A large model management platform based on artificial intelligence

Abstract

The present invention discloses a large model management platform based on artificial intelligence, which relates to the technical field of large model management. The platform discloses a model service module, a model management module, and a service adjustment module. By setting the model service module and the model management module, the management cycle of the large model can be dynamically adjusted by analyzing the model service records of each large model to ensure accurate and appropriate management of each large model. By setting the service adjustment module, the load large model can be analyzed in time, and the users in the load large model can be deeply analyzed to determine the deployment users existing in the load large model, and then some of the deployment users can be allocated to the appropriate large model for subsequent data processing, which can not only avoid the situation where the data processing load of the large model is too high, but also optimize the balance of data processing of each large model.

Description

Technical Field

[0001] The present invention relates to the field of large model management technology, and more specifically, to a large model management platform based on artificial intelligence. Background Art

[0002] Artificial intelligence big models refer to machine learning models with super-large-scale parameters (usually more than one billion) and super-powerful computing resources, which can process massive amounts of data and complete various complex tasks. Different types of big models can process different types of data, such as language big models for processing language-type data, and visual big models for processing visual-type data. Currently, users send the data to be processed to commonly used big models for data processing, which will cause some big models to be overloaded. How to reasonably manage big models and avoid overloading some big models is an urgent problem that needs to be solved. Summary of the invention

[0003] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a large model management platform based on artificial intelligence.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A large model management platform based on artificial intelligence, including a model service module, a model management module, and a service adjustment module;

[0006] The model service module is used to obtain the model service record of the large model;

[0007] The model management module is used to dynamically obtain the model service value of the large model;

[0008] The service adjustment module is used to obtain the model load value of the large model in real time, and based on the comparison result between the model load value and its threshold, determine whether to mark the large model as a large-load model. When the large model is marked as a large-load model, analyze the deployment users of the large-load model and determine whether to allocate the subsequent data of the deployment users to the remaining large models.

[0009] Furthermore, the model service record of the large model is obtained. Specifically, when the user sends data to the large model, the large model processes the data sent by the user and generates the model service record of the large model.

[0010] Furthermore, the model service record includes the user ID to which the data belongs, the data type, the model service time, and the data volume.

[0011] Furthermore, the model service value of the large model is dynamically obtained, specifically: based on the P period interval, at each periodic node, the model service value of the large model is obtained, and a node log is synchronously generated, the node log including the model service value and the node time;

[0012] Set the model service threshold. When the model service value of the large model is less than the model service threshold, calculate the ratio of the model service value to the model service threshold to obtain the adjustment index tz. At the cycle interval, when the next cycle node arrives, the model service value of the large model is obtained.

[0013] Furthermore, the model service value is obtained in the following manner: The model service value of the large model is obtained in the following manner: obtain all model service records of the large model in the current period, mark the total number of model service records as TSG, obtain the data volume of the model service record, set a data volume threshold, and when the data volume of the model service record is greater than or equal to the data volume threshold, mark the model service record as an excess service record, sort all excess service records in the order of model service time, calculate the difference between the model service times of two adjacent excess service records after sorting, and obtain the excess service time difference, sum and average all excess service time differences, obtain the average excess service time difference and mark it as EZC, and use the formula The model service value RGB of the large model is obtained, where a1 is the total number coefficient of model service records and a2 is the average excess service time difference coefficient.

[0014] Furthermore, the model load value of the large model is obtained in real time, and a model load threshold is set. When the model load value of the large model is greater than or equal to the model load threshold, the large model is marked as a large-load model.

[0015] Furthermore, the model load value of the large model is obtained in the following way: obtain all node logs of the large model within H time length before the current time of the system, sort all node logs in the order of node time, calculate the time difference between the node times of two adjacent node logs after sorting, obtain the node time difference, sum up all node time differences and take the average value to obtain the average node time difference, and mark it as PKC, sum up the model service values ​​of all node logs and take the average value to obtain the average model service value, and mark it as SZY, and use the formula The model load value FSE of the large model is obtained, where a1 is the average node time difference coefficient and a2 is the average model service value coefficient.

[0016] Furthermore, when the large model is marked as a large load model, the deployment users of the large load model are analyzed, specifically: when the large model is marked as a large load model, all model service records of the large load model within H time periods before the current system time are obtained, the user id to which the data of the model service record belongs is obtained, the model service records of the same user id are marked as same-household service records, the user rating of each user is obtained, the user threshold rating is set, and when the user's user rating is greater than or equal to the user threshold rating, the user is marked as a deployment user.

[0017] Furthermore, the user evaluation of the user is obtained in the following way:

[0018] All the same-household service records are sorted in the order of the model service time, and the time difference between the model service times of two adjacent same-household service records after sorting is calculated to obtain the same-household service interval. All the same-household service intervals are summed and averaged to obtain the average same-household service interval, which is marked as ZSK. The data volume of two adjacent same-household service records after sorting is calculated by difference and the absolute value is taken to obtain the data fluctuation amount. The data fluctuation amount threshold is set. When the data fluctuation amount is greater than or equal to the data fluctuation amount threshold, the number of data fluctuations is increased by one. The number of data fluctuations is summed to obtain the total number of data fluctuations, which is marked as TLS. The total number of same-household service records is marked as RTC. The formula is used The user evaluation CAE of the user is obtained, where c1 is the average same-household service interval, c2 is the total number of data fluctuation coefficients, and c3 is the total number of same-household service records coefficient.

[0019] Further, it is determined whether to allocate subsequent data of the deployed user to the remaining large models, specifically: taking the device location of the deployed user as the center and drawing a circle with a preset radius to obtain the service deployment range, marking the remaining large models of the same type as the load large model within the service deployment range as matching large models, obtaining the matching value of the matching large model, setting the matching threshold, and when the matching value of the matching large model is greater than or equal to the matching threshold, marking the matching large model as the target large model, marking the target large model with the largest matching value as the allocation large model, and directly allocating the subsequent data of the deployed user to the allocation large model, and at the same time, the allocation large model generates a user allocation record, and the user allocation record includes the user allocation time;

[0020] When the target large model does not exist within the service deployment range, the subsequent data of the deployment user will continue to be allocated to the load large model;

[0021] The optional value of the optional large model is obtained by the following method: obtain the model load value FSE of the optional large model, obtain all user allocation records of the optional large model within H time before the current system time, mark the total number of user allocation records as KGB, and use the formula The matching value XPZ of the matching large model is obtained, where d1 is the model load value coefficient and d2 is the total number coefficient of user allocation records.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. Set up the model service module and the model management module. By analyzing the model service records of each large model, the management cycle of the large model can be dynamically adjusted to ensure accurate and appropriate management of each large model;

[0024] 2. Setting up a service adjustment module can timely analyze the large load model, conduct in-depth analysis of the users in the large load model, determine the deployment users in the large load model, and then allocate some of the deployment users to the appropriate large model for subsequent data processing. This can not only avoid the situation where the data processing load of the large model is too high, but also optimize the balance of data processing of each large model. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flowchart for dynamically obtaining the model service value of a large model;

[0026] Figure 2 Schematic diagram of the service adaptation module. DETAILED DESCRIPTION

[0027] Reference Figure 1-Figure 2 , a large model management platform based on artificial intelligence, including a model service module, a model management module, and a service adjustment module.

[0028] Model service module: When a user sends data to the big model, the big model processes the data sent by the user and generates a model service record for the big model. The model service record includes the user ID to which the data belongs, the data type (different types of big models process different types of data, such as language big models for processing language data, and visual big models for processing visual data), model service time (the time when the big model processes the data sent by the user), and the data volume.

[0029] Model management module: Based on the P periodic interval, at each periodic node, the model service value of the large model is obtained, and the node log is generated synchronously. The node log includes the model service value and the node time (the time corresponding to the periodic node).

[0030] The model service value of the large model is obtained in the following way: obtain all model service records of the large model in the current cycle, mark the total number of model service records as TSG, obtain the data volume of the model service record, set the data volume threshold, the data volume threshold is the system setting threshold, and can be modified according to actual needs. When the data volume of the model service record is greater than or equal to the data volume threshold, the model service record is marked as an excess service record. When the data volume of the model service record is less than the data volume threshold, no corresponding processing is performed. All excess service records are sorted in the order of model service time, and the model service time difference of two adjacent excess service records after sorting is calculated to obtain the excess service time difference. All excess service time differences are summed and averaged to obtain the average excess service time difference and marked as EZC. The formula is used The model service value RGB of the large model is obtained, where a1 is the total number coefficient of model service records, a2 is the average excess service time difference coefficient, the value of a1 is 0.52, and the value of a2 is 0.83.

[0031] Set the model service threshold. The model service threshold is the system setting threshold and can be modified according to actual needs.

[0032] When the model service value of the large model is greater than or equal to the model service threshold, no corresponding processing is performed.

[0033] When the model service value of the large model is less than the model service threshold, the model service value is calculated by comparing it with the model service threshold to obtain the adjustment index tz. At the cycle interval, when the next cycle node arrives, the model service value of the large model is obtained.

[0034] By setting up the model service module and the model management module, we can dynamically adjust the management cycle of large models through the analysis of the model service records of each large model, and ensure accurate and appropriate management of each large model.

[0035] Service adjustment module: obtains the model load value of the large model in real time and sets the model load threshold. The model load threshold is the system setting threshold and can be modified according to actual needs.

[0036] When the model load value of the large model is greater than or equal to the model load threshold, the large model is marked as a large load model.

[0037] When the model load value of the large model is smaller than the model load value, no corresponding processing is performed.

[0038] The model load value of the large model is obtained in the following way: obtain all node logs of the large model within H time before the current time of the system, sort all node logs in the order of node time, calculate the time difference between the node times of two adjacent node logs after sorting, obtain the node time difference, sum up all node time differences and take the average value to obtain the average node time difference, and mark it as PKC, sum up the model service values ​​of all node logs and take the average value to obtain the average model service value, and mark it as SZY, and use the formula The model load value FSE of the large model is obtained, where a1 is the average node time difference coefficient, a2 is the average model service value coefficient, the value of a1 is 0.74, and the value of a2 is 0.81.

[0039] When the large model is marked as a large load model, obtain all model service records of the large load model within H time periods before the current system time, obtain the user id to which the data of the model service record belongs, mark the model service records of the same user id as same-household service records, obtain the user rating of each user, set the user threshold rating, and the user threshold rating is a system-set threshold that can be modified according to actual needs. When the user's user rating is greater than or equal to the user threshold rating, the user is marked as a deployment user.

[0040] When the user rating of a user is less than the user threshold rating, no corresponding processing is performed.

[0041] The user rating of the user is obtained in the following way:

[0042] All the same-household service records are sorted in the order of the model service time, and the model service time of the two adjacent same-household service records after sorting is calculated to obtain the same-household service interval. All the same-household service intervals are summed up and the average is taken to obtain the average same-household service interval, which is marked as ZSK. The data volume of the two adjacent same-household service records after sorting is calculated and the absolute value is taken to obtain the data fluctuation amount. The data fluctuation amount threshold is set. The data fluctuation amount threshold is the system setting threshold and can be modified according to actual needs. When the data fluctuation amount is greater than or equal to the data fluctuation amount threshold, the number of data fluctuations is increased once. When the data fluctuation amount is less than the data fluctuation amount threshold, no corresponding processing is performed. The number of data fluctuations is summed up to obtain the total number of data fluctuations, which is marked as TLS. The total number of same-household service records is marked as RTC. Using the formula The user evaluation CAE of the user is obtained, where c1 is the average service interval between the same household, c2 is the total number coefficient of data fluctuations, and c3 is the total number coefficient of the same household service records. The value of c1 is 0.71, the value of c2 is 0.52, and the value of c3 is 0.69.

[0043] With the device location of the deployment user as the center, draw a circle with a preset radius (the preset radius can be modified according to actual needs) to obtain the service deployment range, mark the remaining large models (large model server locations) of the same type as the load large model within the service deployment range as optional large models, obtain the matching value of the optional large model, set the matching threshold, which is a system-set threshold and can be modified according to actual needs. When the matching value of the optional large model is greater than or equal to the matching threshold, mark the optional large model as the target large model. When the matching value of the optional large model is less than the matching threshold, no corresponding processing is performed, and the target large model with the largest matching value is marked as the allocation large model, and the subsequent data of the deployment user is directly allocated to the allocation large model. At the same time, the allocation large model generates a user allocation record, which includes the user allocation time (i.e., the time when the deployment user is allocated to the allocation large model).

[0044] When the target large model does not exist within the service deployment scope, the subsequent data of the deployment user will continue to be allocated to the load large model.

[0045] The optional value of the optional large model is obtained by the following method: obtain the model load value FSE of the optional large model, obtain all user allocation records of the optional large model within H time before the current system time, mark the total number of user allocation records as KGB, and use the formula The matching value XPZ of the matching large model is obtained, where d1 is the model load value coefficient, d2 is the total number coefficient of user allocation records, the value of d1 is 1.32, and the value of d2 is 1.56.

[0046] By setting up a service adjustment module, we can analyze the large load model in a timely manner and conduct in-depth analysis of the users in the large load model to determine the deployment users that exist in the large load model, and then allocate some of the deployment users to the appropriate large model for subsequent data processing. This can not only avoid the situation where the data processing load of the large model is too high, but also optimize the balance of data processing for each large model.

[0047] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0048] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0049] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0050] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0051] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0052] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0053] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0054] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A large model management platform based on artificial intelligence, characterized in that: Including model service module, model management module, and service adjustment module; The model service module is used to obtain the model service record of the large model; The model management module is used to dynamically obtain the model service value of the large model; The service adjustment module is used to obtain the model load value of the large model in real time, and based on the comparison result of the model load value and its threshold, determine whether to mark the large model as a large-load model. When the large model is marked as a large-load model, analyze the deployment user of the large-load model and determine whether to allocate the subsequent data of the deployment user to the remaining large models. Model service records include the user ID to which the data belongs, data type, model service time, and data volume; The model service value is obtained in the following way: The model service value of the large model is obtained in the following way: obtain all model service records of the large model in the current period, mark the total number of model service records as TSG, obtain the data volume of the model service record, set the data volume threshold, when the data volume of the model service record is greater than or equal to the data volume threshold, mark the model service record as an excess service record, sort all excess service records in the order of model service time, calculate the difference between the model service times of two adjacent excess service records after sorting, and obtain the excess service time difference, sum up all excess service time differences and take the average, obtain the average excess service time difference and mark it as EZC, and use the formula Get the model service value RGB of the large model, where a1 is the total number coefficient of model service records, and a2 is the average excess service time difference coefficient; The model load value of the large model is obtained in the following way: obtain all node logs of the large model within H time before the current time of the system, sort all node logs in the order of node time, calculate the time difference between the node times of two adjacent node logs after sorting, obtain the node time difference, sum up all node time differences and take the average value to obtain the average node time difference, and mark it as PKC, sum up the model service values ​​of all node logs and take the average value to obtain the average model service value, and mark it as SZY, and use the formula The model load value FSE of the large model is obtained, where b1 is the average node time difference coefficient and b2 is the average model service value coefficient.

2. According to the large model management platform based on artificial intelligence in claim 1, it is characterized in that: The model service record of the large model is obtained, specifically: when the user sends data to the large model, the large model processes the data sent by the user and generates the model service record of the large model.

3. The large model management platform based on artificial intelligence according to claim 1 is characterized in that: Dynamically obtain the model service value of the large model, specifically: based on the P period interval, at each periodic node, obtain the model service value of the large model, and synchronously generate a node log, which includes the model service value and node time; Set the model service threshold. When the model service value of the large model is less than the model service threshold, calculate the ratio of the model service value to the model service threshold to obtain the adjustment index tz. At the cycle interval, when the next cycle node arrives, the model service value of the large model is obtained.

4. The large model management platform based on artificial intelligence according to claim 1 is characterized in that: The model load value of the large model is obtained in real time, and a model load threshold is set. When the model load value of the large model is greater than or equal to the model load threshold, the large model is marked as a large-load model.

5. The large model management platform based on artificial intelligence according to claim 1 is characterized in that: When the large model is marked as a large load model, the deployment user of the large load model is analyzed, specifically: when the large model is marked as a large load model, all model service records of the large load model within H time before the current system time are obtained, the user id to which the data of the model service record belongs is obtained, the model service records of the same user id are marked as the same-household service records, the user rating of each user is obtained, the user threshold rating is set, and when the user's user rating is greater than or equal to the user threshold rating, the user is marked as a deployment user.

6. The large model management platform based on artificial intelligence according to claim 5 is characterized in that: The user rating of the user is obtained in the following way: All the same-household service records are sorted in the order of the model service time, and the time difference between the model service times of two adjacent same-household service records after sorting is calculated to obtain the same-household service interval. All the same-household service intervals are summed and averaged to obtain the average same-household service interval, which is marked as ZSK. The data volume of two adjacent same-household service records after sorting is calculated and the absolute value is taken to obtain the data fluctuation amount. The data fluctuation amount threshold is set. When the data fluctuation amount is greater than or equal to the data fluctuation amount threshold, the number of data fluctuations is increased by one. The number of data fluctuations is summed to obtain the total number of data fluctuations, which is marked as TLS. The total number of same-household service records is marked as RTC. The formula is used The user evaluation CAE of the user is obtained, where c1 is the average same-household service interval coefficient, c2 is the total number of data fluctuation coefficient, and c3 is the total number of same-household service records coefficient.

7. The large model management platform based on artificial intelligence according to claim 1 is characterized in that: Determine whether to allocate subsequent data of the deployed user to the remaining large models, specifically: take the device location of the deployed user as the center, draw a circle with a preset radius, obtain the service deployment range, mark the remaining large models of the same type as the load large model within the service deployment range as matching large models, obtain the matching value of the matching large model, set the matching threshold, when the matching value of the matching large model is greater than or equal to the matching threshold, mark the matching large model as the target large model, mark the target large model with the largest matching value as the allocation large model, directly allocate the subsequent data of the deployed user to the allocation large model, and at the same time, the allocation large model generates a user allocation record, and the user allocation record includes the user allocation time; When the target large model does not exist within the service deployment range, the subsequent data of the deployment user will continue to be allocated to the load large model; The optional value of the optional large model is obtained by the following method: obtain the model load value FSE of the optional large model, obtain all user allocation records of the optional large model within H time before the current system time, mark the total number of user allocation records as KGB, and use the formula The matching value XPZ of the matching large model is obtained, where d1 is the model load value coefficient and d2 is the total number coefficient of user allocation records.

Images