with the attributes of the model to ensure that the target model meets the task requirements of the subtask, and fully utilizes the model capabilities of the target model to improve the utilization of computing resources, and thus in the face of high load and tight computing resources, it can still maintain high-performance reasoning capabilities, improve the stability and reliability of the online reasoning system, and reduce operation and maintenance costs.

[0057] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0059] Figure 1 An optional schematic diagram of an application scenario in an embodiment of the present application;

[0060] Figure 2 is an implementation flow chart of an online reasoning method in an embodiment of the present application;

[0061] Figure 3 A schematic diagram of the structure of an online reasoning system in an embodiment of the present application;

[0062] Figure 4 It is a schematic diagram of the overall process of an online reasoning method in an embodiment of the present application;

[0063] Figure 5 A schematic diagram of the structure of an online reasoning device in an embodiment of the present application;

[0064] Figure 6 A schematic diagram of a hardware structure of an electronic device to which an embodiment of the present application is applied;

[0065] Figure 7 The present invention is a schematic diagram of the hardware structure of another electronic device to which the embodiments of the present application are applied. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the technical solution of the present application, rather than all of the embodiments. Based on the embodiments recorded in the application documents, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the technical solution of the present application.

[0067] The following is an introduction to some concepts involved in the embodiments of the present application.

[0068] Cloud Native: It is a method of building and running applications, which takes into account the characteristics of the cloud environment from the beginning of the application design, such as elasticity, scalability, distributed deployment, etc.

[0069] Online Inference: refers to the ability of a machine learning model to process and output results immediately after receiving real-time data input. It is often used in application scenarios that require fast response.

[0070] Model Scheduling: refers to the process of dynamically allocating and adjusting model operations based on task requirements and resource conditions.

[0071] Resource Allocation: refers to the process of reasonably allocating computing resources (such as CPU, memory, GPU) to different tasks or models in the system.

[0072] Load Balancing: is a method to evenly distribute workloads to multiple computing resources to optimize resource usage and improve system throughput and availability.

[0073] Microservices Architecture: is a method of developing an application as a set of small services, each of which runs in its own independent process and collaborates with each other through lightweight communication mechanisms.

[0074] Containerization: It is a technology that packages software applications and their dependencies in containers. The containers are isolated from the underlying infrastructure to ensure the consistency of applications in different environments.

[0075] Continuous Integration / Continuous Deployment (CI / CD): is a software development practice that improves the speed and quality of software delivery by automating code building, testing, and deployment processes.

[0076] Service Mesh: A dedicated infrastructure layer that handles communication between services and provides network functions such as service discovery, load balancing, and fault recovery.

[0077] The following is an introduction to the design concept of this application:

[0078] With the rapid development of artificial intelligence technology, online reasoning has become one of the key applications in the field of machine learning and deep learning. Online reasoning refers to the ability of a model to process and output results immediately after receiving real-time data input. This is crucial for applications that require fast response, such as real-time monitoring, autonomous vehicles, and intelligent customer service.

[0079] Currently, online reasoning systems usually adopt the following architecture:

[0080] Centralized Server: The model is deployed on one or more high-performance servers, which are responsible for processing all incoming data requests.

[0081] Model management: A central management system is responsible for the deployment, updating and monitoring of models.

[0082] Data stream processing: Data is processed through a specific data stream processing framework to ensure the real-time and accuracy of the data.

[0083] When receiving online reasoning tasks, existing online reasoning systems usually assign online reasoning tasks to various models for processing in sequence. However, due to the different parameter scales of different models, the time and accuracy of task processing are also different. If online reasoning tasks with high real-time requirements are assigned to models with long processing time, the real-time requirements of the tasks cannot be met. If online reasoning tasks with small data volumes are assigned to models with large parameter scales, the computing resources of the models cannot be fully utilized.

[0084] And the existing online reasoning systems still face the following challenges and limitations:

[0085] Scalability: As the number of users and data volumes increase, centralized servers may have difficulty scaling to meet the growing demand.

[0086] Flexibility: Existing systems are often difficult to adapt to different operating environments and platforms, limiting their deployment in multi-cloud or hybrid cloud environments.

[0087] Resource utilization: Fixed server resources may result in wasted or insufficient resources when demand fluctuates.

[0088] Updates and maintenance: In a centralized architecture, model updates and maintenance may require downtime, affecting service continuity.

[0089] Cost: Maintaining high-performance servers and data centers is expensive, especially when resources are not fully utilized.

[0090] Fault tolerance: A single point failure may cause service interruption of the entire system, affecting the stability and reliability of the system.

[0091] To overcome these challenges, a more flexible, scalable, and cost-effective online inference approach is needed.

[0092] In view of this, the embodiments of the present application provide an online reasoning method, device, electronic device and storage medium, which divide the online reasoning task to obtain multiple subtasks; based on the task attribute characteristics of each subtask and the model attribute characteristics of each candidate model, determine the target model that matches the subtask, and assign the subtask to the target model for processing to obtain the processing result of the online reasoning task. Based on the above method, by dividing the online reasoning task, the online reasoning task can be assigned to multiple models for processing, improving the task processing efficiency, and matching the attributes of each subtask with the attributes of the model to ensure that the target model meets the task requirements of the subtask, and fully utilize the model capabilities of the target model to improve the utilization of computing resources, and thus in the face of high load and tight computing resources, it can still maintain high-performance reasoning capabilities, improve the stability and reliability of the online reasoning system, and reduce operation and maintenance costs.

[0093] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.

[0094] like Figure 1 As shown, it is a schematic diagram of an application scenario of an embodiment of the present application. The application scenario diagram includes two terminal devices 110 and a server 120.

[0095] In the embodiments of the present application, the terminal device includes but is not limited to mobile phones, tablet computers, laptop computers, desktop computers, e-book readers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals and other devices; the terminal device may be installed with a client related to online reasoning, which may be software (such as a browser, navigation software, etc.), or a web page, a small program, etc. The server is a background server corresponding to the software or web page, small program, etc., or a server specifically used for online reasoning, which is not specifically limited in the present application. The server may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0096] It should be noted that the online reasoning method in the embodiment of the present application can be executed by an electronic device, which can be a server or a terminal device, that is, the method can be executed by a server or a terminal device alone, or by a server and a terminal device together. For example, when executed by a server and a terminal device together, the terminal device sends an online reasoning task to the server, and the server divides the online reasoning task based on the task content of the online reasoning task to obtain multiple subtasks; for each subtask, based on the respective weights of the task attribute features of each subtask, the attribute features of each task are fused to obtain a comprehensive attribute feature; based on the comprehensive attribute feature and the model attribute features of each candidate model, a target model matching the subtask is determined from each candidate model, and the subtask is assigned to the target model for processing; based on the processing results of the target models of each of the multiple subtasks, the processing result of the online reasoning task is obtained, and the processing result is returned to the client.

[0097] In an optional implementation, the terminal device and the server may communicate via a communication network.

[0098] In an optional implementation, the communication network is a wired network or a wireless network.

[0099] It should be noted that Figure 1 The figure is only an example. In fact, the number of terminal devices and servers is not limited and is not specifically limited in the embodiments of the present application.

[0100] In addition, the embodiments of the present application can be applied to various scenarios, including not only online reasoning scenarios, but also including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving and other scenarios.

[0101] The following describes the online reasoning method provided by the exemplary embodiment of the present application in combination with the application scenarios described above and with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the implementation of the present application is not limited in this respect.

[0102] See also Figure 2 As shown, it is an implementation flow chart of an online reasoning method provided in an embodiment of the present application. The specific implementation process of the method includes the following steps S21-S24:

[0103] S21: dividing the online reasoning task based on the task content of the online reasoning task to obtain multiple subtasks;

[0104] Specifically, in order to improve the processing efficiency of online reasoning tasks and improve the processing capabilities of complex tasks, the online reasoning tasks are divided in this application. For example, the task content of the online reasoning task is: generate an introductory article containing pictures and texts on the theme of apples, which can be divided into: picture generation subtask and text generation subtask according to the task content.

[0105] Based on the above method, it supports multiple models to collaboratively process complex tasks. By decomposing tasks and assigning them to different models, efficient reasoning can be achieved. This can fully utilize the advantages of each model and improve the processing efficiency and accuracy of complex tasks.

[0106] In addition, after receiving the online reasoning task to be processed, necessary preprocessing can be performed on it, such as data formatting and verification.

[0107] After dividing the online reasoning task into subtasks, each subtask is processed to obtain the task attribute characteristics of a subtask through the following steps S211-S212:

[0108] S211: extracting features of the subtask based on each preset feature extraction rule to obtain a feature vector of the subtask under each feature extraction rule;

[0109] S212: Mapping each obtained feature vector respectively to obtain attribute features of each task.

[0110] Specifically, the feature vector of a subtask may include task complexity, task data volume, real-time requirements, accuracy requirements, etc. Each feature extraction rule is used to extract a feature vector. For example, feature extraction rule 1 is used to extract task complexity. Feature extraction rule 1 is: task type 1-task complexity 001, task type 2-task complexity 002, task type 3-task complexity 003. The task type of subtask 1 is 2, and the feature vector is 002.

[0111] Mapping the feature vector may be mapping the feature vector to a higher dimension to increase the weight of the data, for example, multiplying the feature vector by 2, or mapping the feature vector to a lower dimension to reduce the complexity of data processing, or mapping the feature vector to the same numerical range to facilitate unified processing, which is not specifically limited here.

[0112] Based on the above method, it is possible to analyze task attributes in real time and respond quickly to task requirements, provide instant data support for the decision-making process, identify task features more quickly, speed up decision-making, and meet real-time or near real-time reasoning needs.

[0113] Optionally, step S212 may be implemented as follows:

[0114] Map each eigenvector to a preset numerical range to obtain the eigenvalue of each task.

[0115] Among them, the task attribute feature can be a task feature value. The feature vector is converted to a unified dimension, that is, mapped to a preset numerical range, and the obtained task feature value belongs to the preset numerical range, for example, the feature vector is normalized to a data range of 0-1.

[0116] S22: for each subtask, based on the weights of the attribute features of each task of the subtask, the attribute features of each task are integrated to obtain a comprehensive attribute feature;

[0117] Specifically, different weights can be pre-set for different task attribute features. These weights can affect the importance of task attribute features in model matching. For example, the weight of task attribute features that characterize task complexity is set to 0.5, the weight of task attribute features that characterize task data volume is set to 0.3, and the weight of task attribute features that characterize task real-time is set to 0.2. The weights can be set according to demand. For example, if the real-time requirement is high, the task attribute features that characterize task real-time are set higher. The way to fuse task attribute features can be dot product, dot multiplication, addition, etc., which are not specifically limited here.

[0118] Optionally, when the task attribute feature includes a task feature value, step S22 may be implemented as follows: based on the weight of each task attribute feature, weighted summing up the task feature values ​​to obtain a comprehensive attribute feature.

[0119] For example, Score = w1×real-time requirement score+w2×accuracy requirement score, where Score represents comprehensive attribute characteristics, the real-time requirement score is the task characteristic value representing the real-time requirement, the accuracy requirement score is the task characteristic value representing the accuracy requirement, and w1 and w2 are weights.

[0120] For another example, task feature value 1 is 0.5, weight 0.20, task feature value 2 is 0.7, weight 0.3, task feature value 3 is 0.4, weight 0.4, task feature value 4 is 0.5, weight 0.1, then the weighted sum comprehensive attribute feature is 0.52.

[0121] Based on the above method, the real-time, data volume, complexity and accuracy requirements of the task are comprehensively considered, and a quantitative score is provided for each task, which more accurately evaluates the model's ability to handle specific tasks, achieves the optimal match between tasks and models, improves the efficiency of resource allocation, reduces waiting time and waste of computing resources, and improves the overall performance of the online reasoning system.

[0122] S23: Based on the comprehensive attribute features and the model attribute features of each candidate model, determine the target model matching the subtask from each candidate model, and assign the subtask to the target model for processing;

[0123] Specifically, the comprehensive attribute characteristics can characterize the processing requirements of the subtask, and the model attribute characteristics of the candidate model characterize the processing capabilities of the model, which can be evaluated based on the model's calculation speed, maximum processing data volume, accuracy, etc. If the task requires high real-time performance and lower accuracy, it may be more suitable for a small model. If the task requires high accuracy and processing of a large amount of data, it may be more suitable for a large model. Assigning subtasks to the target model for processing can also be called assigning subtasks to the target model for execution.

[0124] In this application, a model characteristic database can also be established to record the performance indicators and applicable scenarios of each model, providing rich reference information for intelligent decision-making, making model selection more accurate and optimizing resource utilization.

[0125] Optionally, when the comprehensive attribute feature includes a comprehensive feature value, and the model attribute feature includes a model feature value, when there is a model feature value that is the same as the comprehensive feature value, the candidate model corresponding to the model feature value that is the same as the comprehensive feature value is directly used as the target model matching the subtask, and when there is no model feature value that is the same as the comprehensive feature value, step S23 can be implemented as follows:

[0126] Sort the model eigenvalues ​​in order of size to obtain the sorted model eigenvalues; construct multiple eigenvalue groups based on the model eigenvalues ​​except the last model eigenvalue, each eigenvalue group includes: a model eigenvalue and the model eigenvalue following the model eigenvalue; determine the target eigenvalue group from the multiple eigenvalue groups based on the comprehensive eigenvalue, wherein the comprehensive eigenvalue is not greater than the larger model eigenvalue in the target eigenvalue group and is not less than the smaller model eigenvalue in the target eigenvalue group; and use the candidate model corresponding to the larger model eigenvalue as the target model matching the subtask.

[0127] For example, the comprehensive eigenvalue is 4, the model eigenvalue 1 is 3, the model eigenvalue 2 is 7, the model eigenvalue 3 is 8, the model eigenvalue 4 is 2, the model eigenvalue 5 is 5, and the model eigenvalue 6 is 6. Taking the order from small to large as an example, the sorted model eigenvalues ​​are: 2, 3, 5, 6, 7, 8, and the eigenvalue groups are: [2, 3], [3, 5], [5, 6], [6, 7], [7, 8], and the target eigenvalue group is [3, 5]. In order to meet the processing requirements of the task, the candidate model corresponding to the larger model eigenvalue, that is, the candidate model 5 corresponding to the model eigenvalue 5 is selected as the target model.

[0128] Based on the above approach, the problems of resource waste and processing delay caused by model scale mismatch in related technologies are solved, ensuring that models of different scales can collaborate intelligently according to task characteristics, optimizing resource allocation and reasoning performance. Through this method, even in resource-constrained or high-load environments, fast and accurate reasoning task processing can be achieved, significantly improving overall efficiency and response speed.

[0129] Optionally, step S23 may be implemented as follows:

[0130] Based on the current resource usage status and the resource requirements of the target model, determine the resource allocation strategy of the target model; schedule the cloud native platform to execute the resource allocation strategy, and use the allocated computing resources to run the target model processing subtasks.

[0131] Specifically, the cloud native platform dynamically allocates computing resources and implements load balancing through model scheduling strategies to optimize resource utilization and response speed, considering the current resource usage status, such as the utilization rate of the central processing unit (CPU), memory, and graphics processing unit (GPU) to ensure that the resource allocation strategy conforms to the actual resource situation.

[0132] The above decision-making process takes into account the current system resource status, including the availability and load of computing resources. Through resource status perception, resource overload is avoided, the rationality of task allocation is ensured, and the stability and reliability of the system are improved.

[0133] In the embodiments of this application, the cloud-native architecture features such as microservices, containerization, continuous integration and continuous deployment (CI / CD) are fully utilized to simplify model management and automated deployment in the cloud-native environment, and efficient collaboration of multiple models in the cloud-native architecture is achieved, further improving the scalability and flexibility of the online reasoning system. The innovative collaboration mechanism and automation technology not only improve the stability and reliability of the system, but also reduce the operation and maintenance costs, providing users with an economical, efficient, easy-to-maintain and scalable online reasoning solution.

[0134] S24: Based on the processing results of the target models of the multiple subtasks, the processing result of the online reasoning task is obtained.

[0135] Among them, the processing results of the target models of the subtasks are summarized to obtain the processing results of the online reasoning task.

[0136] Optionally, the weight of each task attribute feature is obtained based on the feature category to which the task attribute feature belongs, and the processing status information of the target model in the process of processing the subtask is obtained; based on the processing status information, the weight corresponding to each feature category is adjusted.

[0137] Specifically, as mentioned above, the weight of task attribute features can affect the importance of task attribute features in model matching. Therefore, it is also necessary to detect the processing status information of the target model, such as execution time, whether lag occurs, etc., and dynamically adjust the weight of feature categories. For example, when it is detected that the processing time of the target model is long, the weight of the feature category with real-time requirements can be increased. When the next online reasoning task is received, the comprehensive attribute characteristics of the task can be calculated more accurately according to the adjusted weight, thereby improving the matching degree between the task and the model.

[0138] Based on this approach, the online reasoning method in this application has an adaptive learning function, which can adjust the scoring formula and decision logic according to historical tasks and model performance, so that the system can continuously optimize the decision-making process and adapt to the ever-changing task characteristics and system environment.

[0139] In an embodiment of the present application, the online reasoning task is divided to obtain multiple subtasks; based on the attribute characteristics of each subtask and the model attribute characteristics of each candidate model, the target model matching the subtask is determined, and the subtask is assigned to the target model for processing to obtain the processing result of the online reasoning task. Based on the above method, by dividing the online reasoning task, the online reasoning task can be assigned to multiple models for processing, improving the task processing efficiency, and matching the attributes of each subtask with the attributes of the model to ensure that the target model meets the task requirements of the subtask, and fully utilize the model capabilities of the target model to improve the utilization of computing resources, and thus in the face of high load and tight computing resources, it can still maintain high-performance reasoning capabilities, improve the stability and reliability of the online reasoning system, and reduce operation and maintenance costs.

[0140] like Figure 3 As shown, it is a structural diagram of an online reasoning system in an embodiment of the present application, including an online reasoning controller, an external business system, a cloud native platform, and a model warehouse, wherein the online reasoning controller is used to execute the online reasoning method in the present application, manage and coordinate the loading, updating, and online reasoning task allocation of model services. It includes a task collaboration management module, a model data management module, a scheduling strategy control module, a decision algorithm management module, a model service management module, and a task queue management module. Among them:

[0141] Task coordination management module: receives online reasoning tasks, decomposes task attributes, and is responsible for subtask allocation, status tracking, and dependency management;

[0142] Task queue management module: It is a model registry that records the attributes, capabilities, and applicable scenarios of each model. It monitors the running status of the model in real time to ensure the smooth execution of tasks. It collects and synchronizes the reasoning results (processing results) of each model, integrates them effectively, and obtains the processing results of online reasoning tasks;

[0143] Model data management module: maintains various model image files, with functions such as upload and delete. It is the static data of the model;

[0144] Model service management module: maintains dynamic data of various models under running status;

[0145] Decision algorithm management module: Based on the subtask attribute characteristics and the model attribute characteristics of the model, a model selection algorithm is formed to automatically match the most appropriate model;

[0146] Scheduling policy control module: It is a resource allocator that dynamically adjusts resources according to model requirements and system resource status.

[0147] External business system: interacts with the online inference controller, sends online inference tasks and receives processing results.

[0148] Model warehouse: A management warehouse for trained model files and model metadata information by external model builders for specific machine learning tasks based on different frameworks, which are delivered to the online inference controller to obtain pre-trained models or upgrade iteratively developed models.

[0149] Cloud native platform: A cloud computing environment that supports microservices, containerization, and continuous integration deployment. It receives model scheduling strategies (resource allocation strategies) issued by online inference servers, allocates resource scheduling models, provides inference services, and outputs subtask processing results. The Kubernetes (K8s) application programming interface server (Application Programming Interface Server) serves as the control center of the K8s cluster, handles the communication and management within the cluster, and receives resource allocation strategies issued by the online inference controller.

[0150] like Figure 4 FIG. 1 is a schematic diagram of the overall process of an online reasoning method in an embodiment of the present application, comprising the following steps:

[0151] S41: receiving online reasoning tasks;

[0152] S42: dividing the online reasoning task into multiple subtasks based on the task content of the online reasoning task;

[0153] S43: Obtaining fitness scores based on the attributes and weights of the subtasks;

[0154] S44: based on the fitness score of the subtask and the model capability of the candidate model, obtaining a target model for processing the subtask;

[0155] S45: assigning subtasks to the target model and allocating computing resources according to the resource requirements of the model;

[0156] S46: Summarize the processing results of each target model to obtain the processing result of the online reasoning task;

[0157] S47: Adjust the weight of the attribute based on the processing result.

[0158] Based on the same inventive concept, the present application embodiment also provides an online reasoning device. Figure 5 As shown, it is a schematic diagram of the structure of the online reasoning device 500, which may include:

[0159] A division unit 501 is used to divide the online reasoning task based on the task content of the online reasoning task to obtain multiple subtasks;

[0160] A fusion unit 502 is used to fuse the attribute features of each subtask based on the weights of the attribute features of each subtask to obtain a comprehensive attribute feature;

[0161] A determination unit 503 is used to determine a target model matching the subtask from each candidate model based on the comprehensive attribute characteristics and the model attribute characteristics of each candidate model, and assign the subtask to the target model for processing;

[0162] The summarizing unit 504 is used to obtain the processing result of the online reasoning task based on the processing results of the target models of the multiple subtasks.

[0163] Optionally, the fusion unit 502 is specifically configured to obtain attribute features of each task in the following manner:

[0164] Extract features of the subtasks based on the preset feature extraction rules to obtain feature vectors of the subtasks under the feature extraction rules;

[0165] Map each feature vector obtained to obtain the attribute features of each task.

[0166] Optionally, the task attribute feature includes a task feature value; the fusion unit 502 is specifically used for:

[0167] Mapping each feature vector to a preset numerical interval respectively to obtain a feature value of each task, and each feature value of each task belongs to the preset numerical interval;

[0168] Based on the weights of each task attribute feature of the subtask, the attribute features of each task are fused to obtain comprehensive attribute features, including:

[0169] Based on the weights of the attribute features of each task, the feature values ​​of each task are weighted and summed to obtain the comprehensive attribute features.

[0170] Optionally, the determining unit 503 is specifically configured to:

[0171] Determine the resource allocation strategy of the target model based on the current resource usage status and the resource requirements of the target model;

[0172] The scheduling cloud native platform executes the resource allocation strategy and uses the allocated computing resources to run the target model processing subtasks.

[0173] Optionally, the weight of each task attribute feature is obtained based on the feature category to which the task attribute feature belongs; the device further includes an adjustment unit 505, which is used to:

[0174] Obtain the processing status information of the target model during the processing of subtasks;

[0175] Based on the processing status information, the weights corresponding to each feature category are adjusted.

[0176] Optionally, the comprehensive attribute feature includes a comprehensive feature value, and the model attribute feature includes a model feature value; the determining unit 503 is specifically used to:

[0177] Sort the eigenvalues ​​of each model in order of size to obtain the sorted eigenvalues ​​of each model;

[0178] Based on each model eigenvalue except the last model eigenvalue, a plurality of eigenvalue groups are constructed, each eigenvalue group comprising: a model eigenvalue and a model eigenvalue of the next digit of the model eigenvalue;

[0179] Determining a target eigenvalue group from a plurality of eigenvalue groups based on a comprehensive eigenvalue, wherein the comprehensive eigenvalue is not greater than a larger model eigenvalue in the target eigenvalue group and not less than a smaller model eigenvalue in the target eigenvalue group;

[0180] The candidate model corresponding to the larger model feature value is taken as the target model matching the subtask.

[0181] For the convenience of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, when implementing this application, the functions of each module (or unit) can be implemented in the same or multiple software or hardware.

[0182] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0183] Those skilled in the art will appreciate that various aspects of the present application may be implemented as a system, method or program product. Therefore, various aspects of the present application may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to as "circuit", "module" or "system" herein.

[0184] Based on the same inventive concept as the above method embodiment, an electronic device is also provided in the embodiment of the present application. In one embodiment, the electronic device may be a server, such as Figure 1 In this embodiment, the structure of the electronic device can be as follows: Figure 6 As shown, it includes a memory 601 , a communication module 603 and one or more processors 602 .

[0185] The memory 601 is used to store computer programs executed by the processor 602. The memory 601 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and programs required for running the instant messaging function, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0186] The memory 601 may be a volatile memory, such as a random-access memory (RAM); the memory 601 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or the memory 601 may be any other medium that can be used to carry or store a desired computer program in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 601 may be a combination of the above memories.

[0187] The processor 602 may include one or more central processing units (CPU) or a digital processing unit, etc. The processor 602 is used to implement the above online reasoning method when calling the computer program stored in the memory 601 .

[0188] The communication module 603 is used to communicate with terminal devices and other servers.

[0189] The specific connection medium between the memory 601, the communication module 603 and the processor 602 is not limited in the embodiment of the present application. Figure 6 In the embodiment, the memory 601 and the processor 602 are connected via a bus 604. The bus 604 is connected to the processor 602 via a bus 604. Figure 6 The connections between the other components are only for illustration and are not intended to be limiting. The bus 604 can be divided into an address bus, a data bus, a control bus, etc. For ease of description, Figure 6 The diagram shows that only one thick line is used, but this does not mean that there is only one bus or only one type of bus.

[0190] The memory 601 stores a computer storage medium, and the computer storage medium stores computer executable instructions, and the computer executable instructions are used to implement the online reasoning method of the embodiment of the present application. The processor 602 is used to execute the above-mentioned online reasoning method, such as Figure 2 shown.

[0191] In another embodiment, the electronic device may also be other electronic devices, such as Figure 1 In this embodiment, the structure of the electronic device can be as follows: Figure 7 As shown, it includes: a communication component 710, a memory 720, a display unit 730, a camera 740, a sensor 750, an audio circuit 760, a Bluetooth module 770, a processor 780 and other components.

[0192] The communication component 710 is used to communicate with the server. In some embodiments, a wireless fidelity (WiFi) module may be included. The WiFi module belongs to a short-range wireless transmission technology. The electronic device can help the user to send and receive information through the WiFi module.

[0193] The memory 720 can be used to store software programs and data. The processor 780 executes various functions and data processing of the terminal device by running the software programs or data stored in the memory 720. The memory 720 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. The memory 720 stores an operating system that enables the terminal device to run. In the present application, the memory 720 can store an operating system and various application programs, and can also store a computer program for executing the online reasoning method of the embodiment of the present application.

[0194] The display unit 730 can also be used to display information input by the user or information provided to the user and a graphical user interface (GUI) of various menus of the terminal device. Specifically, the display unit 730 may include a display screen 732 disposed on the front of the terminal device. The display screen 732 may be configured in the form of a liquid crystal display, a light emitting diode, etc. The display unit 730 may be used to display the online reasoning user interface in the embodiment of the present application, etc.

[0195] The display unit 730 can also be used to receive input digital or character information and generate signal input related to user settings and function control of the terminal device. Specifically, the display unit 730 may include a touch screen 731 arranged on the front of the terminal device, which can collect user touch operations on or near it, such as clicking a button, dragging a scroll box, etc.

[0196] The touch screen 731 can be covered on the display screen 732, or the touch screen 731 and the display screen 732 can be integrated to realize the input and output functions of the terminal device, and the integrated display screen can be referred to as a touch display screen. In this application, the display unit 730 can display the application and the corresponding operation steps.

[0197] The camera 740 can be used to capture still images, and users can post comments on the images taken by the camera 740 through the application. There can be one or more cameras 740. The object generates an optical image through the lens and projects it onto the photosensitive element. The photosensitive element can be a charge coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the light signal into an electrical signal, and then passes the electrical signal to the processor 780 to convert it into a digital image signal.

[0198] The terminal device may further include at least one sensor 750, such as an acceleration sensor 751, a distance sensor 752, a fingerprint sensor 753, and a temperature sensor 754. The terminal device may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, a light sensor, and a motion sensor.

[0199] The audio circuit 760, the speaker 761, and the microphone 762 can provide an audio interface between the user and the terminal device. The audio circuit 760 can transmit the electrical signal converted from the received audio data to the speaker 761, which is converted into a sound signal for output. The terminal device can also be equipped with a volume button for adjusting the volume of the sound signal. On the other hand, the microphone 762 converts the collected sound signal into an electrical signal, which is received by the audio circuit 760 and converted into audio data, and then the audio data is output to the communication component 710 to be sent to, for example, another terminal device, or the audio data is output to the memory 720 for further processing.

[0200] The Bluetooth module 770 is used to exchange information with other Bluetooth devices having Bluetooth modules through the Bluetooth protocol. For example, the terminal device can establish a Bluetooth connection with a wearable electronic device (such as a smart watch) that also has a Bluetooth module through the Bluetooth module 770 to exchange data.

[0201] The processor 780 is the control center of the terminal device. It uses various interfaces and lines to connect various parts of the entire terminal. It executes various functions of the terminal device and processes data by running or executing software programs stored in the memory 720 and calling data stored in the memory 720. In some embodiments, the processor 780 may include one or more processing units; the processor 780 may also integrate an application processor and a baseband processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the baseband processor mainly processes wireless communications. It is understandable that the above-mentioned baseband processor may not be integrated into the processor 780. In the present application, the processor 780 can run the operating system, application programs, user interface display and touch response, as well as the online reasoning method of the embodiment of the present application. In addition, the processor 780 is coupled to the display unit 730.

[0202] In some possible implementations, various aspects of the online reasoning method provided by the present application may also be implemented in the form of a program product, which includes a computer program. When the program product is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of the online reasoning method according to various exemplary embodiments of the present application described above in this specification. For example, the electronic device may execute the following steps: Figure 2 Follow the steps shown in .

[0203] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0204] The program product of the embodiment of the present application may adopt a portable compact disk read-only memory (CD-ROM) and include a computer program, and can be run on an electronic device. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which can be used by or in combination with a command execution system, apparatus, or device.

[0205] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable computer program is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0206] The computer program embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0207] Computer programs for performing the operations of the present application may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The computer program may be executed entirely on the user electronic device, partially on the user device, as a separate software package, partially on the user electronic device and partially on a remote electronic device, or entirely on a remote electronic device or server. In the case of a remote electronic device, the remote electronic device may be connected to the user electronic device via any type of network including a local area network (LAN) or a wide area network (WAN), or may be connected to an external electronic device (e.g., via the Internet using an Internet service provider).

[0208] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0209] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0210] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain a computer-usable computer program.

[0211] The present application 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 the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program commands. These computer program commands 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 generate a machine, so that the commands executed by the processor of the computer or other programmable data processing device generate commands 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.

[0212] These computer program commands may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the commands stored in the computer readable memory produce an article of manufacture comprising a command device, the command device implementing the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0213] These computer program commands can also be loaded onto a computer or other programmable data processing device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the commands executed on the computer or other programmable device provide the instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0214] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0215] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. An online reasoning method, It is characterized in that The method comprises: Based on the task content of the online reasoning task, the online reasoning task is divided to obtain multiple subtasks; For each subtask, based on the weights of the respective task attribute features of the subtask, the attribute features of the respective tasks are integrated to obtain a comprehensive attribute feature; Based on the comprehensive attribute features and the model attribute features of each candidate model, determine a target model matching the subtask from the candidate models, and assign the subtask to the target model for processing; Based on the processing results of the target models of the multiple subtasks, the processing result of the online reasoning task is obtained.

2. The method according to claim 1, It is characterized in that The attribute characteristics of each task are obtained by: Performing feature extraction on the subtask based on each preset feature extraction rule to obtain a feature vector of the subtask under each feature extraction rule; The obtained feature vectors are mapped respectively to obtain the attribute features of each task.

3. The method according to claim 2, It is characterized in that The task attribute characteristics include task characteristic values; The obtained feature vectors are mapped respectively to obtain the attribute features of each task, including: Mapping each of the characteristic vectors to a preset numerical interval respectively to obtain each task characteristic value, wherein each task characteristic value belongs to the preset numerical interval; The step of fusing the task attribute features based on the weights of the task attribute features of the subtasks to obtain a comprehensive attribute feature includes: Based on the weights of the respective task attribute features, the task feature values ​​are weighted and summed to obtain the comprehensive attribute feature.

4. The method according to any one of claims 1 to 3, It is characterized in that The allocating the subtasks to the target model for processing includes: Determining a resource allocation strategy for the target model based on the current resource usage status and the resource requirements of the target model; The cloud native platform is scheduled to execute the resource allocation strategy, and use the allocated computing resources to run the target model to process the subtask.

5. The method according to any one of claims 1 to 3, It is characterized in that The weight of each task attribute feature is obtained based on the feature category to which the task attribute feature belongs; the method further includes: Acquiring processing status information of the target model in the process of processing the subtask; Based on the processing status information, the weights corresponding to the respective feature categories are adjusted.

6. The method according to any one of claims 1 to 3, It is characterized in that The comprehensive attribute feature includes a comprehensive feature value, and the model attribute feature includes a model feature value; and determining a target model matching the subtask from each candidate model based on the comprehensive attribute feature and the model attribute features of each candidate model includes: Sort the eigenvalues ​​of each model in order of size to obtain the sorted eigenvalues ​​of each model; Based on each model eigenvalue except the last model eigenvalue, a plurality of eigenvalue groups are constructed, each eigenvalue group comprising: a model eigenvalue and a model eigenvalue next to the model eigenvalue; Determining a target feature value group from the plurality of feature value groups based on the comprehensive feature value, wherein the comprehensive feature value is not greater than a larger model feature value in the target feature value group and not less than a smaller model feature value in the target feature value group; The candidate model corresponding to the larger model feature value is used as the target model matching the subtask.

7. An online reasoning device, It is characterized in that include: A division unit, used for dividing the online reasoning task based on the task content of the online reasoning task to obtain a plurality of subtasks; A fusion unit, for fusing the attribute features of each subtask based on the weights of the attribute features of each subtask to obtain a comprehensive attribute feature; a determination unit, configured to determine a target model matching the subtask from the candidate models based on the comprehensive attribute feature and the model attribute features of each candidate model, and assign the subtask to the target model for processing; A summarizing unit is used to obtain the processing result of the online reasoning task based on the processing results of the target models of the multiple subtasks.

8. An electronic device, It is characterized in that It comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the methods of claims 1 to 6.

9. A computer-readable storage medium, It is characterized in that It comprises a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of any one of the methods of claims 1 to 6.

10. A computer program product, It is characterized in that The method comprises a computer program stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device executes the steps of any one of the methods described in claims 1 to 6.

Images