ChatGlm3-based adaptive containerized cloud model parameter adjustment method and system, and electronic equipment

Abstract

The invention discloses an adaptive containerized cloud model parameter adjustment method and system based on ChatGlm < 3 > and electronic equipment, in a ChatGlm < 3 > scene, an algorithm based on elastic resource management is adopted for parameter adjustment of a containerized model adaptive to containerized cloud, performance indexes and resource use conditions of the containerized model are dynamically monitored, and the parameter adjustment accuracy of the containerized model is improved. According to a preset rule or by learning historical data, distribution of container virtual resources is automatically adjusted, a customer service level protocol is improved, scheduling based on priorities is ensured through distribution of pre-virtual CPUs, and dynamic scheduling based on the priorities is achieved through a task load monitoring mechanism. According to the method, the parameters can be automatically adjusted on the containerized cloud platform, so that the performance is improved, the resource consumption is reduced, the iteration period is accelerated, different workload and performance requirements can be met, and the problem of complexity of resource configuration and scheduling is effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing technology, and in particular to a method, system and electronic device for adjusting parameters of a containerized cloud model based on ChatGlm3. Background Art

[0002] In today's era of booming cloud computing and artificial intelligence, model parameter adjustment technology adapted to containerized cloud computing has become an important area of ​​concern. With the increasing demand of enterprises to use machine learning and deep learning models to solve practical problems, and the popularity of cloud computing platforms, it has become a trend to containerize machine learning models and deploy them in the cloud. Behind this trend is that container technology can provide a lightweight, portable, and scalable deployment environment, while cloud computing platforms provide enterprises with flexible and elastic resource management capabilities.

[0003] ChatGLM3 is a conversation pre-training model jointly released by Zhipu AI and Tsinghua University KEG Laboratory. ChatGLM3-6B is an open source model in the ChatGLM3 series, which retains many excellent features of the previous two generations of models, such as smooth conversation and low deployment threshold. Traditional model parameter adjustment is often performed in a fixed hardware environment, which is inconsistent with the flexibility and dynamism of cloud container deployment. Therefore, model parameter adjustment technology adapted to containerization and cloud computing has emerged. This technology is committed to solving the balance between model performance and resource utilization in containerized deployment scenarios. By dynamically adjusting and optimizing model parameters, the model can achieve optimal performance in a cloud container environment.

[0004] There are doubts about the applicability of traditional model parameter adjustment methods in cloud container environments. Traditional resource management methods usually use static configuration and rely on fixed resource allocation strategies. This often leads to inflexible resource allocation in dynamically changing cloud environments and cannot effectively respond to changes in workloads, thus affecting model performance or causing resource waste. Therefore, researchers began to explore how to take advantage of container technology and combine the needs of machine learning model parameter adjustment to develop a set of model parameter adjustment technologies suitable for cloud container environments. Researchers focus on how to use container orchestration tools, monitoring systems, and adaptive algorithms to achieve dynamic adjustment of model parameters, and propose a series of methods and algorithms, such as container-aware adaptive parameter adjustment algorithms, which use container internal resource usage to dynamically adjust model parameters to optimize performance. At the same time, they also focus on how to ensure the stability, reliability, and security of the model during container deployment, and achieve this goal through container orchestration tools and container security technologies.

[0005] The background of model parameter tuning technology adapted to containerization on the cloud can be summarized as the integration of cloud computing and artificial intelligence, as well as the challenge and reflection of traditional model parameter tuning methods driven by the demand for deploying machine learning models on the cloud. Researchers continue to explore and innovate, and strive to find model parameter tuning solutions suitable for cloud container environments to meet the needs of enterprises for high performance and high efficiency. The complexity of resource configuration and scheduling is a very important issue in model parameter tuning technology adapted to containerization on the cloud. In a multi-tenant, resource-sharing cloud environment, how to effectively allocate appropriate resources to each containerized model and how to adjust these resources to adapt to changes in model performance have become a difficult point, especially in the face of sudden high-concurrency requests or dynamically changing data streams. Real-time adjustment of resources becomes particularly critical, but also extremely challenging. This not only affects the execution efficiency of the model, but may also lead to resource waste or reduced service quality. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for adjusting parameters of an adaptive containerized cloud model based on ChatGlm3, which adopts an algorithm based on elastic resource management to dynamically monitor the performance indicators and resource usage of the containerized model, and automatically adjust resource allocation according to preset rules or by learning historical data.

[0007] To achieve the above purpose, the technical solution provided by the present invention is:

[0008] The first aspect of the present application provides a parameter adjustment method for a containerized cloud-adapted model based on ChatGlm3. In the ChatGlm3 scenario, the parameter adjustment of the containerized model adapted for containerized cloud-adaptation adopts an algorithm based on elastic resource management, dynamically monitors the performance indicators and resource usage of the containerized model, and automatically adjusts the allocation of container virtual resources according to preset rules or by learning historical data to improve the customer service level agreement, and ensures priority-based scheduling through the allocation of pre-virtualized CPUs, and realizes priority-based dynamic scheduling through the task load monitoring mechanism.

[0009] To optimize the above technical solutions, the specific measures taken also include:

[0010] An elastic scaling strategy is formed through an elastic resource management algorithm to scale container virtual resources. When it is detected that the performance of the containerized model has degraded due to insufficient container virtual resources, the container virtual resource allocation is automatically increased. When it is detected that the containerized model is wasting due to insufficient use of container virtual resources, the container virtual resource allocation is automatically reduced.

[0011] The described automatic adjustment of the allocation of container virtual resources according to preset rules or by learning historical data to improve customer service level protocol adopts a multi-group coyote optimization algorithm, which optimizes the parameters in the containerization model parameter adjustment by simulating the behavior of the coyote group, combining the multi-group and population update strategy, and realizes the solution of the multi-objective optimization problem.

[0012] Furthermore, the multi-group coyote optimization algorithm specifically includes the following steps:

[0013] Initialize the position and speed of the coyote individual. The position of the individual represents the allocation ratio of the container's virtual resources, and the speed represents the rate of change of the container's virtual resources.

[0014] Calculate the fitness of each individual, use the performance indicators under the current container virtual resource configuration to evaluate the performance of the containerized model, and obtain the fitness value of each individual to guide the update of the individual position;

[0015] According to the current position of the best individual and the best group, the speed update formula is used to update the position and speed of each individual.

[0016] The pre-virtual CPU allocation ensures priority-based scheduling using a pigeon flock heuristic optimization algorithm, which performs task clustering and a fast adaptive feedback loop neural network for pre-virtual CPU allocation to ensure priority-based scheduling.

[0017] Furthermore, the pigeon flock heuristic optimization algorithm specifically includes the following steps:

[0018] The tasks are clustered using the pigeon flock heuristic optimization algorithm, similar tasks are assigned to the same group, and the foraging behavior of pigeon flocks is simulated to search for the optimal clustering solution;

[0019] A fast adaptive feedback loop neural network model is designed and trained to predict the pre-virtualized CPU resources required for each task and achieve fast adaptive feedback.

[0020] The task load monitoring mechanism is used to realize priority-based dynamic scheduling using a deep convolutional neural network, and the deep convolutional neural network is used to realize priority-based dynamic scheduling, thereby improving the resource allocation and scheduling efficiency of containerized applications in a cloud environment.

[0021] Furthermore, the deep convolutional neural network specifically includes the following steps:

[0022] Collect task load data of containerized applications in cloud environments and pre-process the data;

[0023] A deep convolutional neural network model suitable for task load monitoring is designed. The features of the received task are taken as input, feature extraction and learning are performed, and the load status of the task is output. By training the deep convolutional neural network model and optimizing the loss function, the deep convolutional neural network model can accurately predict the load status of the task.

[0024] The second aspect of the present application provides a container-based cloud model parameter adjustment system based on ChatGlm3, including:

[0025] The container virtual resource dynamic adjustment module is used to automatically adjust the allocation of container virtual resources according to preset rules or by learning historical data to improve customer service level agreements;

[0026] Task clustering priority scheduling module, used to ensure priority-based scheduling by allocating pre-virtualized CPUs;

[0027] Real-time dynamic resource scheduling module, which is used to realize priority-based dynamic scheduling through task load monitoring mechanism;

[0028] The container virtual resource dynamic adjustment module adopts a multi-group coyote optimization algorithm, the task clustering priority scheduling module adopts a pigeon flock heuristic optimization algorithm, and the real-time dynamic resource scheduling module adopts a deep convolutional neural network.

[0029] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the adapted containerized cloud model parameter adjustment method based on ChatGlm3 as described in the first aspect of the present application is implemented.

[0030] The fourth aspect of the present application provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the adapted containerized cloud model parameter adjustment method based on ChatGlm3 as described in the first aspect of the present application.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The present invention provides a model parameter adjustment technology based on ChatGlm3 adapted for containerized cloud migration. The model parameter adjustment technology mainly adapted for containerized cloud migration is applied to various scenarios, especially in cloud computing environments, where machine learning models need to be adjusted and optimized quickly and flexibly to adapt to different workloads and data characteristics.

[0033] This technology enables the model to automatically adjust parameters on a containerized cloud platform, thereby improving performance, reducing resource consumption and accelerating iteration cycles. Its application scenarios include but are not limited to automated machine learning (Auto ML), real-time data processing and analysis, and large-scale parallel computing tasks. Through this technology, enterprises and research institutions can more efficiently deploy and manage their machine learning models to ensure optimal performance in a dynamically changing cloud environment.

[0034] The present invention adopts an algorithm based on elastic resource management to dynamically monitor the performance indicators and resource usage of the containerized model, and automatically adjusts resource allocation according to preset rules or by learning historical data; using elastic scaling strategies, when it is detected that the model's performance has degraded due to insufficient resources, resource allocation is automatically increased; conversely, when resource usage is insufficient and wasteful, resource allocation is reduced. In this way, not only can the model be guaranteed to maintain efficient operation in a changing cloud environment, but resource usage can also be optimized to reduce costs. The specific algorithm implementation can adopt a machine learning algorithm, such as reinforcement learning, to adapt to different workloads and performance requirements by continuously learning the optimal resource adjustment strategy, thereby effectively solving the complexity of resource configuration and scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 : Flow chart of the parameter adjustment method of the adapted containerized cloud model based on ChatGlm3 of the present invention.

[0036] Figure 2 : Schematic diagram of the process flow of multi-group coyote optimization algorithm of the present invention.

[0037] Figure 3 : Schematic diagram of the process flow of the pigeon flock heuristic optimization algorithm of the present invention.

[0038] Figure 4 : Schematic diagram of the process flow of the deep convolutional neural network of the present invention. DETAILED DESCRIPTION

[0039] The above contents of the present invention are further described in detail below in the form of embodiments, but this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the above contents of the present invention belong to the scope of the present invention.

[0040] This application proposes a model parameter adjustment technology based on ChatGlm3 that is adapted to containerization and cloud migration. By dynamically adjusting container virtual resources, task clustering priority scheduling, and real-time dynamic resource scheduling, the performance of containerized applications in cloud environments is improved and resource utilization is optimized. This application conducts in-depth research on various types of businesses and addresses challenges faced in the process of containerization and cloud migration, including but not limited to: model performance degradation, low resource utilization, poor operating efficiency, etc. Through in-depth understanding of different businesses, model parameters are adjusted in a targeted manner to solve performance problems in different business scenarios, thereby improving the overall performance and efficiency of containerized applications.

[0041] The embodiments of the present application comprehensively utilize the model parameter adjustment technology based on ChatGlm3, the improved multi-group coyote optimization (MMCO) algorithm, the pigeon flock heuristic optimization (MPIO) algorithm, the fast adaptive feedback recurrent neural network (FARNN) and the deep convolutional neural network (DCNN) and other technologies to achieve dynamic adjustment of virtual resources, task clustering priority scheduling and real-time dynamic resource scheduling, thereby significantly improving the performance of containerized applications in cloud environments, optimizing resource utilization and reducing costs.

[0042] The embodiment of the present application provides a model parameter adjustment technology based on ChatGlm3 for adapting containerized cloud computing. The improved multi-group coyote optimization (MMCO) algorithm can dynamically adjust virtual resources to meet service level agreements and improve customer satisfaction; the pigeon flock heuristic optimization (MPIO) algorithm is combined with the fast adaptive feedback recurrent neural network (FARNN) to improve processing efficiency through task clustering and priority scheduling; the deep convolutional neural network (DCNN) is used to monitor task loads and realize real-time dynamic resource scheduling; the present application can significantly improve the performance of containerized applications in the cloud environment, while optimizing resource utilization, reducing costs, and ensuring the efficient operation and economy of the cloud computing environment, which is crucial for supporting large-scale, dynamically changing cloud computing application scenarios.

[0043] In one embodiment, the present invention proposes a parameter adjustment method for a containerized cloud model based on ChatGlm3. The flowchart is as follows: Figure 1 As shown in the figure, the parameter adjustment method for adapting the containerized cloud model based on ChatGlm3 includes the following steps:

[0044] The MMCO algorithm scales container virtual resources:

[0045] like Figure 2 As shown, in the model parameter adjustment for adapting containerization to the cloud, the improved multi-group coyote optimization (MMCO) algorithm is used to scale the virtual resources of the container, thereby improving the customer service level agreement.

[0046] In the ChatGlm3 scenario, the machine learning model is deployed in a containerized environment, and the container virtual resources are dynamically adjusted to optimize the model performance. Specifically, the MMCO algorithm simulates the behavior of coyote groups and combines multi-group and population update strategies to achieve efficient solutions to multi-objective optimization problems. In the model parameter adjustment of containerization on the cloud, the MMCO algorithm can be used to optimize model parameters and dynamically adjust the container virtual resource allocation based on real-time monitored performance indicators to meet the changing workload and performance requirements.

[0047] The MMCO algorithm involves some key steps and formula derivations:

[0048] First, define the objective function of the problem, such as minimizing the training time of the model or maximizing the accuracy of the model;

[0049] Then, through the cooperation and competition between coyote individuals and groups, the objective function is gradually optimized. In the MMCO algorithm, the position update between individuals is realized by the following formula:

[0050] xi(t+1)=xi(t)+vi(t+1)

[0051] Speed ​​update formula:

[0052] Vi(t+1)=w*vi(t)+c1*r1*(pbesti(t)-xi(t))+c2*r2*(gbest(t)-xi(t))

[0053] Among them, xi(t) represents the position of individual i at time t; vi(t+1) represents the speed of individual at time t+1; w represents the inertia weight; c1 represents the individual learning factor; c2 represents the group learning factor; r1 and r2 represent random numbers; pbesti(t) represents the best position of individual i at time t; gbest(t) represents the best position of the entire group at time t.

[0054] Initialize the position and speed of the individual coyote according to the actual situation. The position of the individual represents the allocation ratio of the container's virtual resources, and the speed represents the rate of change of the container's virtual resources. For example, if the CPU and memory resource allocation ratios of a container are CPU and "Memory" respectively, then the position of the individual is a two-dimensional vector (CPU, Memory), and the speed can be expressed as the rate of change of the container's virtual resource allocation ratio, which is used to guide the update of the individual position.

[0055] Next, calculate the fitness of each individual, that is, the performance indicator of the model under the current resource configuration; in this process, allocate CPU and Memory according to the current resources, and use appropriate performance indicators to evaluate the performance of the model, such as minimizing training time or maximizing model accuracy. In this way, the fitness value of each individual can be obtained to guide the update of the individual position.

[0056] Then, according to the current position of the best individual pbest and the best group qbest, the speed update formula is used to update the position and speed of each individual. The speed update formula includes the inertia weight w, the individual learning factor c1, the group learning factor c2, and the random numbers r1 and r2. These parameters will affect the update speed and direction of the individual position, thereby guiding the individual to move in the direction of better performance.

[0057] In the actual iteration process, the new position of each individual is obtained according to the speed update formula, and then the performance indicators of the model are recalculated, and the positions of the best individual pbest and the best group qbest are updated according to these indicators. In this way, the algorithm can continuously search for the optimal resource allocation plan to improve model performance and optimize resource utilization.

[0058] In general, the allocation ratio of container virtual resources is continuously adjusted through the iterative optimization process to improve model performance and optimize resource utilization, thereby achieving the goal of model parameter adjustment that adapts to containerization and cloud computing.

[0059] Applying the MMCO algorithm to model parameter adjustment for container-based cloud computing can achieve dynamic scaling of container virtual resources to adapt to different workloads and performance requirements. By monitoring the model's performance indicators and container virtual resource utilization, the MMCO algorithm can automatically adjust the container virtual resource allocation to ensure the efficient operation of the model in the cloud environment, enabling enterprises to better utilize the elasticity and flexibility of the cloud computing platform, achieve optimized deployment of machine learning models, improve customer service levels, and reduce resource costs.

[0060] MPIO algorithm improves FARNN to ensure priority:

[0061] like Figure 3 As shown, the improved pigeon flock heuristic optimization (MPIO) algorithm is used for task clustering and the fast adaptive feedback recurrent neural network (FARNN) is used for pre-virtual CPU allocation to ensure priority-based scheduling.

[0062] During the implementation, the MPIO algorithm is first used to cluster tasks and assign similar tasks to the same group for subsequent priority scheduling;

[0063] Then, a FARNN model is designed and trained to predict the CPU resources required for each task and achieve fast adaptive feedback.

[0064] In the task clustering stage, tasks are grouped according to their similarities and a unique ID is assigned to each group; the population is initialized and the MPIO algorithm is used, the core of which is to simulate the foraging behavior of pigeon flocks to search for the optimal clustering solution.

[0065] The specific implementation is as follows:

[0066] A certain number of initial pigeon groups are randomly generated, and each pigeon represents a task; the positions of these tasks represent their positions in the feature space; the operation is to calculate the fitness of each pigeon (task) based on the similarity between tasks; fitness can be measured based on the feature similarity between tasks, and the more similar tasks are assigned to the same group, the higher the fitness; the position of each pigeon is updated according to the current position and speed update formula; in this process, the position of the pigeon (task) will move in the direction of higher fitness to find a better task clustering solution; the speed of each pigeon is updated according to the current speed and a certain acceleration; this process will affect the moving speed and direction of the pigeon (task) to better explore the task clustering space; in each iteration process, check whether the conditions for stopping iteration are met, such as reaching the maximum number of iterations or the fitness reaching a certain threshold; if the stopping conditions are not met, repeat steps 2 to 5 until the optimal task clustering solution is found; in the whole process, the MPIO algorithm is used, and its core is to simulate the foraging behavior of the pigeon group to search for the optimal clustering solution.

[0067] Repeat the above steps until the stopping condition is met.

[0068] The MPIO algorithm can be used to obtain the clustering results of tasks and assign a unique ID to each task to prepare for subsequent CPU allocation.

[0069] Next, in the design and training phase of the FARNN model, a training data set is prepared, including the characteristics of the task and the corresponding CPU resource requirements; then, a FARNN model suitable for task demand prediction is designed, and the model is trained using the training data set; the hidden layer state update formula and output layer prediction formula of the FARNN model are as follows:

[0070] h t =f(W hh h t-1 +W xh x t +b h )

[0071] y t =g(W hyh t +b y )

[0072] Among them, h t-1 is the hidden layer state, xt is the input feature, yt is the output, i.e., the CPU resource requirement, W hh , W xh and W hy is the weight matrix, b h and b y is the bias term, f and g are the hidden layer activation function and the output layer activation function respectively.

[0073] After training is completed, the trained FARNN model is used to predict the CPU resource requirements of each task. In the actual task scheduling process, tasks are sorted and CPU resources are allocated according to their priorities and predicted CPU resource requirements to ensure priority-based scheduling. At the same time, the execution of tasks is monitored and the FARNN model is updated in a timely manner to adapt to changes in task requirements.

[0074] Regarding the parameter adjustment of the MPIO algorithm, the population size is set to 100, which can ensure that there are enough pigeon groups to search the space of task clustering without being too large to cause excessive computational burden; the number of iterations is set to 200, which is usually enough for the algorithm to converge to a good solution; the inertia weight w can be set to 0.7, the individual learning factor and the group learning factor can be set to 1.5 and 2 respectively, which can balance the influence of individuals and groups on the current best solution and the global best solution, and improve the search efficiency and convergence of the algorithm.

[0075] For the parameter adjustment of the FARNN model, the number of hidden layer neurons is selected as 50, which can avoid overfitting while maintaining a certain model complexity; the learning rate is set to 0.001, which is usually a small value and can ensure the stability of model parameter updates; ReLU is selected as the activation function, which can effectively handle nonlinear relationships and improve the expressiveness of the model; the setting of these parameters is based on a comprehensive consideration of the experimental results, which can enable the algorithm and model to achieve good results in task clustering and CPU resource prediction.

[0076] In summary, the implementation process of this solution includes the task clustering stage and the design and training stage of the FARNN model. By clustering tasks through the MPIO algorithm and using the FARNN model to predict the CPU resource requirements of tasks, fast adaptive feedback can be achieved, thereby ensuring priority-based task scheduling.

[0077] DCNN implements dynamic priority scheduling:

[0078] like Figure 4As shown in the figure, a task load monitoring mechanism based on deep convolutional neural network (DCNN) is designed to realize priority-based dynamic scheduling; these strategies are used in combination to improve the resource allocation and scheduling efficiency of containerized applications in cloud environments.

[0079] Design a task load monitoring mechanism based on a deep convolutional neural network (DCNN) and implement priority-based dynamic scheduling using the following steps:

[0080] Collect task load data of containerized applications in cloud environments, including task type, priority, resource requirements, and other information;

[0081] Then, the data is preprocessed, including data cleaning, feature extraction, and normalization, to facilitate the training and prediction of the DCNN model;

[0082] A deep convolutional neural network (DCNN) model suitable for task load monitoring is designed. The model takes the features of the received tasks as input, such as task type, priority, etc., and then extracts and learns the features through structures such as convolutional layers, pooling layers, and fully connected layers, and finally outputs the load status of the task; the DCNN model is trained using a labeled task load dataset, and the loss function is optimized so that the model can accurately predict the load status of the task.

[0083] In order to implement task load monitoring and priority-based dynamic scheduling based on deep convolutional neural network (DCNN), the following sample data was collected:

[0084] Task 1 is image processing, with a priority of 4 (high), and resource requirements of 80% CPU, 200MB memory, and 10GB storage; Task 2 is data analysis, with a priority of 3 (medium), and resource requirements of 60% CPU, 300MB memory, and 5GB storage; Task 3 is model training, with a priority of 5 (highest), and resource requirements of 90% CPU, 400MB memory, and 20GB storage. A DCNN model is designed, which takes task features as input and can predict the load status of tasks after training. For example, the predicted load status of Task 1 is 90% CPU, 180MB memory, 9GB storage, and priority 4 (high); the model is deployed in a cloud environment to monitor the load status of tasks in real time, and a priority-based dynamic scheduling strategy is adopted according to the priority of the task and the predicted load status to achieve priority execution and resource allocation of tasks.

[0085] After training is completed, the DCNN model is deployed to the cloud environment and the load status of the task is monitored in real time. Whenever a new task arrives, the characteristics of the task are input into the DCNN model, and the model is used to predict the load status of the task, including resource requirements and priority. In the task load monitoring stage, a priority-based dynamic scheduling strategy is adopted in combination with the task priority information. Specifically, the execution order and resource allocation of tasks are dynamically adjusted according to the priority of the task and the predicted load status; tasks with high priority will be scheduled first and allocated more resources to ensure their priority execution under limited resources.

[0086] In the actual implementation process, the performance of the DCNN model is continuously monitored, and the model is tuned according to real-time data. Other task scheduling algorithms and strategies, such as load balancing algorithms and dynamic resource adjustment, are also combined to further improve the resource allocation and scheduling efficiency of containerized applications. By designing a DCNN-based task load monitoring mechanism and combining it with a priority-based dynamic scheduling strategy, more efficient containerized application resource allocation and scheduling can be achieved in a cloud environment.

[0087] In another embodiment, the present invention proposes a parameter adjustment system for a model based on ChatGlm3 adapted for containerization and cloud computing, comprising:

[0088] The container virtual resource dynamic adjustment module is used to automatically adjust the allocation of container virtual resources according to preset rules or by learning historical data to improve customer service level agreements;

[0089] Task clustering priority scheduling module, used to ensure priority-based scheduling by allocating pre-virtualized CPUs;

[0090] Real-time dynamic resource scheduling module, which is used to realize priority-based dynamic scheduling through task load monitoring mechanism;

[0091] The container virtual resource dynamic adjustment module adopts a multi-group coyote optimization algorithm, the task clustering priority scheduling module adopts a pigeon flock heuristic optimization algorithm, and the real-time dynamic resource scheduling module adopts a deep convolutional neural network.

[0092] In another embodiment, the present invention proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned method for adjusting parameters of the adapted containerized cloud model based on ChatGlm3 is implemented.

[0093] In another embodiment, the present invention proposes a computer-readable storage medium storing a computer program, which enables a computer to execute the above-mentioned method for adjusting parameters of a containerized cloud model based on ChatGlm3.

[0094] In the model parameter adjustment technology adapted to containerization on the cloud, this application adopts MMCO to achieve efficient allocation and optimized use of resources, especially for complex technology stacks such as ChatGPT. The improved multi-group coyote optimization (MMCO) algorithm can dynamically adjust virtual resources, meet service level agreements, and improve customer satisfaction; the pigeon flock heuristic optimization (MPIO) algorithm is combined with the fast adaptive feedback loop neural network (FARNN) to improve processing efficiency through task clustering and priority scheduling; the deep convolutional neural network (DCNN) monitors the task load and realizes real-time dynamic resource scheduling, which can significantly improve the performance of containerized applications in the cloud environment, while optimizing resource utilization, reducing costs, and ensuring the efficient operation and economy of the cloud computing environment, which is very important for supporting large-scale and dynamically changing cloud computing application scenarios. The overall technical architecture of the embodiment of the present application is the improved multi-group coyote optimization (MMCO) algorithm + pigeon flock heuristic optimization (MPIO) algorithm + deep convolutional neural network (DCNN) to work together to solve the complex scheduling problem of resource allocation in the model parameter adjustment technology adapted to containerization on the cloud, and ensure the efficient operation and economy of the cloud computing environment.

[0095] The application scenarios of this application include various cloud computing platforms, online service providers, and large-scale network applications. For example, when deploying complex machine learning models or online applications in the cloud, this technology can dynamically adjust model parameters to adapt to different business needs and changing environments. It is also suitable for microservice architectures in containerized environments to more effectively manage and optimize resources and improve the scalability and stability of applications.

[0096] This application can help enterprises improve the performance, resource utilization and flexibility of containerized applications, thereby reducing costs and improving competitiveness. By optimizing model parameters, enterprises can more efficiently deploy and manage applications in a cloud computing environment, reduce resource waste and operation and maintenance costs, and improve user experience and service quality. Therefore, this technology is of great significance for supporting large-scale, dynamically changing cloud computing application scenarios, and can bring long-term business value and competitive advantages to enterprises.

[0097] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0098] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in any form. Any technician familiar with the profession, without departing from the scope of the technical solution of the present invention and based on the technical essence of the present invention, may make any simple modifications, equivalent substitutions and improvements to the above embodiments, which are still within the protection scope of the technical solution of the present invention.

Claims

1. A parameter adjustment method for a containerized cloud model based on ChatGlm3, characterized in that: In the ChatGlm3 scenario, the parameter adjustment of the containerized model adapted to the containerized cloud uses an algorithm based on elastic resource management, dynamically monitors the performance indicators and resource usage of the containerized model, and automatically adjusts the allocation of container virtual resources according to preset rules or by learning historical data to improve customer service level agreements. It also ensures priority-based scheduling through the allocation of pre-virtualized CPUs, and implements priority-based dynamic scheduling through the task load monitoring mechanism.

2. The parameter adjustment method for the adapted containerized cloud model based on ChatGlm3 according to claim 1 is characterized in that: An elastic scaling strategy is formed through an elastic resource management algorithm to scale container virtual resources. When it is detected that the performance of the containerized model has degraded due to insufficient container virtual resources, the container virtual resource allocation is automatically increased. When it is detected that the containerized model is wasting due to insufficient use of container virtual resources, the container virtual resource allocation is automatically reduced.

3. The parameter adjustment method for the adapted containerized cloud model based on ChatGlm3 according to claim 1 or 2 is characterized in that: The described automatic adjustment of the allocation of container virtual resources according to preset rules or by learning historical data to improve customer service level protocol adopts a multi-group coyote optimization algorithm, which optimizes the parameters in the containerization model parameter adjustment by simulating the behavior of the coyote group, combining the multi-group and population update strategy, and realizes the solution of the multi-objective optimization problem.

4. The parameter adjustment method for the adapted containerized cloud model based on ChatGlm3 according to claim 3 is characterized in that: The multi-group coyote optimization algorithm specifically comprises the following steps: Initialize the position and speed of the coyote individual. The position of the individual represents the allocation ratio of the container's virtual resources, and the speed represents the rate of change of the container's virtual resources. Calculate the fitness of each individual, use the performance indicators under the current container virtual resource configuration to evaluate the performance of the containerized model, and obtain the fitness value of each individual to guide the update of the individual position; According to the current position of the best individual and the best group, the speed update formula is used to update the position and speed of each individual.

5. The parameter adjustment method for the adapted containerized cloud model based on ChatGlm3 according to claim 1 is characterized in that: The allocation of pre-virtualized CPUs ensures priority-based scheduling by using a pigeon flock heuristic optimization algorithm, which performs task clustering and a fast adaptive feedback loop neural network for the allocation of pre-virtualized CPUs to ensure priority-based scheduling.

6. The parameter adjustment method for the adapted containerized cloud model based on ChatGlm3 according to claim 5 is characterized in that: The pigeon flock heuristic optimization algorithm specifically comprises the following steps: The tasks are clustered using the pigeon flock heuristic optimization algorithm, similar tasks are assigned to the same group, and the foraging behavior of pigeon flocks is simulated to search for the optimal clustering solution; A fast adaptive feedback loop neural network model is designed and trained to predict the pre-virtualized CPU resources required for each task and achieve fast adaptive feedback.

7. The parameter adjustment method for the adapted containerized cloud model based on ChatGlm3 according to claim 1 is characterized in that: The task load monitoring mechanism is used to realize priority-based dynamic scheduling using a deep convolutional neural network, and the deep convolutional neural network is used to realize priority-based dynamic scheduling, thereby improving the resource allocation and scheduling efficiency of containerized applications in a cloud environment.

8. The parameter adjustment method for the adapted containerized cloud model based on ChatGlm3 according to claim 7 is characterized in that: The deep convolutional neural network specifically includes the following steps: Collect task load data of containerized applications in cloud environments and pre-process the data; A deep convolutional neural network model suitable for task load monitoring is designed. The features of the received task are taken as input, feature extraction and learning are performed, and the load status of the task is output. By training the deep convolutional neural network model and optimizing the loss function, the deep convolutional neural network model can accurately predict the load status of the task.

9. A containerized cloud-based model parameter adjustment system based on ChatGlm3, characterized in that: include: The container virtual resource dynamic adjustment module is used to automatically adjust the allocation of container virtual resources according to preset rules or by learning historical data to improve customer service level agreements; Task clustering priority scheduling module, used to ensure priority-based scheduling by allocating pre-virtualized CPUs; Real-time dynamic resource scheduling module, which is used to realize priority-based dynamic scheduling through task load monitoring mechanism; The container virtual resource dynamic adjustment module adopts a multi-group coyote optimization algorithm, the task clustering priority scheduling module adopts a pigeon flock heuristic optimization algorithm, and the real-time dynamic resource scheduling module adopts a deep convolutional neural network.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for adjusting parameters of the adapted containerized cloud model based on ChatGlm3 as described in any one of claims 1 to 8 is implemented.

Images