Method and system for optimizing configuration of multi-gpu network service based on large model

Abstract

The present application relates to a large model-based multi-GPU network service configuration optimization method and optimization system. The optimization method comprises: obtaining multi-GPU computing resource information; obtaining project background data through a large model based on project basic information; obtaining project picture information through a large model based on the obtained text information, and deriving project structured information; based on user historical data, combining project structured information, predicting the multi-GPU computing resources required by the demand side in different time periods in the future; statistically analyzing the size of the project three-dimensional point cloud data uploaded by the user, matching the three-dimensional point cloud data with the project structured information; based on the analysis and prediction, balancing the multi-GPU computing resource demand, and optimizing the multi-GPU computing resource allocation; after balancing and optimizing, according to the actual required GPU computing resource amount of the user and the prediction result, the multi-GPU computing resource usage in different time periods is priced in stages; real-time monitoring of multi-GPU computing resource usage and alarming when abnormal usage is detected.

Description

Technical Field

[0001] This invention relates to the field of graphics processor resource configuration optimization technology, and in particular to a method and system for optimizing the configuration of multi-GPU network services based on a large model. Background Technology

[0002] With the development of information technology, large-scale 3D point cloud data is being used more and more widely in various fields, such as geographic information systems, engineering and construction, and autonomous driving. Processing large-scale 3D point cloud data typically requires high-performance graphics processing units (GPUs) for computation. However, current GPU computing resource allocation methods often fail to meet the needs of different users at different times, leading to unreasonable resource allocation, wasting resources, and increasing data processing costs.

[0003] Therefore, the multi-GPU network service configuration optimization system and working method based on large models have great research value. Summary of the Invention

[0004] This invention aims to provide a multi-GPU network service configuration optimization method based on large models, which can accurately predict and dynamically adjust the multi-GPU computing resource requirements of different users at different time periods. This solves the problem of multi-GPU computing resource configuration optimization, as well as the resource waste and inefficiency caused by unreasonable resource allocation during large-scale 3D point cloud data processing.

[0005] The present invention also aims to provide an optimization system for implementing the above-described improved multi-GPU network service configuration optimization method.

[0006] According to one aspect of the present invention, a method for optimizing the configuration of multi-GPU network services based on a large model is provided, comprising the following steps: obtaining available multi-GPU computing resource information; receiving basic project information input by a user, and obtaining project background data based on the basic project information through a large model; obtaining existing project image information through the large model based on the text information associated with the basic project information and the project background data, and deriving project structured information accordingly; reading user historical data, combining it with the project structured information, constructing a user behavior profile, and predicting the multi-GPU computing resources required by the demand side at different future time periods; statistically analyzing the scale of the project 3D point cloud data uploaded by the user, and matching the 3D point cloud data with the project structured information; balancing the demand for multi-GPU computing resources based on the analysis and prediction, and optimizing the allocation of multi-GPU computing resources; after balancing and optimizing the multi-GPU computing resources, implementing tiered pricing for the use of multi-GPU computing resources at different time periods according to the amount of GPU computing resources required by the user and the prediction results; monitoring the usage of multi-GPU computing resources in real time, and issuing an alarm when abnormal usage is detected.

[0007] The following technical effects can be achieved by using the optimization method in this scheme:

[0008] 1. The multi-GPU network service configuration optimization method based on a large model provided by this invention acquires the overall multi-GPU computing resources of the system and predicts the multi-GPU computing resources required by the demand side at different time periods based on historical user data, thereby constructing a user behavior profile. Furthermore, this invention can perform real-time statistics on 3D point cloud data uploaded by different users, thereby calculating the multi-GPU resources required by each user, ensuring the real-time nature of resource configuration, and further improving the overall performance of the system.

[0009] After receiving 3D point cloud data uploaded by different users, the system calculates the GPU computing resources required by each user by statistically analyzing the scale of their 3D point cloud data. Based on this, the method can balance the demand for multi-GPU computing resources across different time periods, thereby optimizing the allocation of multi-GPU computing resources on the demand side. This invention employs a Monte Carlo algorithm to balance the demand for multi-GPU computing resources across different time periods, resulting in a more rational allocation of multi-GPU computing resources. This avoids resource waste and idleness, thereby improving the overall utilization rate of multi-GPU computing resources.

[0010] By optimizing the configuration of multi-GPU computing resources, resource costs for processing large-scale 3D point cloud data were saved. After resource coordination, tiered pricing can be implemented for different time periods based on the user's required multi-GPU computing resource segments. This allows users to choose the appropriate pricing scheme according to their actual needs when using multi-GPU computing resources, thereby reducing usage costs. The overall optimization scheme significantly improved operational and economic efficiency.

[0011] 2. The multi-GPU computing resource allocation optimization method of this invention can dynamically adjust the allocation of multi-GPU computing resources based on user historical data and real-time uploaded 3D point cloud data, better adapting to the needs of different users and application scenarios, and exhibiting high adaptability and flexibility. By optimizing the configuration of multi-GPU computing resources, large-scale 3D point cloud data processing tasks can be completed in a shorter time, improving data processing speed and efficiency, thereby providing users with a better experience. The multi-GPU configuration optimization method of this invention is not only applicable to fields such as geographic information systems, engineering construction, and autonomous driving, but can also be extended to other industries requiring large-scale 3D point cloud data processing, such as medical, film and television, and gaming fields, demonstrating strong versatility and broad application prospects.

[0012] In some embodiments, the optimization method further includes the following step: periodically generating a comprehensive report on multi-GPU computing resource usage, project progress, and forecast data.

[0013] In some embodiments, the multi-GPU computing resource information includes GPU device type, quantity, performance parameters, current usage status, and GPU device address.

[0014] In some embodiments, after obtaining the project background data, the method further includes receiving modified project parameters input by a user for at least one project parameter contained in the project background data.

[0015] In some embodiments, the step of predicting the GPU computing resources required by the demand side in different future time periods includes predicting the total multi-GPU computing resource demand for all time periods, and predicting the multi-GPU computing resource demand for each specific time period.

[0016] In some embodiments, the step of statistically analyzing the scale of the three-dimensional point cloud data includes:

[0017] Calculate the total amount and density of the three-dimensional point cloud data;

[0018] The complexity of processing 3D point cloud data is analyzed, including the computational load required for cleaning, registration, and classification of 3D point cloud data.

[0019] Analysis of the time requirements for processing 3D point cloud data.

[0020] In some embodiments, the step of balancing the computational demand of the multi-GPU computing resources and optimizing the allocation of the multi-GPU computing resources includes:

[0021] Balance the multi-GPU computing resource requirements at different time periods;

[0022] Dynamically adjust the multi-GPU computing resource requirements based on the balanced calculation results;

[0023] Develop strategies to improve the efficiency of multi-GPU computing resource utilization.

[0024] In some embodiments, the step of balancing the demand for multi-GPU computing resources and optimizing the allocation of multi-GPU computing resources includes: preloading user demand information and performing actual loading when the user actually needs it.

[0025] According to another aspect of the present invention, a multi-GPU network service configuration optimization system based on a large model is provided, the optimization system being used to perform the aforementioned optimization method, the optimization system comprising:

[0026] The GPU computing resource management module is configured to obtain information on available multi-GPU computing resources.

[0027] The large model-based project background information acquisition module is configured to acquire project background data through the large model based on the basic project information input by the user, and acquire existing project image information through the large model based on the text information associated with the basic project information and the project background data, and derive the project structured information accordingly.

[0028] The user-side data upload and statistics module is configured to receive 3D point cloud data of a project uploaded by a user, perform statistical analysis on the scale of the received 3D point cloud data, and match the 3D point cloud data with the project's structured information; the user-side data upload and statistics module is also configured to analyze the user's historical data.

[0029] The demand-side resource prediction module is configured to construct a user behavior profile based on the user's historical data and the project's structured information, and to predict the multi-GPU computing resources required by the demand side at different future time periods.

[0030] The balanced computing optimization module is configured to prepare and optimize the allocation of multi-GPU computing resources in advance based on the prediction.

[0031] The tiered pricing and resource allocation optimization module is configured to, after completing demand-side resource coordination, implement tiered pricing for the use of multi-GPU computing resources in different time periods based on the amount of multi-GPU computing resources required by the user.

[0032] The system resource monitoring module is configured to monitor GPU computing resource usage in real time and issue an alarm when abnormal usage is detected.

[0033] In some embodiments, the optimization system further includes a report generation module configured to periodically generate comprehensive reports on multi-GPU computing resource usage, project progress, and forecast data.

[0034] Other features and advantages of the present invention will partly become apparent to those skilled in the art upon reading this application, and partly will be described in conjunction with the accompanying drawings in the detailed description below. Attached Figure Description

[0035] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings, wherein:

[0036] Figure 1 This is a schematic diagram of the architecture of a multi-GPU network service configuration optimization system based on a large model according to an embodiment of the present invention;

[0037] Figure 2 This is a flowchart of a multi-GPU network service configuration optimization method based on a large model according to an embodiment of the present invention. Detailed Implementation

[0038] Referring now to the accompanying drawings, a schematic scheme of the multi-GPU network service configuration optimization method and system based on a large model disclosed in this invention will be described in detail. Although the drawings are provided to illustrate some embodiments of the invention, they are not necessarily drawn to the dimensions of the specific embodiments, and certain features may be enlarged, removed, or partially cut to better illustrate and explain the disclosure of the invention. Some components in the drawings may be repositioned according to actual needs without affecting the technical effect. The phrase "in the drawings" or similar expressions appearing in the specification do not necessarily refer to all drawings or examples.

[0039] Certain directional terms used in the description of the accompanying drawings below, such as “inner,” “outer,” “above,” “below,” and other directional terms, will be understood to have their normal meaning and refer to those directions as normally viewed in the accompanying drawings. Unless otherwise specified, the directional terms used in this specification are generally in accordance with the conventional directions understood by those skilled in the art.

[0040] The terms “first,” “first,” “second,” “second,” and similar terms used in this invention do not indicate any order, quantity, or importance, but are used to distinguish one component from others.

[0041] This invention provides a multi-GPU network service configuration optimization system and method based on large models, aiming to solve the problem of wasted resources caused by unreasonable allocation of multi-GPU computing resources, which makes it difficult to meet the needs of different users at different times. The main objective of this invention is to achieve accurate prediction and dynamic adjustment of the multi-GPU computing resource needs of different users at different times, thereby solving the problem of multi-GPU computing resource configuration optimization and addressing the resource waste and inefficiency caused by unreasonable allocation of multi-GPU computing resources during large-scale 3D point cloud data processing, ultimately improving the overall utilization rate of multi-GPU computing resources.

[0042] like Figure 1 As shown, according to the present invention, the multi-GPU network service configuration optimization system based on a large model includes a resource management module, a project background information acquisition module based on a large model, a user-end data upload and statistics module, a demand-side resource prediction module, a balanced computing optimization module, a tiered pricing and resource allocation optimization module, and a resource monitoring module. It may also additionally include a report generation module. The functions of these modules are described in detail below.

[0043] The resource management module collects information on all available multi-GPU computing resources from the optimization system, including the quantity, performance, and memory of GPU computing resources. Here, "all available multi-GPU computing resources" refers to high-performance GPUs suitable for 3D point cloud data processing, such as NVIDIA's Tesla, Quadro, and RTX series, or AMD's Radeon Instinct series.

[0044] The large-model-based project background information acquisition module automatically retrieves project background data from a large model based on user-inputted basic project information, including the project name. This background data includes, but is not limited to, at least one of the following project parameters: start date, construction scale, construction period, project budget, project team size, and expected project deliverables. Users can also manually correct at least one of these project parameters via wired or wireless input devices, or on the optimized system's touchscreen interface, to ensure the accuracy of the project background data and meet personalized needs. Based on the textual information obtained from the basic project information and project background data, the system automatically retrieves project images and other information from the internet using the large model and summarizes structured information.

[0045] The user-side data upload and statistics module receives 3D point cloud data from projects uploaded by different users. Based on this, the system optimizes the statistical analysis of the scale of the 3D point cloud data and automatically matches it with the project's structured information. This module also analyzes historical data from users' past system usage, such as requested GPU computing resources, usage duration, usage frequency, and usage time periods.

[0046] The demand-side resource prediction module is used to create a summary profile describing typical user behaviors and demand characteristics based on historical user data, thereby identifying user demand patterns and constructing user behavior profiles. Based on this, it uses algorithms such as Bayesian algorithms to predict the multi-GPU computing resources required in different future time periods.

[0047] The system balanced computing optimization module is used to prepare and optimize resource allocation in advance to meet the GPU computing resources that users may need at different times in the future. In this scenario, user information is preloaded, and actual loading is performed when there is actual demand. For example, the Monte Carlo algorithm in this module is used to balance the multi-GPU computing resource demands at different times, so as to achieve efficient utilization of system resources while meeting user needs.

[0048] The system's tiered pricing and resource allocation optimization module, after coordinating demand-side resources, uses tiered pricing to allocate GPU computing resources at different times based on the user's required GPU computing resources, thereby optimizing the allocation of multi-GPU computing resources. This improves the utilization rate of multi-GPU computing resources, reduces user resource costs, and allows users to choose the optimal resource usage plan according to their own needs.

[0049] The system resource monitoring module monitors the usage of multi-GPU computing resources in real time, including parameters such as current utilization, temperature, and power consumption. The optimization system can automatically detect abnormal usage, such as overload, idle time exceeding thresholds, and device malfunctions, and issue timely alerts upon detection. Based on real-time monitoring data, the optimization system can dynamically adjust the allocation strategy of multi-GPU computing resources to cope with sudden demands and resource bottlenecks.

[0050] The report generation module is used to periodically generate comprehensive reports, including multi-GPU computing resource usage, project progress, and forecast data. Based on the analysis results, the optimization system provides specific optimization suggestions to help users use GPU computing resources more efficiently.

[0051] This invention optimizes the configuration of multi-GPU computing resources in an optimized system, pre-allocating GPU computing resources to different tasks based on project type. This saves resource costs for large-scale 3D point cloud data processing and makes task allocation more rational. The optimized multi-GPU computing resource configuration improves response speed, avoids resource waste, and generates economic benefits.

[0052] It is understood that, in addition to the GPU, the hardware configuration of the optimization system of this invention may also include a central processing unit (CPU), memory, hard disk, and network. Specifically, the CPU is selected as a high-performance multi-core processor, such as an Intel Xeon or AMD EPYC series; the memory is configured to be sufficient based on the actual data scale and computing needs, such as 128GB, 256GB, or higher; the hard disk is configured with high-speed, large-capacity storage devices, such as NVMe SSDs or distributed storage systems (such as Ceph, GlusterFS); and the network is configured with high-speed, low-latency network devices, such as 10Gbps, 40Gbps, or higher Ethernet switches and network cards.

[0053] In addition, the software that makes up the optimization system can include an operating system, GPU computing software, a 3D point cloud processing library, a containerization technology module, a task scheduling and management module, and a user management and authentication module. Specifically: the operating system should be a stable operating system compatible with GPU computing, such as Windows, Ubuntu, or CentOS; the GPU computing software refers to the installation of GPU-related software, such as CUDA, cuDNN (for NVIDIA GPUs), or ROCm (for AMD GPUs); the 3D point cloud processing library refers to the installation and configuration of relevant libraries for 3D point cloud data processing, such as PCL (Point Cloud Library) and Open3D; the containerization technology module refers to the use of containerization technologies, such as Docker or Kubernetes, to achieve rapid application deployment and easy management; the task scheduling and management module should select or develop a task scheduling and management system suitable for the user's business scenarios, such as Apache Mesos, Kubernetes, or Slurm; and the user management and authentication module should implement user management and authentication functions, such as using OAuth 2.0 or LDAP technologies.

[0054] like Figure 2 As shown, the optimization method performed by the aforementioned large-model-based multi-GPU network service configuration optimization system may include the following steps:

[0055] S1: Obtain information on the overall available multi-GPU computing resources on the server.

[0056] First, the optimization system collects information on all GPU computing resources, including the types and quantities of available GPU devices, performance parameters (such as memory, computing power, and power consumption), and their current usage status (idle, partially occupied, or fully occupied). In addition, the optimization system records the addresses of the GPU devices for flexible scheduling when needed.

[0057] S2: Based on the basic project information input by the user, including the project name, automatically obtain project background data through a large model, including at least one of the following project parameters: start time, construction scale, construction period, project budget, project team size, and expected project results.

[0058] Specifically, after a user uploads basic project information to the optimization system, the system automatically retrieves background data for the project from relevant websites based on an artificial intelligence model. Key parameters include, but are not limited to, start date, construction scale, construction period, project budget, project team size, and expected project results.

[0059] Users can also manually modify at least one of these parameters via wired or wireless devices or on the system's touchscreen interface to ensure data accuracy and personalization.

[0060] S3: Based on textual information obtained from basic project information and background data, the system automatically retrieves project image information from the internet using a large model and summarizes the structured information of the project. In response, the optimization system analyzes the user's project progress and integrates historical data from past use of the optimization system, such as requested GPU computing resources, usage duration, frequency, time period, task type, and task complexity. This analysis helps the optimization system understand the user's usage patterns and needs, providing a basis for subsequent multi-GPU computing resource scheduling.

[0061] S4: Based on historical user data and the obtained structured project information, the optimization system predicts the multi-GPU computing resources required for different time periods in the future. By reading historical user data and combining it with the obtained structured project information, the optimization system can build user behavior profiles that go beyond traditional methods. For example, using Bayesian algorithms, it can predict the multi-GPU computing resources required for different time periods in the future. The prediction results include not only the overall GPU computing resource demand over time but also the GPU computing resource demand for each specific time period. Based on this, the optimization system constructs user behavior profiles, describing user demand trends and preferences, as well as possible sudden demand and resource peak periods.

[0062] S5: After a user uploads 3D point cloud data for their project to the system, the optimization system statistically analyzes the scale of the 3D point cloud data and automatically matches it with the project's structured information. Specifically, after a user uploads 3D point cloud data, the optimization system statistically analyzes the scale of the 3D point cloud data uploaded by each user, including the total amount and density of the 3D point cloud data, the processing complexity of the 3D point cloud data (e.g., the computational load required for cleaning, registration, and classification of the 3D point cloud data), and the time requirements for 3D point cloud data processing (real-time processing or batch processing). These analysis results are used to accurately estimate the amount of GPU computing resources required by each user, ensuring the accuracy and effectiveness of resource allocation.

[0063] S6: Balance the computational resource requirements of multiple GPUs and optimize GPU computational resource configuration. Based on the aforementioned analysis and prediction results, the optimization system can use the Monte Carlo algorithm to balance the GPU computational resource requirements at different time periods, thereby dynamically adjusting resource requirements and formulating strategies to improve resource utilization efficiency. Through the optimization algorithm, the system ensures that GPU computational resources are used reasonably and efficiently at any time, minimizing resource waste and bottlenecks.

[0064] S7: Implement a tiered pricing strategy. After resource coordination is complete, the optimization system will implement a tiered pricing strategy for different time periods based on the user's required multi-GPU computing resources and prediction results. This includes a resource premium strategy during peak hours, a resource discount strategy during off-peak hours, and a discount strategy for users who use the system long-term. This strategy helps the system optimize the configuration of multi-GPU computing resources, improve resource utilization, reduce user resource costs, and enable users to choose the optimal resource usage plan according to their own needs.

[0065] S8: Real-time monitoring of multi-GPU computing resource usage. The optimization system can monitor the usage of multi-GPU computing resources in real time, including parameters such as current utilization, temperature, and power consumption. The optimization system can automatically detect abnormal usage (such as overload, prolonged idle time exceeding a preset period, and equipment failure) and issue timely alerts. Based on real-time monitoring data, the optimization system can dynamically adjust resource allocation strategies to cope with sudden demand and resource bottlenecks. The system can temporarily allocate additional GPU resources to cope with sudden peak task periods, or perform maintenance and upgrades when resources are idle.

[0066] S9: Regularly generate comprehensive reports. The optimization system can generate comprehensive reports periodically as needed, including multi-GPU computing resource usage, project progress, and forecast data. Based on the analysis results, the optimization system provides specific optimization suggestions to help users use GPU computing resources more efficiently.

[0067] It should be understood that although this specification describes various embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0068] The above description is merely an illustrative embodiment of the present invention and is not intended to limit the scope of the invention. Any equivalent changes, modifications, and combinations made by those skilled in the art without departing from the concept and principles of the present invention should fall within the scope of protection of the present invention.

Claims

1. A method for optimizing the configuration of multi-GPU network services based on a large model, characterized in that, Includes the following steps: Obtain information on available multi-GPU computing resources; Receive basic project information input by the user, and obtain project background data based on the basic project information through a large model; Based on the text information associated with the basic project information and the project background data, the existing project image information is obtained through the large model, and the project structured information is derived accordingly. Read user historical data, combine it with the project's structured information, construct user behavior profiles, and predict the multi-GPU computing resources required by the demand side at different future time periods; The scale of the 3D point cloud data of the project uploaded by the user is statistically analyzed, and the 3D point cloud data is matched with the structured information of the project. Based on the analysis and prediction, the demand for multi-GPU computing resources is calculated in a balanced manner, and the allocation of multi-GPU computing resources is optimized. After balancing and optimizing multi-GPU computing resources, tiered pricing is implemented for the use of multi-GPU computing resources in different time periods based on the actual amount of GPU computing resources required by users and the prediction results. Monitor the usage of multi-GPU computing resources in real time and issue alerts when abnormal usage is detected.

2. The multi-GPU network service configuration optimization method based on a large model according to claim 1, characterized in that, The optimization method further includes the following steps: Regularly generate comprehensive reports on multi-GPU computing resource usage, project progress, and forecast data.

3. The multi-GPU network service configuration optimization method based on a large model according to claim 1, characterized in that, The multi-GPU computing resource information includes GPU device type, quantity, performance parameters, current usage status, and GPU device address.

4. The multi-GPU network service configuration optimization method based on a large model according to claim 1, characterized in that, After obtaining the project background data, the method also includes receiving corrected project parameters input by the user for at least one project parameter contained in the project background data.

5. The multi-GPU network service configuration optimization method based on a large model according to claim 1, characterized in that, The steps for predicting the GPU computing resources required by the demand side in different future time periods include predicting the total multi-GPU computing resource demand for all time periods, and predicting the multi-GPU computing resource demand for each specific time period.

6. The multi-GPU network service configuration optimization method based on a large model according to claim 1, characterized in that, The steps for statistical analysis of the scale of the 3D point cloud data include: Calculate the total amount and density of the three-dimensional point cloud data; The complexity of processing 3D point cloud data is analyzed, including the computational load required for cleaning, registration, and classification of 3D point cloud data. Analysis of the time requirements for processing 3D point cloud data.

7. The multi-GPU network service configuration optimization method based on a large model according to claim 1, characterized in that, The steps of balancing the demand for multi-GPU computing resources and optimizing the allocation of multi-GPU computing resources include: Balance the multi-GPU computing resource requirements at different time periods; Dynamically adjust the multi-GPU computing resource requirements based on the balanced calculation results; Develop strategies to improve the efficiency of multi-GPU computing resource utilization.

8. The multi-GPU network service configuration optimization method based on a large model according to claim 1, characterized in that, The steps of balancing the computational resource requirements of the multi-GPU systems and optimizing the allocation of these resources include: This is to preload user demand information and then load it when the user actually needs it.

9. A multi-GPU network service configuration optimization system based on a large model, characterized in that, The optimization system is used to execute the optimization method according to any one of claims 1 to 8, and the optimization system comprises: The GPU computing resource management module is configured to obtain information on available multi-GPU computing resources. The large model-based project background information acquisition module is configured to acquire project background data through the large model based on the basic project information input by the user, and acquire existing project image information through the large model based on the text information associated with the basic project information and the project background data, and derive the project structured information accordingly. The user-side data upload and statistics module is configured to receive 3D point cloud data of a project uploaded by a user, perform statistical analysis on the scale of the received 3D point cloud data, and match the 3D point cloud data with the project's structured information; the user-side data upload and statistics module is also configured to analyze the user's historical data. The demand-side resource prediction module is configured to construct a user behavior profile based on the user's historical data and the project's structured information, and to predict the multi-GPU computing resources required by the demand side at different future time periods. The balanced computing optimization module is configured to prepare and optimize the allocation of multi-GPU computing resources in advance based on the prediction. The tiered pricing and resource allocation optimization module is configured to, after completing demand-side resource coordination, implement tiered pricing for the use of multi-GPU computing resources in different time periods based on the amount of multi-GPU computing resources required by the user. The system resource monitoring module is configured to monitor GPU computing resource usage in real time and issue an alarm when abnormal usage is detected.

10. The multi-GPU network service configuration optimization system based on a large model according to claim 9, characterized in that, It also includes a report generation module, configured to periodically generate comprehensive reports on multi-GPU computing resource usage, project progress, and forecast data.

Images