A resource configuration updating method and apparatus, a storage medium, and an electronic device

Abstract

The application discloses a resource configuration updating method and device, a storage medium and an electronic device, relates to the technical field of education, and comprises the following steps: obtaining historical task execution data of an agent to be updated; comprehensively evaluating the historical task execution of the agent to be updated according to the historical task execution data, obtaining execution capability data representing the task execution capability of the agent to be updated; determining a level updating type of the agent to be updated, and updating the resource configuration level of the agent to be updated according to the level updating type, to obtain a target resource configuration level corresponding to the agent to be updated; and performing resource configuration on the agent to be updated according to the target resource configuration level, to determine target resource configuration information corresponding to the agent to be updated. The application updates the resource configuration of the agent by determining the level updating type and updating the resource configuration level, realizes reasonable allocation of computing power resources, and guarantees the business processing capability of the agent.

Description

Technical Field

[0001] This application relates to the field of educational technology, and in particular to a method, apparatus, storage medium, and electronic device for updating resource allocation. Background Technology

[0002] With the development of artificial intelligence technology, AI agents are widely used in educational business scenarios such as teaching assistance, academic affairs processing, and student management.

[0003] Currently, the AI ​​agent hosting platform allocates initial computing resources to each AI agent during the creation phase, and then adjusts these resources based on parameters such as the amount of tokens consumed by each AI agent during task execution. AI agents with high token consumption are allocated sufficient computing resources, while those with low token consumption are allocated less computing resources.

[0004] However, the frequency of AI agent usage varies across different educational business scenarios, and the usage requirements of AI agents within the same educational business scenario will change dynamically over time. Existing technologies cannot adapt and adjust resources according to the differences in the frequency of AI agent usage and the changes in usage requirements, which will lead to a waste of computing resources or a decline in the business performance of AI agents. Summary of the Invention

[0005] In view of this, this application provides a resource configuration update method, apparatus, storage medium, and electronic device. The main purpose is to improve the technical problem that the usage frequency of AI agents varies in different educational business scenarios, and the usage needs of AI agents in the same educational business scenario will change dynamically with the business time. Existing technologies cannot adapt and adjust resources according to the differences in the usage frequency and changes in usage needs of AI agents, which will lead to the waste of computing resources or the decline in business performance of AI agents.

[0006] Firstly, this application provides a resource configuration update method, including:

[0007] Obtain historical task execution data of the agent to be updated. The historical task execution data includes agent operation data of the agent to be updated executing historical tasks, as well as evaluation data of the user who initiated the historical task on the historical task execution status of the agent to be updated.

[0008] Based on the historical task execution data, a comprehensive evaluation of the historical task execution status of the agent to be updated is performed to obtain execution capability data representing the task execution capability of the agent to be updated;

[0009] Based on the execution capability data, the level update type of the agent to be updated is determined, and the resource configuration level of the agent to be updated is updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated.

[0010] Based on the target resource configuration level, the agent to be updated is configured with resources to determine the target resource configuration information corresponding to the agent to be updated. The target resource configuration information is used by the agent to be updated to execute the task to be executed.

[0011] Optionally, the step of comprehensively evaluating the historical task execution status of the agent to be updated based on the historical task execution data to obtain execution capability data representing the task execution capability of the agent to be updated includes:

[0012] Based on the target resource consumption data of the agent to be updated, the resource consumption data is normalized to obtain the normalized resource consumption data of the agent to be updated.

[0013] The performance data of the agent to be updated and the task complexity data of the historical tasks are processed separately to obtain comprehensive performance data and comprehensive task complexity data.

[0014] The normalized resource consumption data, comprehensive performance data, comprehensive task complexity data, and evaluation data are fused together to obtain execution capability data representing the task execution capability of the agent to be updated.

[0015] Optionally, the process of comprehensively processing the performance data of the agent to be updated and the task complexity data of the historical tasks to obtain comprehensive performance data and comprehensive task complexity data includes:

[0016] The comprehensive performance data is obtained by weighting the response latency data, the generation time of task reply information, and the success rate of the tool call by the agent to be updated in the performance data generated by the agent to be updated executing historical tasks.

[0017] The confidence and complexity of each intent in the historical tasks are weighted to obtain the basic task complexity data;

[0018] Based on the basic task complexity data and the contextual association data of the historical tasks, the comprehensive task complexity data is obtained.

[0019] Optionally, the normalized resource consumption data, comprehensive performance data, comprehensive task complexity data, and evaluation data are fused together to obtain execution capability data representing the task execution capability of the agent to be updated, including:

[0020] Based on a preset task difficulty correction coefficient, the comprehensive task complexity data is corrected to obtain task difficulty correction data. The evaluation data and the task difficulty correction data are then processed to obtain evaluation difficulty fusion data.

[0021] Based on the normalized resource consumption data, resource consumption constraint data is obtained; based on the evaluation difficulty fusion data and the resource consumption constraint data, resource constraint evaluation data is obtained.

[0022] The resource constraint evaluation data and the comprehensive performance data are processed to obtain the previous execution capability data of the intelligent agent to be updated in executing the historical task;

[0023] Based on a preset smoothing coefficient, the execution capability data other than the previous execution capability data in the previous execution capability data and historical execution capability data are weighted to obtain execution capability data representing the task execution capability of the agent to be updated.

[0024] Optionally, the step of configuring resources for the agent to be updated based on the target resource configuration level to determine the target resource configuration information corresponding to the agent to be updated, wherein the target resource configuration information is used by the agent to be updated to execute the task to be executed, includes:

[0025] If the target resource configuration level is higher or lower than the current resource configuration level, the resource configuration information is adjusted according to the target resource configuration level to obtain the target resource configuration information;

[0026] If the target resource configuration level is equal to the current resource configuration level, the resource configuration information corresponding to the current resource configuration level shall be determined as the target resource configuration information.

[0027] Optionally, the step of determining the level update type of the agent to be updated based on the execution capability data, and updating the resource configuration level of the agent to be updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated, includes:

[0028] If the execution capability data is greater than or equal to the upgrade level threshold, the level update type of the agent to be updated is determined to be the upgrade level type, and the resource configuration level of the agent to be updated is upgraded based on the execution capability data to obtain the first resource configuration level of the agent to be updated after the upgrade.

[0029] If the execution capability data is greater than or equal to the reduction level threshold and less than the increase level threshold, the level update type of the agent to be updated is determined to be the maintenance level type, and the resource configuration level of the agent to be updated is determined to be the second target resource configuration level based on the execution capability data.

[0030] If the execution capability data is less than the downgrade threshold, the level update type of the agent to be updated is determined to be a downgrade type, and the resource configuration level of the agent to be updated is downgraded based on the execution capability data to obtain the third resource configuration level of the agent to be updated after downgrading.

[0031] Optionally, when the execution capability data is less than the downgrade threshold, determining that the level update type of the agent to be updated is a downgrade type, and downgrading the resource configuration level of the agent to be updated based on the execution capability data to obtain the downgraded third resource configuration level of the agent to be updated, includes:

[0032] Based on the execution capability data, the number of tasks and task permissions for the agent to be updated to execute the tasks to be executed are reduced, and the execution status of the tasks to be executed is monitored;

[0033] In response to the detection that the level update type of the task to be performed by the agent to be updated is either the level upgrade type or the level maintenance type, the current resource configuration level of the agent to be updated is determined as the target resource configuration level;

[0034] In response to the detection that the level update type of the task to be performed by the agent to be updated is a downgrade type, the resource configuration level of the agent to be updated is downgraded to obtain the third resource configuration level of the agent to be updated.

[0035] Optionally, before obtaining the historical task execution data of the agent to be updated, the method further includes:

[0036] Obtain the rating data determined by the user based on the historical task execution results, response speed, and interactive experience of the intelligent agent to be updated;

[0037] The interaction text information between the user and the agent to be updated during the execution of the historical task is obtained, and the interaction text information is subjected to semantic parsing and sentiment analysis to obtain sentiment change data.

[0038] The rating data and the emotion change data are weighted and fused to obtain the evaluation data.

[0039] Secondly, this application provides a resource configuration update apparatus, comprising:

[0040] The acquisition module is configured to acquire historical task execution data of the agent to be updated. The historical task execution data includes agent operation data of the agent to be updated executing historical tasks, and evaluation data of the user who initiated the historical task on the historical task execution status of the agent to be updated.

[0041] The evaluation module is configured to comprehensively evaluate the historical task execution status of the agent to be updated based on the historical task execution data, and obtain execution capability data representing the task execution capability of the agent to be updated;

[0042] The update module is configured to determine the level update type of the agent to be updated based on the execution capability data, and update the resource configuration level of the agent to be updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated.

[0043] The determination module is configured to configure resources for the agent to be updated based on the target resource configuration level, and determine the target resource configuration information corresponding to the agent to be updated. The target resource configuration information is used by the agent to be updated to execute the task to be executed.

[0044] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the resource configuration update method described in the first aspect.

[0045] Fourthly, this application provides an electronic device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the resource configuration update method described in the first aspect.

[0046] Fifthly, this application provides a computer program product, which includes a computer program that, when executed by a processor, implements the resource configuration update method described in the first aspect.

[0047] By means of the above technical solution, this application provides a resource configuration update method, apparatus, storage medium, and electronic device, comprising: acquiring historical task execution data of the intelligent agent to be updated, the historical task execution data including intelligent agent operation data of the intelligent agent to be updated executing historical tasks, and evaluation data of the user who initiated the historical task on the historical task execution status of the intelligent agent to be updated; comprehensively evaluating the historical task execution status of the intelligent agent to be updated based on the historical task execution data to obtain execution capability data representing the task execution capability of the intelligent agent to be updated; determining the level update type of the intelligent agent to be updated based on the execution capability data, and updating the resource configuration level of the intelligent agent to be updated according to the level update type to obtain the target resource configuration level corresponding to the intelligent agent to be updated; configuring resources for the intelligent agent to be updated according to the target resource configuration level to determine the target resource configuration information corresponding to the intelligent agent to be updated, the target resource configuration information being used by the intelligent agent to be updated to execute the task to be executed. Compared with existing technologies, this application obtains execution capability data by comprehensively evaluating the historical task execution of the agent based on historical task execution data, providing a precise basis for the level update of the agent to be updated; by determining the level update type based on the execution capability data and updating the resource configuration level to obtain the target resource configuration level, the agent's resources are configured according to the target resource configuration level, realizing the adaptation between the agent's resource configuration and the target resource configuration level, so that the resource configuration fits the actual needs of the agent to execute the task to be executed, realizing the rational allocation of computing resources and ensuring the agent's business processing capabilities. Attached Figure Description

[0048] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart illustrating a resource configuration update method provided in an embodiment of this application is shown.

[0051] Figure 2 A flowchart illustrating a resource configuration update method provided in an embodiment of this application is shown.

[0052] Figure 3 This paper shows a schematic diagram of the structure of a resource configuration update device provided in an embodiment of this application;

[0053] Figure 4A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0054] The embodiments of this application will now be described in more detail with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0055] To address the issue that existing technologies cannot adapt and adjust resources based on varying usage frequencies and changing needs of AI agents across different educational scenarios, leading to wasted computing resources or decreased performance, this embodiment provides a resource configuration update method. Figure 1 As shown, the method includes:

[0056] Step 101: Obtain historical task execution data of the agent to be updated.

[0057] The historical task execution data includes the agent operation data of the agent to be updated executing historical tasks, as well as the evaluation data of the user who initiated the historical task on the historical task execution of the agent to be updated.

[0058] In this embodiment, the intelligent agent to be updated can be various AI agents on an intelligent agent hosting platform. These agents can be used to perform business tasks in different vertical fields such as education, customer service, healthcare, and finance. For example, the intelligent agent to be updated in this embodiment can specifically be an educational scenario intelligent assistant A serving homeroom teachers. This agent can be used to complete education-related tasks such as student performance analysis, writing home-school notices, responding to emergencies, and managing mental health records.

[0059] In this embodiment, historical task execution data can be data generated during the execution of historical business tasks by the agent to be updated. Historical task execution data can be used to provide data support for the comprehensive performance evaluation and resource allocation update of the agent to be updated.

[0060] In the embodiments of this application, the agent operation data can be indicator data related to the operation status, resource consumption, and task difficulty generated during the execution of historical tasks by the agent to be updated. The agent operation data can be used to reflect the actual operation status, resource cost, and task processing difficulty of the agent. For example, the agent operation data in the embodiments of this application may specifically include resource consumption data, performance data, and task complexity data of historical tasks.

[0061] In this embodiment of the application, the evaluation data can be feedback data provided by the user who initiated the historical task regarding the task execution results and service experience of the agent to be updated. The evaluation data can be used to reflect the user's satisfaction with the agent's service. For example, the evaluation data in this embodiment of the application can be generated by fusing the user's explicit rating data and implicit emotion change data obtained based on the analysis of the user's interaction text with the agent.

[0062] Step 102: Based on the historical task execution data, conduct a comprehensive evaluation of the historical task execution status of the agent to be updated, and obtain execution capability data representing the task execution capability of the agent to be updated.

[0063] In this embodiment, the execution capability data can be an indicator obtained by quantifying the overall task execution level of the agent to be updated. The execution capability data can be used to characterize the overall ability of the agent to be updated to handle various business tasks. For example, the execution capability data in this embodiment can be a comprehensive quantitative score that integrates multiple dimensions of indicators such as resource consumption, execution efficiency, task difficulty, and user evaluation. This score can be within a preset numerical range of the platform, and the higher the score, the stronger the task execution capability of the agent.

[0064] In the embodiments of this application, a comprehensive evaluation of the historical task execution of the agent to be updated can be completed through the multi-dimensional dynamic performance evaluation engine of the agent hosting platform. The engine can construct a three-dimensional evaluation system that includes difficulty incentives, cost constraints, and efficiency corrections. Specifically, it can include: standardizing and normalizing the collected historical task execution data, and then using a preset fusion algorithm to fuse and calculate the multi-dimensional preprocessed data to obtain execution capability data that reflects the actual task execution capability of the agent to be updated.

[0065] Step 103: Based on the execution capability data, determine the level update type of the agent to be updated, and update the resource configuration level of the agent to be updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated.

[0066] In this embodiment, the level update type can be a change type of the resource configuration level of the agent to be updated, determined based on execution capability data. For example, the level update type in this embodiment can specifically include three types: level upgrade, level maintenance, and level downgrade.

[0067] In this embodiment, the resource configuration level can be a hierarchy of computing power resource allocation divided by the intelligent agent hosting platform for intelligent agents with different capabilities and business needs. The resource configuration level can be used to match the task execution capabilities and computing power resources of the intelligent agents to realize the allocation of computing power resources. For example, the resource configuration level in this embodiment can specifically include five core job levels: Intern L1, Junior L2, Backbone L3, Senior L4, and Expert L5, and can also include two special states: Performance Improvement Plan (PIP) and Frozen. Different job levels can correspond to different computing power resource allocation standards.

[0068] In this embodiment, the target resource configuration level can be the resource configuration level matched by the agent to be updated after the level update operation. For example, the target resource configuration level in this embodiment can be the intelligent assistant A upgraded from intern level L1 to backbone level L3, or the PIP state that has entered the rectification period due to a decline in performance.

[0069] In the embodiments of this application, the determination of the level update type and the update of the resource configuration level can be completed through the intelligent agent job level state machine and lifecycle manager of the intelligent agent hosting platform. The manager can maintain a highly available finite state machine and adopt a jitter-free state transition mechanism based on hysteresis comparators. For each resource configuration level, a corresponding level upgrade threshold and a level down threshold are preset. When the execution capability data continuously reaches the threshold conditions and meets the preset stability test cycle requirements, the corresponding level update type can be determined, ensuring the accuracy and stability of the level update determination.

[0070] Step 104: Configure resources for the agent to be updated according to the target resource configuration level, determine the target resource configuration information corresponding to the agent to be updated, and use the target resource configuration information to execute the task to be executed by the agent to be updated.

[0071] In the embodiments of this application, the target resource configuration information may be a set of computing power resource configuration parameters that match the target resource configuration level. The target resource configuration information can be used to allocate corresponding computing power resources for the agent to be updated to perform subsequent tasks. For example, the target resource configuration information in the embodiments of this application may specifically include core configuration parameters such as base model specifications, context window parameters, tool call permissions, and concurrent quotas of queries per second (QPS).

[0072] In the embodiments of this application, the resource orchestration executor of the intelligent agent hosting platform can perform specific resource configuration operations. This executor can serve as the final execution terminal for the interaction between the system and the intelligent agent hosting platform, transforming the decisions of the intelligent agent job level state machine and lifecycle manager into specific underlying resource change operations. It can realize operations such as hot switching of the base model, elastic scaling of the context window, dynamic configuration of the tool authentication gateway, and updating of computing power routing policies by calling the interface (Admin API) of the intelligent agent hosting platform. Moreover, the end-to-end effective time of configuration changes can be controlled within a preset range, while maintaining the affinity of the ongoing session and avoiding service interruption.

[0073] Compared with existing technologies, this embodiment obtains execution capability data by comprehensively evaluating the historical task execution status of the agent based on historical task execution data, providing a precise basis for the level update of the agent to be updated; by determining the level update type based on the execution capability data and updating the resource configuration level to obtain the target resource configuration level, the agent's resources are configured according to the target resource configuration level, realizing the adaptation between the agent's resource configuration and the target resource configuration level, so that the resource configuration fits the actual needs of the agent to execute the task to be executed, realizing the rational allocation of computing resources and ensuring the agent's business processing capabilities.

[0074] As an optional approach, when performing the task of "comprehensively evaluating the historical task execution performance of the agent to be updated based on historical task execution data to obtain execution capability data representing the task execution capability of the agent to be updated," the following methods can be used, but are not limited to these: Figure 2 As shown, the method includes:

[0075] Step 201: Based on the target resource consumption data of the agent to be updated, normalize the resource consumption data to obtain the normalized resource consumption data of the agent to be updated.

[0076] In this embodiment of the application, the target resource consumption data can be the resource consumption benchmark data set by the intelligent agent hosting platform for the intelligent agent to be updated. The target resource consumption data can be used as a reference for the normalization processing of resource consumption data to eliminate the difference in the dimensions of resource consumption under different levels of intelligent agents and different task scenarios.

[0077] In this embodiment, resource consumption data can be various resource consumption-related data generated during the execution of historical tasks by the agent to be updated. This resource consumption data can reflect the actual resource cost of the agent executing tasks. For example, the resource consumption data in this embodiment may specifically include the number of input tokens, the number of output tokens, and the unit price data of the corresponding tokens generated during the execution of historical tasks.

[0078] In this embodiment, the normalized resource consumption data can be standardized data obtained by normalizing the resource consumption data by comparing it with the target resource consumption data. The normalized resource consumption data can be used to eliminate differences in dimensions, which facilitates subsequent fusion calculations with other dimensional indicators.

[0079] For the embodiments of this application, the calculation formula for normalized resource consumption data can be as shown in Formula 1, wherein... This can represent the number of input tokens used by the agent to be updated to perform historical tasks. This can represent the unit price of the input tokens used by the agent to be updated to perform historical tasks. This can represent the number of output tokens from the historical tasks executed by the agent to be updated. This can represent the unit price of the output tokens of the agent to be updated when executing historical tasks. It can represent the target resource consumption data, which is the baseline cost of resource consumption set by the system.

[0080] (Formula 1)

[0081] Step 202: Perform data processing on the performance data of the agent to be updated and the task complexity data of historical tasks to obtain comprehensive performance data and comprehensive task complexity data.

[0082] In the embodiments of this application, performance data can be various performance-related metrics generated when the agent to be updated executes historical tasks. Performance data can objectively reflect the agent's response speed, tool invocation capabilities, and other execution performance characteristics. For example, the performance data in the embodiments of this application may specifically include response latency data, task response information generation time, and the success rate of the agent calling tools.

[0083] In this embodiment, task complexity data can be data characterizing the difficulty of historical tasks performed by the agent to be updated. This data can provide a basis for difficulty-based incentives in agent performance evaluation, avoiding evaluation bias caused by differences in task difficulty. For example, the task complexity data in this embodiment may specifically include the confidence level of each intention in historical tasks, the complexity of each intention, and contextual association data of historical tasks.

[0084] In this embodiment, the comprehensive performance data can be standardized performance index data obtained by weighting the performance data. The comprehensive performance data can be used to comprehensively reflect the overall execution performance level of the intelligent agent to be updated.

[0085] In the embodiments of this application, the comprehensive task complexity data can be standardized difficulty index data obtained by weighting and correcting the task complexity data in multiple dimensions. The comprehensive task complexity data can be used to accurately characterize the overall execution difficulty of historical tasks.

[0086] Step 203: The normalized resource consumption data, comprehensive performance data, comprehensive task complexity data, and evaluation data are fused and processed to obtain execution capability data representing the task execution capability of the agent to be updated.

[0087] In this embodiment, the fusion processing of normalized resource consumption data, comprehensive performance data, comprehensive task complexity data, and evaluation data may include steps such as task difficulty correction, resource consumption constraints, performance correction, and long-term data smoothing to obtain single execution capability data of the agent to be updated for executing a single historical task; combined with the agent's historical execution capability data, weighted smoothing is performed to obtain execution capability data that reflects the agent's long-term comprehensive task execution capability.

[0088] Optionally, when performing the step of "combining the performance data of the agent to be updated and the task complexity data of historical tasks to obtain comprehensive performance data and comprehensive task complexity data", the following methods can be used, but are not limited to these: weighting the response latency data, task response information generation time, and the success rate of the agent to be updated calling tools in the performance data generated by the agent to be updated executing historical tasks to obtain comprehensive performance data; weighting the confidence and complexity of each intent in historical tasks to obtain basic task complexity data; and obtaining comprehensive task complexity data based on the basic task complexity data and the contextual association data of historical tasks.

[0089] In this embodiment of the application, the formula for calculating the comprehensive performance data can be as shown in Formula 2, wherein, It can represent response latency data. It can indicate the generation time of the task response information. This can represent the success rate of the agent initiating the update calling the tool. It can represent the weight of response latency data. This can represent the weight of the generation time of the task response information. It can represent the weight of the success rate of the agent calling the tool, which is to be updated.

[0090] For the embodiments of this application, the formula for calculating the comprehensive task complexity data can be as shown in Formula 3, where... It can represent the confidence level of the i-th type of intent in a historical task. This can represent the basic complexity weight of the i-th type of intent. It can represent the context correction factor (the value can be between 0.1 and 0.3). It can represent the context window occupancy rate (i.e., the ratio of the current number of context tokens to the maximum number of context window tokens), and is used to summarize task complexity data. The value range can be 0.1-1.0.

[0091] (Formula 2)

[0092] (Formula 3)

[0093] Optionally, when performing the process of "fusion processing of normalized resource consumption data, comprehensive performance data, comprehensive task complexity data, and evaluation data to obtain execution capability data representing the task execution capability of the agent to be updated," the following methods may be used, but are not limited to: adjusting the comprehensive task complexity data based on a preset task difficulty adjustment coefficient to obtain task difficulty adjustment data; processing the evaluation data and task difficulty adjustment data to obtain evaluation difficulty fusion data; obtaining resource consumption constraint data based on normalized resource consumption data; obtaining resource constraint evaluation data based on the evaluation difficulty fusion data and resource consumption constraint data; processing the resource constraint evaluation data and comprehensive performance data to obtain the previous execution capability data of the agent to be updated for executing historical tasks; and weighting the previous execution capability data and historical execution capability data (excluding the previous execution capability data) based on a preset smoothing coefficient to obtain the execution capability data representing the task execution capability of the agent to be updated.

[0094] In this embodiment of the application, the previous execution capability data can be the execution capability data obtained from the previous task executed by the agent to be updated. The calculation formula can be shown in Formula 4, where, It can represent the weight of the rating data. It can represent the weights of emotion change data. It can represent data on changes in emotions. It can represent rating data. It can represent normalized resource consumption data. It can represent the incentive factor for task complexity. It can represent comprehensive performance data. It can represent comprehensive task complexity data. By fusing evaluation data, task complexity incentive factors, and comprehensive task complexity data, we can obtain evaluation difficulty fused data; by fusing evaluation difficulty fused data and normalized resource consumption data, we can obtain resource constraint evaluation data; by fusing resource constraint evaluation data and comprehensive performance data, we can obtain the previous execution capability data of the agent to be updated in executing historical tasks.

[0095] (Formula 4)

[0096] In this embodiment of the application, the formula for calculating the execution capability data can be as shown in Formula 5, wherein, It can represent the smoothing coefficient. This can represent the execution capability data for the t-th execution. It can represent the execution capability data in the historical execution capability data except for the t-th execution. The execution capability data of the t-th execution and the execution capability data of the (t-1)-th execution are merged to obtain the execution capability data.

[0097] (Formula 5)

[0098] Optionally, when executing the process of "configuring resources for the agent to be updated based on the target resource configuration level, determining the target resource configuration information corresponding to the agent to be updated, and using the target resource configuration information to execute the task to be executed by the agent to be updated", the following methods may be used, but are not limited to these: if the target resource configuration level is higher or lower than the current resource configuration level, adjust the resource configuration information according to the target resource configuration level to obtain the target resource configuration information; if the target resource configuration level is equal to the current resource configuration level, determine the resource configuration information corresponding to the current resource configuration level as the target resource configuration information.

[0099] In the embodiments of this application, when the target resource configuration level is higher than the current resource configuration level, that is, when the agent's level update type is an upgrade type, the resource configuration information of the agent can be upgraded and adjusted according to the preset configuration standard corresponding to the target resource configuration level. Specifically, this may include upgrading the base model specifications, expanding the context window parameters, opening more tool call permissions, increasing the QPS concurrency quota, etc. For example, when upgrading the agent from L1 intern level to L3 backbone level, the base model can be switched from a lightweight model to a medium model, the context window can be expanded from 4K to 32K, the extended toolset can be unlocked, and the QPS concurrency quota can be increased.

[0100] In the embodiments of this application, when the target resource configuration level is lower than the current resource configuration level, that is, when the level update type of the agent is a downgrade type, the resource configuration information of the agent can be downgraded according to the preset configuration standard corresponding to the target resource configuration level. Specifically, this may include downgrading the base model specifications, reducing the context window parameters, reducing tool call permissions, reducing QPS concurrency quota, etc. The downgrade adjustment can adopt a soft landing approach, first limiting the QPS concurrency quota and stripping the tool call permissions, and only adjusting the base model specifications after exhausting other means.

[0101] In this embodiment, when the target resource configuration level is equal to the current resource configuration level, i.e. the agent's level update type is the maintenance level type, there is no need to adjust the agent's resource configuration information. The resource configuration information corresponding to the current resource configuration level is directly determined as the target resource configuration information to maintain the agent's computing power resource supply stability.

[0102] In this embodiment, the adjustment of resource configuration information can be completed by the resource orchestration executor by calling the Admin API of the intelligent agent hosting platform. This can realize functions such as hot switching of the base model and elastic scaling of the context window. Furthermore, the friendliness of the ongoing session can be maintained during the adjustment process, and a smooth transition window can be set to ensure that the service is not interrupted.

[0103] Optionally, when performing the action of "determining the level update type of the agent to be updated based on execution capability data, and updating the resource configuration level of the agent to be updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated", the following methods can be used, but are not limited to these: if the execution capability data is greater than or equal to the level increase threshold, determine that the level update type of the agent to be updated is the level increase type, and upgrade the resource configuration level of the agent to be updated based on the execution capability data to obtain the first resource configuration level of the agent to be updated after the upgrade; if the execution capability data is greater than or equal to the level decrease threshold and less than the level increase threshold, determine that the level update type of the agent to be updated is the level maintenance type, and determine the resource configuration level of the agent to be updated as the second target resource configuration level based on the execution capability data; if the execution capability data is less than the level decrease threshold, determine that the level update type of the agent to be updated is the level decrease type, and downgrade the resource configuration level of the agent to be updated based on the execution capability data to obtain the third resource configuration level of the agent to be updated after the downgrade.

[0104] In this embodiment, the grade increase threshold and grade decrease threshold are preset values ​​for the intelligent agent hosting platform to configure a grade for each resource. A jitter-resistant state transition mechanism based on a hysteresis comparator can be adopted, with a preset hysteresis bandwidth between the grade increase threshold and the grade decrease threshold. The execution capability data must continuously reach the threshold conditions and meet the preset stability test cycle requirements to determine the corresponding grade update type. The number of stability test cycles can be set to 3-5 evaluation cycles according to business needs to avoid misjudgment of the grade due to fluctuations in a single execution capability data instance.

[0105] In the embodiments of this application, when the execution capability data is continuously greater than or equal to the upgrade level threshold, the level update type can be determined as the upgrade level type, and the resource configuration level of the agent is upgraded. The upgraded level is the first resource configuration level.

[0106] Optionally, a shadow test verification mechanism based on probability sampling can be triggered before upgrading the resource configuration level of the agent. For example, a shadow container with the target job level configuration is deployed for the agent to be upgraded, and a certain proportion of real-time traffic is mirrored to the shadow container. The response quality and resource consumption of the original container and the shadow container are compared to calculate the return on investment (ROI). The formal upgrade is only performed when the ROI is greater than a preset threshold and the sampling period reaches a preset time. If the agent has frequently triggered computing power lending recently and the lending profit and loss assessment is excellent, the shadow test can be waived and the upgrade can be performed directly.

[0107] In this embodiment of the application, when the execution capability data is continuously greater than or equal to the decrease level threshold and less than the increase level threshold, the level update type can be determined as the maintenance level type, and the current resource configuration level of the agent is determined as the second target resource configuration level, so as to keep the agent level unchanged.

[0108] In this embodiment, when the execution capability data is consistently below the downgrade threshold, the downgrade type can be determined as a downgrade type, and the resource configuration level of the agent is downgraded. The downgraded level is the third resource configuration level. The downgrade process can involve transferring the agent into a PIP (Progressive Processing Initiation) remediation period, employing a tiered intervention soft landing strategy for downgrading.

[0109] Optionally, when executing the statement "when the execution capability data is less than the downgrade threshold, determine that the level update type of the agent to be updated is a downgrade type, and downgrade the resource configuration level of the agent to be updated based on the execution capability data to obtain the third resource configuration level of the agent to be updated after downgrading", the following methods can be used, but are not limited to these: reducing the number of tasks and task permissions of the agent to be updated to execute tasks based on the execution capability data, and monitoring the execution status of the tasks to be executed; in response to detecting that the level update type of the agent to be updated to execute tasks is an upgrade type or a maintenance type, determining the current resource configuration level of the agent to be updated as the target resource configuration level; in response to detecting that the level update type of the agent to be updated to execute tasks is a downgrade type, downgrading the resource configuration level of the agent to be updated to obtain the third resource configuration level of the agent to be updated after downgrading.

[0110] For the embodiments of this application, a three-stage soft landing graded intervention strategy can be adopted for downgrade assessment. The first stage of the soft landing graded intervention strategy can reduce the number of tasks to be performed by the agent to be updated, for example, by 50%. The agent's performance of tasks to be performed is continuously monitored and its performance data is recalculated and the grade update type is determined. If the grade update type is an upgrade or maintenance type, the agent to be updated will not be downgraded. If the grade update type is a downgrade type, the agent's task permissions to be performed are reduced (i.e., the second stage of the soft landing graded intervention strategy). If the grade update type is detected to change to an upgrade or maintenance type, it indicates that the agent's task execution capability has been restored. Its current resource configuration level is determined as the target resource configuration level, the PIP rectification period is lifted, and the original number of tasks and task permissions are restored.

[0111] For example, if the performance data of the intelligent assistant A falls below the threshold due to changes in the API interface of the academic affairs system, after the developer repairs the interface adapter, its performance in performing tasks such as attendance query and learning analysis will return to normal, and the level update type will change to the maintenance level type. Its current L3 level can be determined as the target resource configuration level.

[0112] In this embodiment of the application, if it is detected that the execution capability data of the agent to be updated has not recovered after the task permissions are reduced, the level update type is still the downgrade type, and it is found that the tasks recently faced by the agent are generally simple, the resource configuration level of the agent can be formally downgraded (i.e., the third stage of the soft landing graded intervention strategy) to obtain the third resource configuration level.

[0113] Optionally, before executing "obtaining historical task execution data of the agent to be updated", the following methods may be used, but not limited to: obtaining rating data determined by the user based on the historical task execution results, response speed, and interactive experience of the agent to be updated; obtaining interactive text information between the user and the agent to be updated during the historical task execution process, and performing semantic parsing and sentiment analysis on the interactive text information to obtain emotion change data; and performing weighted fusion processing on the rating data and emotion change data to obtain evaluation data.

[0114] In this embodiment of the application, the rating data can be explicit feedback data from the user to the intelligent agent. The user can rate the intelligent agent based on dimensions such as the accuracy of the historical task execution results, the speed of response, and the quality of the interactive experience. The value range of the rating data can be normalized to 0-1.

[0115] In this embodiment, the emotion change data can be implicit feedback data from the user to the agent. Acquiring emotion change data involves collecting all interactive text information between the user and the agent during the execution of historical tasks. A sentiment analysis engine based on an asynchronous bypass architecture performs semantic parsing and sentiment analysis on the interactive text information. The analysis process can employ a first-and-last anchor point sliding window sampling mechanism, extracting the interactive text from the first N rounds and the last N rounds of the conversation as anchor points, where N can range from 2 to 5. A goal-oriented aspect-level sentiment analysis algorithm eliminates invalid negative emotional interference from the user regarding external objective facts, retaining only the sentiment scores pointing to service-related entities such as the agent, responses, and services, thus obtaining the average sentiment valence of the initial anchor point, the average sentiment valence of the final anchor point, and the sentiment improvement value. The emotion change data is obtained by comprehensively calculating the average sentiment valence of the initial anchor point, the average sentiment valence of the final anchor point, and the sentiment improvement value. The calculation formula can be shown in Formula Six, where, It can represent the weighting coefficient. It can represent the mood improvement value. It can be calculated from the average emotional valence of the initial anchor point and the average emotional valence of the final anchor point. It can represent the average emotional valence of the termination anchor point.

[0116] (Formula 6)

[0117] Optionally, the fusion weight for weighted fusion of rating data and emotion change data can be dynamically adjusted based on the existence of rating data. When there are valid explicit user ratings, the weight of the rating data can be increased. When there are no explicit user ratings, emotion change data can be used as the main evaluation basis. The evaluation data obtained in this way can improve the accuracy of the task execution ability of the agent to be updated.

[0118] As an optional approach, this application embodiment also provides an example of intelligent agent computing power lending, the specific steps of which may include:

[0119] Step 1: After the agent to be updated receives a task initiated by the user, the system can first use the global interaction perception and feature extraction module to extract and analyze the features of the task to be executed, and calculate the actual task complexity data. The calculation method of this data can be consistent with the task complexity data of historical tasks, and quantification is achieved through analysis of dimensions such as task intent and contextual association data. At the same time, the system can also retrieve the task complexity carrying threshold corresponding to the current resource configuration level of the agent to be updated. The task complexity carrying threshold can represent the upper limit of task complexity that an agent of the corresponding level can efficiently handle.

[0120] Step 2: The system can compare the actual task complexity data of the task to be executed with the current task complexity capacity threshold of the agent to be updated. If the actual task complexity data does not exceed the capacity threshold, it indicates that the current computing resources of the agent to be updated are sufficient to handle the task to be executed, and it can be processed by the computing resources currently configured by the agent according to the normal routing strategy. If the actual task complexity data is greater than the capacity threshold, the system can further detect the current state of the agent to be updated. If the agent is in a frozen state, it indicates that it has been inactive for a long time or has been continuously substandard in the archived state, and the system will directly refuse to allocate additional computing resources to it. If the agent is in a non-frozen state, including normal L1-L5 job level states or PIP state during the rectification period, it can be determined that the agent to be updated meets the triggering conditions for computing power lending and start the computing power lending process.

[0121] Step 3: Based on the actual task complexity data of the task to be executed, the system can determine the target lending resource configuration level that matches the task processing requirements. The target lending resource configuration level can be the resource configuration level corresponding to the computing power resources that the agent to be updated needs to temporarily borrow. The target lending resource configuration level can be flexibly matched according to the task complexity. For example, when an L1-level agent receives a high-difficulty task with a task complexity close to 1.0, the L1-level agent can be matched to the L5 expert-level target lending resource configuration level.

[0122] Step 4: The system can issue a unique computing power lending token for this computing power lending. This token is a one-time privileged token and may contain information such as the token's unique identifier, the ID of the agent to be updated, the agent's original resource configuration level, the matching target lending resource configuration level, the resource amount of this lending, the maximum allowed token consumption, the maximum duration, the list of temporarily open tool call permissions, the token issuance time and expiration time, and can also configure a traceability ID to associate with the original request of this task to be executed, so as to realize the full-process traceability of the lending behavior.

[0123] Step 5: After the computing power lending token is issued, the system can inject the token into the request header of the task to be executed. At the same time, the resource orchestration executor modifies the computing power routing strategy to route the task request carrying the computing power lending token to the higher-level computing power resource pool corresponding to the target lending resource configuration level. The higher-level computing power resource pool will provide computing power support for the task to be executed, and the tool call permissions configured in the token will be temporarily opened to ensure that the agent to be updated can efficiently complete the highly complex task to be executed by relying on the lending computing power resources.

[0124] Step 6: After the agent to be updated completes the task to be executed and returns the response result to the user by relying on the computing power resources borrowed, the system can trigger the verification process of the computing power lending token, extract the token information of this loan from the request metadata and make it invalid, and at the same time calculate the actual resource consumption cost incurred in this computing power lending process, including the actual token consumption, computing power occupation time, etc., and record the cost information in the technical debt ledger of the agent to be updated.

[0125] Step 7: After completing token verification and cost statistics, the system can evaluate the profit and loss of this computing power loan and calculate the return on investment (ROI). Specifically, it can be obtained by dividing the single execution capability data of the task to be executed by the ratio of the loan cost to the baseline cost. If the calculated ROI is lower than the preset qualified threshold, the computing power loan can be identified as inefficient, and a penalty coefficient can be added to the execution capability data of the agent to be updated to deduct it. If the ROI is higher than the preset excellent threshold, the computing power loan can be identified as efficient, and an incentive coefficient can be added to the execution capability data. If the ROI is between the qualified threshold and the excellent threshold, the execution capability data remains unchanged. In this way, the computing power loan behavior of the agent can be quantitatively evaluated and rewarded or punished.

[0126] Step 8: The system can continuously monitor the computing power lending status of each agent to be updated within a preset rolling time window, including the lending frequency and cumulative lending cost within that time window. The rolling time window can be set to 24 hours, and a maximum lending frequency and a maximum cumulative lending cost are preset for each agent. If the lending frequency of an agent within the window does not exceed the maximum, subsequent computing power lending can be triggered normally. If the lending frequency of an agent within the window reaches or exceeds the maximum, subsequent lending requests can be rejected. If it is detected that the computing power lending frequency of an agent to be updated exceeds the preset threshold within the rolling time window, and the profit and loss assessment results of multiple computing power lendings are all excellent, it can be indicated that the actual business needs of the agent have exceeded its current resource allocation level. The system can automatically trigger the promotion assessment signal for the agent, transforming the temporary computing power lending needs into a permanent resource allocation quota assessment basis.

[0127] Compared with existing technologies, this embodiment normalizes the resource consumption data of the agent to be updated, processes performance data and task complexity data separately to obtain comprehensive performance data and comprehensive task complexity data, and fuses multi-dimensional data to obtain execution capability data, thereby achieving a multi-dimensional comprehensive evaluation of the agent's task execution capability and improving the accuracy of the execution capability data; by adjusting or determining the target resource configuration information differently based on the comparison results between the target resource configuration level and the current level, the execution capability level and resource configuration of the agent are updated, improving the adaptability of the agent's execution capability and resource configuration; by reducing the number of tasks and permissions of agents whose execution capability data does not meet the standards and monitoring the execution status, and then determining whether to downgrade based on the monitoring results, the resource configuration level of the agent is gradually downgraded; by weightedly fusing user rating data and emotion change data obtained based on interactive text to obtain evaluation data, the completeness and accuracy of the evaluation data are improved.

[0128] Furthermore, as Figure 1 and Figure 2 To specifically implement the method shown, this embodiment provides a resource configuration update device, such as... Figure 3 As shown, the device includes: an acquisition module 31, an evaluation module 32, an update module 33, and a determination module 34.

[0129] The acquisition module 31 is configured to acquire historical task execution data of the agent to be updated. The historical task execution data includes agent operation data of the agent to be updated executing historical tasks, as well as evaluation data of the user who initiated the historical task on the historical task execution of the agent to be updated.

[0130] Evaluation module 32 is configured to comprehensively evaluate the historical task execution status of the agent to be updated based on historical task execution data, and obtain execution capability data representing the task execution capability of the agent to be updated;

[0131] The update module 33 is configured to determine the level update type of the agent to be updated based on the execution capability data, and update the resource configuration level of the agent to be updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated.

[0132] The determination module 34 is configured to configure resources for the agent to be updated based on the target resource configuration level, and determine the target resource configuration information corresponding to the agent to be updated. The target resource configuration information is used by the agent to be updated to execute the task to be executed.

[0133] In some examples of this embodiment, the evaluation module 32 is specifically configured to: normalize the resource consumption data based on the target resource consumption data of the agent to be updated to obtain normalized resource consumption data of the agent to be updated; process the performance data of the agent to be updated and the task complexity data of historical tasks to obtain comprehensive performance data and comprehensive task complexity data; and fuse the normalized resource consumption data, comprehensive performance data, comprehensive task complexity data and evaluation data to obtain execution capability data representing the task execution capability of the agent to be updated.

[0134] In some examples of this embodiment, the evaluation module 32 is further configured to weight the response latency data, the generation time of task response information, and the success rate of the agent calling the tool in the performance data generated by the agent to be updated executing historical tasks to obtain comprehensive performance data; to weight the confidence and complexity of each intent in the historical tasks to obtain basic task complexity data; and to obtain comprehensive task complexity data based on the basic task complexity data and the contextual association data of the historical tasks.

[0135] In some examples of this embodiment, the evaluation module 32 is further configured to: correct the comprehensive task complexity data based on a preset task difficulty correction coefficient to obtain task difficulty correction data; process the evaluation data and the task difficulty correction data to obtain evaluation difficulty fusion data; obtain resource consumption constraint data based on normalized resource consumption data; obtain resource constraint evaluation data based on the evaluation difficulty fusion data and resource consumption constraint data; process the resource constraint evaluation data and comprehensive performance data to obtain the previous execution capability data of the agent to be updated for executing historical tasks; and perform weighted processing on the previous execution capability data and the historical execution capability data other than the previous execution capability data based on a preset smoothing coefficient to obtain execution capability data representing the task execution capability of the agent to be updated.

[0136] In some examples of this embodiment, the determining module 34 is specifically configured to adjust the resource configuration information according to the target resource configuration level when the target resource configuration level is higher or lower than the current resource configuration level, so as to obtain the target resource configuration information; and to determine the resource configuration information corresponding to the current resource configuration level as the target resource configuration information when the target resource configuration level is equal to the current resource configuration level.

[0137] In some examples of this embodiment, the update module 33 is specifically configured to: when the execution capability data is greater than or equal to the upgrade level threshold, determine that the level update type of the agent to be updated is the upgrade level type, and upgrade the resource configuration level of the agent to be updated based on the execution capability data to obtain the first resource configuration level of the agent to be updated after the upgrade; when the execution capability data is greater than or equal to the downgrade level threshold and less than the upgrade level threshold, determine that the level update type of the agent to be updated is the maintenance level type, and determine the resource configuration level of the agent to be updated as the second target resource configuration level based on the execution capability data; when the execution capability data is less than the downgrade level threshold, determine that the level update type of the agent to be updated is the downgrade level type, and downgrade the resource configuration level of the agent to be updated based on the execution capability data to obtain the third resource configuration level of the agent to be updated after the downgrade.

[0138] In some examples of this embodiment, the update module 33 is further configured to reduce the number of tasks and task permissions of the agent to be updated to perform tasks based on the execution capability data, and monitor the execution status of the tasks to be performed; in response to detecting that the level update type of the agent to be updated performing tasks is an upgrade level type or a maintenance level type, the current resource configuration level of the agent to be updated is determined as the target resource configuration level; in response to detecting that the level update type of the agent to be updated performing tasks is a downgrade type, the resource configuration level of the agent to be updated is downgraded to obtain the third resource configuration level of the agent to be updated after downgrading.

[0139] In some examples of this embodiment, the acquisition module 31 is specifically configured to acquire rating data determined by the user based on the historical task execution results, response speed, and interactive experience of the intelligent agent to be updated; acquire interactive text information between the user and the intelligent agent to be updated during the historical task execution process, and perform semantic parsing and sentiment analysis on the interactive text information to obtain emotion change data; and perform weighted fusion processing on the rating data and emotion change data to obtain evaluation data.

[0140] It should be noted that other corresponding descriptions of the functional units involved in the resource configuration update device provided in this embodiment can be found in [reference]. Figure 1 and Figure 2The corresponding descriptions in [the document] will not be repeated here.

[0141] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method shown.

[0142] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0143] like Figure 4 The diagram shown is a hardware structure schematic of an electronic device according to the present invention, comprising:

[0144] At least one processor 401; and,

[0145] Memory 402 is communicatively connected to at least one processor 401; wherein,

[0146] The memory 402 stores instructions that can be executed by at least one processor to enable the at least one processor to perform the resource configuration update method as described above.

[0147] Figure 4 Take a processor 401 as an example.

[0148] The electronic device may also include an input device 403 and an output device 404.

[0149] The processor 401, memory 402, input device 403, and output device 404 can be connected via a bus or other means. Figure 4 Taking the example of a connection between China and Israel via a bus.

[0150] Memory 402, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the resource configuration update method in the embodiments of this application, for example, Figure 1 and Figure 2 The method flow is shown. The processor 401 executes various functional applications and data processing by running non-volatile software programs, instructions, and modules stored in the memory 402, thereby implementing the resource configuration update method in the above embodiments.

[0151] Memory 402 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the use of the resource configuration update method. Furthermore, memory 402 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 402 may optionally include memory remotely located relative to processor 401, and these remote memories may be connected via a network to the apparatus performing the resource configuration update method. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0152] Input device 403 can receive user clicks and generate signal inputs related to user settings and function control for resource configuration update methods. Output device 404 may include display devices such as a display screen.

[0153] One or more modules are stored in memory 402, and when run by one or more processors 401, the resource configuration update method in any of the above method embodiments is executed.

[0154] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0155] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0156] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0157] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or it can be implemented by hardware. By applying the solution of this embodiment, compared with the prior art, this embodiment obtains execution capability data by comprehensively evaluating the historical task execution status of the intelligent agent based on historical task execution data, providing a precise basis for the level update of the intelligent agent to be updated; by determining the level update type based on the execution capability data and updating the resource configuration level to obtain the target resource configuration level, the intelligent agent's resource configuration is configured according to the target resource configuration level, realizing the adaptation between the intelligent agent's resource configuration and the target resource configuration level, so that the resource configuration fits the actual needs of the intelligent agent in executing the tasks to be executed, realizing the reasonable allocation of computing resources, and ensuring the intelligent agent's business processing capabilities; by normalizing the resource consumption data of the intelligent agent to be updated, the comprehensive performance data is obtained by processing the performance data and task complexity data separately. Based on comprehensive task complexity data, multi-dimensional data is integrated to obtain execution capability data, enabling a multi-dimensional comprehensive evaluation of the agent's task execution capability and improving the accuracy of the execution capability data. By adjusting or determining the target resource configuration information based on the comparison results between the target resource configuration level and the current level, the execution capability level and resource configuration of the agent are updated, improving the adaptability of the agent's execution capability and resource configuration. For agents whose execution capability data does not meet the standards, the number of tasks and permissions are first reduced and the execution status is monitored, and then a decision is made on whether to downgrade based on the monitoring results, thus achieving a gradual downgrade of the agent's resource configuration level. By weightedly integrating user rating data and emotion change data obtained based on interactive text, evaluation data is obtained, improving the completeness and accuracy of the evaluation data.

[0158] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0159] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A resource allocation update method, characterized in that, include: Obtain historical task execution data of the agent to be updated. The historical task execution data includes agent operation data of the agent to be updated executing historical tasks, as well as evaluation data of the user who initiated the historical task on the historical task execution status of the agent to be updated. Based on the historical task execution data, a comprehensive evaluation of the historical task execution status of the agent to be updated is performed to obtain execution capability data representing the task execution capability of the agent to be updated; Based on the target resource consumption data of the agent to be updated, the resource consumption data of the agent to be updated is normalized to obtain the normalized resource consumption data of the agent to be updated. The performance data of the agent to be updated and the task complexity data of the historical tasks are processed separately to obtain comprehensive performance data and comprehensive task complexity data. The comprehensive performance data is obtained by weighting the response latency data, the generation time of task reply information, and the success rate of the tool call by the agent to be updated in the performance data generated by the agent to be updated executing historical tasks. The confidence and complexity of each intent in the historical tasks are weighted to obtain the basic task complexity data; Based on the basic task complexity data and the contextual association data of the historical tasks, the comprehensive task complexity data is obtained. The normalized resource consumption data, comprehensive efficiency data, comprehensive task complexity data, and evaluation data are fused together to obtain execution capability data representing the task execution capability of the agent to be updated. Based on a preset task difficulty correction coefficient, the comprehensive task complexity data is corrected to obtain task difficulty correction data. The evaluation data and the task difficulty correction data are then processed to obtain evaluation difficulty fusion data. Based on the normalized resource consumption data, resource consumption constraint data is obtained; based on the evaluation difficulty fusion data and the resource consumption constraint data, resource constraint evaluation data is obtained. The resource constraint evaluation data and the comprehensive performance data are processed to obtain the previous execution capability data of the intelligent agent to be updated in executing the historical task; Based on a preset smoothing coefficient, the execution capability data other than the previous execution capability data in the previous execution capability data and the historical execution capability data are fused to obtain execution capability data representing the task execution capability of the agent to be updated. Based on the execution capability data, the level update type of the agent to be updated is determined, and the resource configuration level of the agent to be updated is updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated. Based on the target resource configuration level, the agent to be updated is configured with resources to determine the target resource configuration information corresponding to the agent to be updated. The target resource configuration information is used by the agent to be updated to execute the task to be executed.

2. The method according to claim 1, characterized in that, The step of configuring resources for the agent to be updated based on the target resource configuration level, and determining the target resource configuration information corresponding to the agent to be updated, includes: If the target resource configuration level is higher or lower than the current resource configuration level, the resource configuration information is adjusted according to the target resource configuration level to obtain the target resource configuration information; If the target resource configuration level is equal to the current resource configuration level, the resource configuration information corresponding to the current resource configuration level shall be determined as the target resource configuration information.

3. The method according to claim 1, characterized in that, The step of determining the level update type of the agent to be updated based on the execution capability data, and updating the resource configuration level of the agent to be updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated, includes: If the execution capability data is greater than or equal to the upgrade level threshold, the level update type of the agent to be updated is determined to be the upgrade level type, and the resource configuration level of the agent to be updated is upgraded based on the execution capability data to obtain the first resource configuration level of the agent to be updated after the upgrade. If the execution capability data is greater than or equal to the reduction level threshold and less than the increase level threshold, the level update type of the agent to be updated is determined to be the maintenance level type, and the resource configuration level of the agent to be updated is determined to be the second target resource configuration level based on the execution capability data. If the execution capability data is less than the downgrade threshold, the level update type of the agent to be updated is determined to be a downgrade type, and the resource configuration level of the agent to be updated is downgraded based on the execution capability data to obtain the third resource configuration level of the agent to be updated after downgrading.

4. The method according to claim 3, characterized in that, When the execution capability data is less than the downgrade threshold, the step of determining that the level update type of the agent to be updated is a downgrade type, and downgrading the resource configuration level of the agent to be updated based on the execution capability data to obtain the third resource configuration level of the agent to be updated after downgrading, includes: Based on the execution capability data, the number of tasks and task permissions for the agent to be updated to execute the tasks to be executed are reduced, and the execution status of the tasks to be executed is monitored; In response to the detection that the level update type of the task to be performed by the agent to be updated is either the level upgrade type or the level maintenance type, the current resource configuration level of the agent to be updated is determined as the target resource configuration level; In response to the detection that the level update type of the task to be performed by the agent to be updated is a downgrade type, the resource configuration level of the agent to be updated is downgraded to obtain the third resource configuration level of the agent to be updated.

5. The method according to claim 1, characterized in that, Before obtaining the historical task execution data of the agent to be updated, the method further includes: Obtain the rating data determined by the user based on the historical task execution results, response speed, and interactive experience of the intelligent agent to be updated; The interaction text information between the user and the agent to be updated during the execution of the historical task is obtained, and the interaction text information is subjected to semantic parsing and sentiment analysis to obtain sentiment change data. The rating data and the emotion change data are weighted and fused to obtain the evaluation data.

6. A resource allocation and updating device, characterized in that, include: The acquisition module is configured to acquire historical task execution data of the agent to be updated. The historical task execution data includes agent operation data of the agent to be updated executing historical tasks, and evaluation data of the user who initiated the historical task on the historical task execution status of the agent to be updated. The evaluation module is configured to comprehensively evaluate the historical task execution status of the agent to be updated based on the historical task execution data, and obtain execution capability data representing the task execution capability of the agent to be updated; Based on the target resource consumption data of the agent to be updated, the resource consumption data of the agent to be updated is normalized to obtain the normalized resource consumption data of the agent to be updated. The performance data of the agent to be updated and the task complexity data of the historical tasks are processed separately to obtain comprehensive performance data and comprehensive task complexity data. The comprehensive performance data is obtained by weighting the response latency data, the generation time of task reply information, and the success rate of the tool call by the agent to be updated in the performance data generated by the agent to be updated executing historical tasks. The confidence and complexity of each intent in the historical tasks are weighted to obtain the basic task complexity data; Based on the basic task complexity data and the contextual association data of the historical tasks, the comprehensive task complexity data is obtained. The normalized resource consumption data, comprehensive efficiency data, comprehensive task complexity data, and evaluation data are fused together to obtain execution capability data representing the task execution capability of the agent to be updated. Based on a preset task difficulty correction coefficient, the comprehensive task complexity data is corrected to obtain task difficulty correction data. The evaluation data and the task difficulty correction data are then processed to obtain evaluation difficulty fusion data. Based on the normalized resource consumption data, resource consumption constraint data is obtained; based on the evaluation difficulty fusion data and the resource consumption constraint data, resource constraint evaluation data is obtained. The resource constraint evaluation data and the comprehensive performance data are processed to obtain the previous execution capability data of the intelligent agent to be updated in executing the historical task; Based on a preset smoothing coefficient, the execution capability data other than the previous execution capability data in the previous execution capability data and the historical execution capability data are fused to obtain execution capability data representing the task execution capability of the agent to be updated. The update module is configured to determine the level update type of the agent to be updated based on the execution capability data, and update the resource configuration level of the agent to be updated according to the level update type to obtain the target resource configuration level corresponding to the agent to be updated. The determination module is configured to configure resources for the agent to be updated based on the target resource configuration level, and determine the target resource configuration information corresponding to the agent to be updated. The target resource configuration information is used by the agent to be updated to execute the task to be executed.

7. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 5.

Images