Computing power resource allocation method of intelligent computing center and model training method and device

Abstract

The invention provides a computing power resource allocation method of an intelligent computing center and a model training method and device.The allocation method comprises the steps that S1, operation data of all computing tasks running on the intelligent computing center at the current moment are obtained, and operation features are extracted from the operation data to obtain operation feature vectors; s2, inputting the operation feature vector into a computing power resource allocation large language model deployed in the intelligent computing center to obtain a computing power resource allocation strategy; the computing power resource allocation strategy represents a computing power resource type and a computing power resource quantity allocated to each computing task by the intelligent computing center; and S3, indicating the intelligent computing center to execute the computing power resource allocation strategy. According to the method, the computing power resource allocation strategy can be efficiently formulated, and the management efficiency of the computing power resources is improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of computing power infrastructure, and in particular to a computing power resource allocation method, model training method and device for an intelligent computing center. Background Art

[0002] With the development of artificial intelligence technology and computing power technology, the concept of intelligent computing center has emerged. "Intelligent computing center" refers to the use of large-scale heterogeneous computing power resources, including general computing power and intelligent computing power, mainly for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning and other scenarios) to provide the required computing power, data and algorithms. Intelligent computing center covers facilities, hardware, software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0003] The existing method for intelligent computing centers to allocate computing resources to computing tasks is to manually formulate computing resource allocation strategies based on experience. The efficiency of formulating computing resource allocation strategies is low, resulting in low efficiency in the management of computing resources. Summary of the invention

[0004] The embodiments of the present invention provide a computing power resource allocation method, a model training method and a device for an intelligent computing center to solve the problem that the existing method for allocating computing power resources to computing tasks in an intelligent computing center is to manually formulate a computing power resource allocation strategy based on experience, and the efficiency of formulating the computing power resource allocation strategy is low, resulting in low management efficiency of computing power resources.

[0005] In order to solve the above-mentioned technical problems, the present invention is achieved as follows:

[0006] In a first aspect, an embodiment of the present invention provides a computing resource allocation method for an intelligent computing center, comprising:

[0007] Step S1: obtaining the operation data of all computing tasks running on the intelligent computing center at the current moment, and extracting operation features from the operation data to obtain an operation feature vector;

[0008] Step S2: inputting the operation feature vector into the computing power resource allocation large language model deployed in the intelligent computing center to obtain a computing power resource allocation strategy; the computing power resource allocation strategy represents the type and amount of computing power resources allocated by the intelligent computing center to each computing task;

[0009] Step S3: instruct the intelligent computing center to execute the computing resource allocation strategy.

[0010] Optionally, after step S3, the following steps are included:

[0011] Step S4: monitor the actual performance data of the intelligent computing center; optimize the training feature set used to train the large language model for computing resource allocation according to the actual performance data, perform training iterations on the large language model for computing resource allocation with the optimized training feature set, and replace the large language model for computing resource allocation with the large language model for computing resource allocation after the training iteration on the intelligent computing center.

[0012] Optionally, the operating characteristics include at least one of the following: CPU usage, memory occupancy, I / O rate, network bandwidth, request arrival rate, and system delay.

[0013] In a second aspect, an embodiment of the present invention provides a method for training a large language model by allocating computing resources, including:

[0014] Step S01: acquiring historical operation data of multiple historical computing tasks that have been run on the intelligent computing center, extracting operation features from the historical operation data to obtain historical operation feature vectors; and acquiring historical computing power resource allocation strategies of the intelligent computing center corresponding to the multiple historical computing tasks one by one;

[0015] Step S02: for each of the historical operation feature vectors, label the historical computing resource allocation strategy of the same historical computing task until all the historical operation feature vectors are labeled; and the labeled historical operation feature vectors are combined to obtain a training feature set;

[0016] Step S03: using the training feature set to train the initial large language model to obtain a computing resource allocation large language model.

[0017] Optionally, the step S03 includes:

[0018] Step S031: After each round of training iteration, check whether the value of the loss function of the current initial large language model is less than a preset loss threshold to obtain a verification result;

[0019] Step S032: If the verification result indicates that the value of the loss function is greater than or equal to the loss threshold, proceed to the next round of training iteration;

[0020] Step S033: If the verification result indicates that the value of the loss function is less than the loss threshold, the next round of training iteration is terminated, and the current initial large language model is used as the computing resource allocation large language model.

[0021] In a third aspect, an embodiment of the present invention provides a computing resource allocation device for an intelligent computing center, including:

[0022] An acquisition module is used to acquire the operation data of all computing tasks running on the intelligent computing center at the current moment, and to extract operation features from the operation data to obtain an operation feature vector;

[0023] An allocation module is used to input the operation feature vector into a computing resource allocation large language model deployed in the intelligent computing center to obtain a computing resource allocation strategy; the computing resource allocation strategy represents the type and amount of computing resources allocated by the intelligent computing center to each computing task;

[0024] An execution module is used to instruct the intelligent computing center to execute the computing power resource allocation strategy.

[0025] In a fourth aspect, an embodiment of the present invention provides a training device for a large language model for allocating computing resources, including:

[0026] A data acquisition module is used to acquire historical operation data of multiple historical computing tasks that have been run on the intelligent computing center, extract operation features from the historical operation data to obtain historical operation feature vectors; and acquire historical computing power resource allocation strategies of the intelligent computing center corresponding to the multiple historical computing tasks one by one;

[0027] A training set generation module is used to label the historical computing resource allocation strategy of the same historical computing task for each of the historical operation feature vectors until all the historical operation feature vectors are labeled; and to combine the labeled historical operation feature vectors to obtain a training feature set;

[0028] The training module is used to train the initial large language model using the training feature set to obtain a computing resource allocation large language model.

[0029] In a fifth aspect, an embodiment of the present invention provides an electronic device, comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor; when the program or instruction is executed by the processor, the steps of the method for allocating computing power resources of an intelligent computing center as described in any one of the first aspects are implemented, or the steps of the method for training a large language model for allocating computing power resources as described in any one of the second aspects are implemented.

[0030] In a sixth aspect, an embodiment of the present invention provides a readable storage medium storing a program or instruction on the readable storage medium, which, when executed by a processor, implements the steps in the method for allocating computing power resources of an intelligent computing center as described in any one of the first aspects, or implements the steps in the method for training a large language model for allocating computing power resources as described in any one of the second aspects.

[0031] In the seventh aspect, an embodiment of the present invention provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the method for allocating computing power resources of an intelligent computing center as described in any one of the first aspects, or implement the steps of the method for training a large language model for allocating computing power resources as described in any one of the second aspects.

[0032] In an embodiment of the present invention, through step S1: obtaining the operation data of all computing tasks running on the intelligent computing center at the current moment, and extracting the operation features from the operation data to obtain the operation feature vector; step S2: inputting the operation feature vector into the large language model of computing power resource allocation deployed in the intelligent computing center to obtain the computing power resource allocation strategy; the computing power resource allocation strategy characterizes the type and amount of computing power resources allocated by the intelligent computing center to each computing task; step S3: instructing the intelligent computing center to execute the computing power resource allocation strategy. The embodiment of the present invention can efficiently formulate the computing power resource allocation strategy and improve the management efficiency of computing power resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0034] Figure 1 A schematic diagram of a flow chart of a computing resource allocation method for an intelligent computing center according to an embodiment of the present invention;

[0035] Figure 2 A flowchart of a method for training a large language model by allocating computing resources according to an embodiment of the present invention;

[0036] Figure 3 Schematic diagram of the architecture of the dynamic isolation and quota management system for computing resources;

[0037] Figure 4 This is a principle block diagram of a computing resource allocation device for an intelligent computing center according to an embodiment of the present invention;

[0038] Figure 5 A block diagram of the principle of a training device for a large language model for allocating computing resources according to an embodiment of the present invention;

[0039] Figure 6 The figure is a principle block diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] The terms "first", "second", etc. of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of one type, and the number of objects is not limited, for example, the first object can be one or more. In addition, "or" in the present application represents at least one of the connected objects. For example, "A or B" covers three schemes, namely, Scheme 1: including A but not including B; Scheme 2: including B but not including A; Scheme 3: including both A and B. The character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0042] In the technical solution of the present disclosure, words such as “connect”, “couple” or “connected” are not limited to physical or mechanical connections, but may include electrical connections.

[0043] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0044] First, the technical terms involved in the present invention are briefly explained below.

[0045] The "computing power" mentioned in the present invention is the ability of computer equipment or computing / data center to process information. It is the ability of computer hardware and software to work together to execute certain computing requirements. It is the computing power to achieve target result output by processing information data. It is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to the society through computing power infrastructure.

[0046] The "computing power" (Computational Power, CP) described in the present invention is the ability of a data center server to process data and output results. It is a comprehensive indicator to measure the computing power of a data center, including general computing power, super computing power and intelligent computing power. The commonly used unit of measurement is the number of floating point operations performed per second (FLOPS: Floating Point Operations Per Second, 1EFLOPS = 10^18FLOPS). The larger the value, the stronger the comprehensive computing power. According to calculations, 1EFLOPS is approximately the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs or 2 million mainstream notebooks. The calculation formula is: CP = CP 通用 +CP 智能 +CP 超级 .

[0047] The "Network Power" (NP) described in the present invention is a manifestation of the data transmission capability of computing facilities, including comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, etc. The carrying capacity involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling capabilities. In the embodiment of the present invention, the carrying capacity uses the memory bandwidth.

[0048] The "Storage Power" (SP) described in the present invention is the comprehensive ability of a data center in terms of data storage capacity, performance, safety and reliability, and green and low-carbon. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and server built-in storage devices. The commonly used unit of measurement for storage capacity is exabyte (EB, 1EB = 2^60bytes), and the commonly used unit of measurement for performance is the number of reads and writes per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB). The disaster recovery ratio is an important manifestation of safety and reliability.

[0049] The "computing power infrastructure" described in the present invention is a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity. It can realize centralized computing, storage, transmission and application of information, and presents characteristics such as diversity and ubiquity, intelligence and agility, security and reliability, and green and low-carbon.

[0050] The “computing power” mentioned in the present invention includes general computing power, intelligent computing power and super computing power.

[0051] The “general computing power” mentioned in the present invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0052] The "intelligent computing power" described in the present invention is a computing platform for large-scale deployment of special chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit) for various innovative applications of artificial intelligence, such as natural language processing and machine vision.

[0053] The "super computing power" mentioned in the present invention is mainly the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for calculations in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0054] The "intelligent computing center" described in the present invention refers to a facility that provides the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning scenarios) by using large-scale heterogeneous computing power resources, including general computing power (CPU: Central Processing Unit) and intelligent computing power (GPU: Graphics Processing Unit, FPGA: Field Programmable Gate Array, ASIC: Application Specific Integrated Circuit, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0055] The "computing resources" mentioned in the present invention refer to the technologies and facilities with information calculation, transmission, storage and application capabilities required for the development of the digital society, including but not limited to computing resources such as CPU (Central Processing Unit), GPU (Graphics Processing Unit), network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, as well as supporting and guarantee resources such as wind, fire, water and electricity.

[0056] The "large language model" described in the present invention refers to a large language model (LLM), which is a language model with a large parameter scale, designed to understand and generate human language, trained with a large amount of text data, and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc. Among them, the large language model can be a large language model such as GPT4, GLM-4 or Llama-3.

[0057] The embodiment of the present invention provides a computing resource allocation method for an intelligent computing center, see Figure 1 As shown, Figure 1 The flowchart of the computing resource allocation method of the intelligent computing center according to the embodiment of the present invention is as follows:

[0058] Step S1: obtaining the operation data of all computing tasks running on the intelligent computing center at the current moment, and extracting operation features from the operation data to obtain an operation feature vector;

[0059] Step S2: Input the running feature vector into the computing resource allocation large language model deployed in the intelligent computing center to obtain the computing resource allocation strategy; the computing resource allocation strategy represents the type and amount of computing resources allocated by the intelligent computing center to each computing task;

[0060] Step S3: Instruct the intelligent computing center to execute the computing resource allocation strategy.

[0061] In an embodiment of the present invention, a training method for a large language model for computing power resource allocation includes: obtaining historical operation data of multiple historical computing tasks that have been run on an intelligent computing center, extracting operation features from the historical operation data to obtain a historical operation feature vector; and obtaining a historical computing power resource allocation strategy of the intelligent computing center corresponding to multiple historical computing tasks one by one; for each historical operation feature vector, marking the historical computing power resource allocation strategy of the same historical computing task until all historical operation feature vectors are marked; and, combining the marked historical operation feature vectors to obtain a training feature set; using the training feature set to train the initial large language model to obtain a computing power resource allocation large language model. Thus, in step S2 of an embodiment of the present invention, the operation feature vector is input into the computing power resource allocation large language model deployed in the intelligent computing center, and the computing power resource allocation large language model can output the computing power resource allocation strategy. By applying the large language model to formulate computing power resource allocation strategies, the embodiment of the present invention improves the efficiency and accuracy of formulating computing power resource allocation strategies.

[0062] In an embodiment of the present invention, through step S1: obtaining the operation data of all computing tasks running on the intelligent computing center at the current moment, and extracting the operation features from the operation data to obtain the operation feature vector; step S2: inputting the operation feature vector into the large language model of computing power resource allocation deployed in the intelligent computing center to obtain the computing power resource allocation strategy; the computing power resource allocation strategy characterizes the type and amount of computing power resources allocated by the intelligent computing center to each computing task; step S3: instructing the intelligent computing center to execute the computing power resource allocation strategy. The embodiment of the present invention can efficiently formulate the computing power resource allocation strategy and improve the management efficiency of computing power resources.

[0063] It should be noted that the existing intelligent computing center allocates computing resources to computing tasks in a static resource allocation method. The static resource allocation method is usually based on predefined resource quotas or task priorities. When the computing task is started, the allocated resources remain unchanged and are not dynamically adjusted according to the load conditions, resulting in low utilization efficiency of computing resources.

[0064] Based on this, the embodiment of the present invention can execute step S1, step S2 and step S3 of the embodiment of the present invention once at a preset time interval, that is, every preset time interval, to dynamically adjust the computing resource allocation strategy according to actual conditions and improve the utilization efficiency of computing resources. The preset time interval, for example, 1 hour, can achieve hourly dynamic resource allocation tracking and adjustment; for another example, 30 minutes, can achieve half-hourly dynamic resource allocation tracking and adjustment, thereby improving the management accuracy of computing resources. In some optional embodiments, the preset time interval can be specifically set by the user according to actual needs, which cannot be enumerated here. It should be noted that the inability to enumerate should not be considered unclear.

[0065] In some embodiments of the present invention, optionally, after step S3, the following steps are included:

[0066] Step S4: monitor the actual performance data of the intelligent computing center; optimize the training feature set used to train the large language model for computing resource allocation according to the actual performance data, iterate the training of the large language model for computing resource allocation with the optimized training feature set, and replace the computing resource allocation large language model with the computing resource allocation large language model after the training iteration on the intelligent computing center.

[0067] In the embodiment of the present invention, the actual performance data of the intelligent computing center may include at least one of the following: task completion time, resource utilization, and service quality.

[0068] It can be understood that the task completion time is the completion time of the computing task. In conjunction with a specific example, the task completion time can be used to determine the completed tasks on the intelligent computing center. When there is a task completion on the intelligent computing center, the computing power resources freed up after the task is completed can be used for unfinished computing tasks. Therefore, in step S4, the training feature set for training the large language model for computing power resource allocation is optimized according to the actual performance data. In this example, it can be specifically to correct the historical computing power allocation strategy of the historical computing task annotations in the training feature set (increase the total amount of computing power resources in the historical computing power allocation strategy, or reduce the resource allocation amount for historical computing tasks equivalent to completed tasks), and then through the subsequent training feature set in step S4, the computing power resource allocation large language model is trained and iterated incremental learning, so that the computing power resource allocation strategy output by the large language model for computing power resource allocation can adapt to the situation where there are computing power resources freed up after the completion of tasks on the intelligent computing center, that is, better fit the application scenario and improve the utilization rate of computing power resources.

[0069] It can be understood that if the resource utilization rate is low, it directly indicates that there are excess computing resources that can be used for computing tasks. Therefore, in step S4, the training feature set used to train the large language model for computing resource allocation is optimized according to the actual performance data. Specifically, the historical computing power allocation strategy annotated with the historical computing tasks in the training feature set can be corrected (the total amount of computing power resources in the historical computing power allocation strategy, or the amount of resource allocation for historical computing tasks equivalent to completed tasks is reduced). Then, through the subsequent incremental learning of the large language model for computing resource allocation with the optimized training feature set in step S4, the computing power resource allocation strategy output by the large language model for computing resource allocation can adapt to the low resource utilization in the intelligent computing center, that is, better fit the application scenario and improve the utilization of computing power resources.

[0070] It should be noted that the service quality can be the evaluation of the computing power resource allocation strategy fed back by the user, such as whether the computing efficiency of a certain A computing task is satisfactory. Exemplarily, if not satisfied, the training feature set used to train the computing power resource allocation large language model according to the actual performance data in step S4 can be specifically to correct the historical computing power allocation strategy annotated by the historical computing tasks in the training feature set (increase the resource allocation amount for the A computing task in the historical computing power allocation strategy), and then through the subsequent incremental learning of the computing power resource allocation large language model with the optimized training feature set in step S4, the computing power resource allocation strategy output by the computing power resource allocation large language model can increase the computing power resource amount of the A computing task, improve the computing efficiency of the computing task, and coordinate the computing power resource allocation of other computing tasks to ensure the high resource utilization of the overall computing power resources.

[0071] In some embodiments of the present invention, optionally, the operating characteristics include at least one of the following: CPU usage, memory occupancy, I / O rate, network bandwidth, request arrival rate, and system delay.

[0072] The above operating characteristics can accurately characterize the operating status of computing tasks in the intelligent computing center, ensure a computing resource allocation strategy with high resource utilization, and improve computing resource utilization.

[0073] It should be noted that in some optional embodiments of the present invention, the operating characteristics include: CPU usage, memory usage, I / O rate, network bandwidth, request arrival rate, and system latency. The above operating characteristics can accurately characterize the operating status of computing tasks on the intelligent computing center, ensure a high resource utilization computing resource allocation strategy, and improve computing resource utilization.

[0074] The embodiment of the present invention provides a method for training a large language model by allocating computing resources. Figure 2 As shown, Figure 2 The flowchart of the method for training a large language model by allocating computing power resources according to an embodiment of the present invention is as follows. The method for training a large language model by allocating computing power resources includes:

[0075] Step S01: obtaining historical operation data of multiple historical computing tasks that have been run on the intelligent computing center, extracting operation features from the historical operation data to obtain historical operation feature vectors; and obtaining historical computing power resource allocation strategies of the intelligent computing center corresponding to the multiple historical computing tasks one by one;

[0076] Step S02: for each historical operation feature vector, annotate the historical computing resource allocation strategy of the same historical computing task until all historical operation feature vectors are annotated; and combine the annotated historical operation feature vectors to obtain a training feature set;

[0077] Step S03: Use the training feature set to train the initial large language model to obtain a computing resource allocation large language model.

[0078] In some embodiments of the present invention, the operation feature may optionally include at least one of the following: CPU usage, memory usage, I / O rate, network bandwidth, request arrival rate, and system latency. Based on the above operation features, the historical operation feature vector obtained by extracting the operation features from the historical operation data can accurately characterize the characteristics of the historical computing tasks, ensuring that the initial large language model accurately learns the historical computing tasks.

[0079] In an embodiment of the present invention, through step S01: historical operation data of multiple historical computing tasks that have been run on the intelligent computing center are obtained, and operation features are extracted from the historical operation data to obtain a historical operation feature vector; and, the historical computing power resource allocation strategy of the intelligent computing center corresponding to multiple historical computing tasks is obtained; step S02: for each historical operation feature vector, the historical computing power resource allocation strategy of the same historical computing task is marked until all historical operation feature vectors are marked; and the marked historical operation feature vectors are combined to obtain a training feature set; step S03: the initial large language model is trained using the training feature set to obtain a computing power resource allocation large language model. The computing power resource allocation large language model trained by the embodiment of the present invention can accurately output the computing power resource allocation strategy, thereby improving the efficiency of formulating the computing power resource allocation strategy.

[0080] In some embodiments of the present invention, optionally, step S03 includes:

[0081] Step S031: After each round of training iteration, check whether the value of the loss function of the current initial large language model is less than a preset loss threshold, and obtain a verification result;

[0082] Step S032: If the verification result indicates that the value of the loss function is greater than or equal to the loss threshold, proceed to the next round of training iteration;

[0083] Step S033: If the verification result indicates that the value of the loss function is less than the loss threshold, the next round of training iteration is terminated, and the current initial large language model is used as the computing resource to allocate the large language model.

[0084] It should be noted that the loss threshold can be determined by the user according to the user's actual needs.

[0085] In the embodiment of the present invention, the iterative training of the model is controlled by using steps S031, S032 and S033 based on the loss threshold, so that the training can be terminated in time after the demand is met, thereby avoiding the waste of training resources and improving the training efficiency.

[0086] In some embodiments of the present invention, optionally, the expression of the loss function L(θ) is as follows:

[0087]

[0088] Among them, A opt (t k ) is time t k The optimal resource allocation strategy; l is the loss function, such as mean square error; f(X(t k );θ) is the expression of the large language model for computing power resource allocation; θ is the model parameter of the large language model for computing power resource allocation.

[0089] The following is a description with reference to specific embodiments.

[0090] See also Figure 3 As shown, Figure 3 The overall architecture of the computing power resource dynamic isolation and quota management system (hereinafter referred to as the management system, or system) based on the intelligent computing center is illustrated. The system realizes real-time monitoring, intelligent prediction, dynamic management and continuous optimization of computing power resources through the collaboration of multiple modules. The following is a detailed description of each module:

[0091] 1. Data Collection Module (DCM)

[0092] This module is responsible for real-time monitoring of computing resource usage and collecting raw data D(t). The collected data includes status information of system resources such as CPU usage, memory usage, disk I / O, network traffic, etc. DCM is the starting point of the system and provides raw data for subsequent feature extraction and prediction processes.

[0093] 2. Feature Extraction Module (FEM)

[0094] FEM preprocesses the raw data D(t) collected by DCM, extracts key features, and constructs feature vector X(t). Features include CPU usage, memory usage, I / O rate, network bandwidth, request arrival rate, and system latency. These features serve as input variables to help the large language model understand the current resource usage of the system.

[0095] 3. Large Language Model (MLM, equivalent to the large language model for computing resource allocation in the embodiment of the present invention)

[0096] This module uses a supervised learning algorithm to process the feature vector X(t) to predict the resource allocation strategy A(t). After model training, MLM can predict the optimal resource allocation strategy based on the real-time feature vector. Through prediction, MLM can quickly determine the current system load demand and optimize resource allocation to ensure system performance.

[0097] 4. Resource Management Module (RMM)

[0098] RMM dynamically adjusts the isolation parameters and quota allocation of computing resources based on the resource allocation strategy A(t) predicted by MLM. Through RMM's adjustment, the system can flexibly allocate resources for different tasks, avoid resource competition and waste, and ensure the rational use of resources and performance guarantee.

[0099] 5. Feedback Module (FBM)

[0100] FBM is responsible for monitoring the system performance indicators P(t) after the strategy is executed, such as task completion time, resource utilization, and service quality. It feeds back the actual system performance data to MLM so that MLM can continuously optimize the model. This feedback loop enables the system to adapt to changes in load and improve prediction accuracy and the effectiveness of resource allocation.

[0101] The data flow starts from DCM, extracts features through FEM, predicts strategies through MLM, and then passes to RMM to execute resource management strategies, and finally to FBM for feedback. FBM data is fed back to MLM, forming a closed loop, allowing the system to continuously optimize itself and adapt to changing load requirements.

[0102] Suppose there is a set of resources R = {r1, r2, ..., r n} and a set of tasks or applications T = {t1, t2, …, t m The goal is to find a resource allocation strategy A(t) = {a ij (t)}, so that the system utility U(t) is maximized.

[0103] The expression of the objective function of MLM training is as follows:

[0104]

[0105] Among them, u i It is the task i The utility function of task t represents its performance or satisfaction. This function depends on task t i The number of resources allocated at time t. The form of the utility function can vary according to the characteristics of different tasks, and is usually used to represent the efficiency or completion of tasks under different resource allocations. For example, for delay-sensitive tasks, the utility function may be proportional to the number of resources, while for tasks with high fault tolerance, the utility function may be smoother. M is the number of tasks.

[0106] The objective function is to maximize the overall utility U(t) of the system by selecting the optimal resource allocation strategy A(t).

[0107] Constraints for MLM model training:

[0108] Total resource limit:

[0109]

[0110] Among them, C j It is a resource j total capacity.

[0111] Resource non-negativity:

[0112]

[0113] The system has demonstrated significant technical effects in computing resource management, including:

[0114] Dynamic Adaptation

[0115] This system uses a large language model for computing resource allocation to predict resource requirements in real time, and adjusts resource allocation strategies in real time according to dynamic changes in actual load. Compared with traditional static or rule-based resource allocation schemes, this system can flexibly adapt to various sudden loads and fluctuating environments. For example, in cloud computing and big data processing scenarios, when user demand or workload increases sharply, the system can quickly identify load changes and adjust resource quotas accordingly to avoid performance degradation due to resource shortages. This dynamic adaptability makes the system perform better in complex environments with high load and high concurrency.

[0116] Improve efficiency

[0117] This system improves overall resource utilization by optimizing resource allocation strategies. The large language model for computing resource allocation can accurately predict load demand by analyzing historical data and real-time features, avoiding waste caused by over-allocation or idle resources in traditional methods. Especially in data centers and cloud computing environments with high resource costs, the system of the present invention effectively reduces unnecessary resource allocation, significantly reduces the operating cost of the system, and ensures that resources are optimally utilized under various workloads.

[0118] Guaranteed performance

[0119] The present invention can not only dynamically allocate resources, but also ensure that the system maintains stable performance under high-load scenarios. The system can meet the performance requirements of different tasks, including low latency, high throughput, and real-time response. For tasks such as big data analysis, streaming media distribution, and online services that require real-time processing, the system avoids resource bottlenecks and performance degradation through intelligent resource scheduling to ensure user experience. Through continuously optimized resource allocation strategies, the overall performance of the system has been significantly improved.

[0120] Fair distribution

[0121] In the resource allocation process, this system introduces a fairness constraint mechanism to ensure the reasonable allocation of resources. This mechanism automatically identifies the task priority and its resource requirements through the large language model algorithm for computing resource allocation, realizes fair resource allocation in a multi-task and multi-user environment, and avoids the problem of unfair resource allocation. For example, on a multi-user shared cloud computing platform, the system can allocate resources according to the actual needs of user tasks, ensuring that low-priority tasks will not be completely ignored due to the resource occupation of high-priority tasks, thereby improving user satisfaction and the overall fairness of the system.

[0122] Adaptive Optimization

[0123] Through the feedback mechanism, the system will continuously monitor the effect of the resource allocation strategy after executing it, and feed back the operating data to the computing resource allocation large language model for optimization. This adaptive optimization capability enables the system to automatically adjust when the load pattern changes, thereby continuously improving the accuracy and real-time performance of the resource allocation strategy. Especially in an environment with frequent demand fluctuations, the system can quickly learn new load characteristics and update model parameters to make resource allocation more in line with actual needs. This adaptability greatly improves the long-term stability of the system and the intelligence level of resource scheduling.

[0124] In the following application scenarios, the dynamic isolation and quota management system for computing resources demonstrates significant applicability and technical advantages:

[0125] A) Cloud computing and big data platforms

[0126] In cloud computing and big data platforms, resource requirements are often highly dynamic. As the number of visits and computing tasks of different users continue to change, the load will also show the characteristics of high-frequency fluctuations. Traditional static allocation strategies are difficult to meet these complex requirements, and the system of the present invention can automatically adjust the resource allocation strategy through real-time monitoring and prediction, so as to quickly expand computing resources during peak periods of resource demand and ensure the consistency of user experience. This on-demand expansion feature makes the system very suitable for elastic computing scenarios, and can effectively optimize resource allocation in both public and private cloud environments.

[0127] B) Online Services and Real-time Applications

[0128] In online services (such as e-commerce, video streaming, game servers, etc.), the amount of user requests varies significantly over time, especially during promotional activities, traffic peaks or emergencies, resource demand can fluctuate dramatically in a short period of time. Through the system of the present invention, sufficient computing resources can be reserved in advance for high-load scenarios based on predicted traffic changes, avoiding performance degradation or service interruption caused by insufficient resources, thereby ensuring the smooth experience of users and the stability of services.

[0129] C) Internet of Things and Edge Computing

[0130] In the Internet of Things and edge computing, the number of devices is huge and widely distributed, and the resource requirements of different nodes vary significantly. This system can perform fine-grained resource management based on the load status of each node, and ensure that devices operate efficiently under limited resource conditions through dynamic isolation and quota adjustment. For example, in smart cities and industrial Internet of Things, the system can provide accurate computing resources for real-time data processing and monitoring applications based on the actual working conditions of the equipment, thereby improving the overall performance of the network.

[0131] D) Financial trading system

[0132] Financial trading systems require extremely high computing speed and stability, and resource requirements usually change with market fluctuations. For example, when trading volume increases dramatically, the system needs to respond quickly to handle a large number of trading requests. The system of the present invention can predict resource requirements based on real-time trading traffic, ensure sufficient computing resources during peak trading periods, reduce delays, and ensure the real-time nature of transactions and the security of the system. This intelligent resource management strategy can effectively support financial applications such as securities trading, futures markets, and cryptocurrency trading.

[0133] D) Intelligent manufacturing and industrial control

[0134] In intelligent manufacturing and industrial control systems, the equipment on the production line may require a large amount of real-time data processing, and the load demand changes with the production tasks. The system of the present invention can adjust resource allocation in real time according to the changes in production tasks, making the production process more efficient and stable. For example, in scenarios such as automobile manufacturing and electronic product production, the system can flexibly allocate resources according to the production rhythm and task priority, thereby ensuring the maximization of production efficiency and timely responding to the dynamic needs of equipment.

[0135] E) Emergency Response and Disaster Recovery

[0136] In emergency response scenarios, such as natural disasters and emergencies, the system load often increases dramatically in a short period of time. For example, during a disaster, rescue organizations and government departments may use system resources on a large scale for tasks such as data analysis and emergency dispatch. The system of the present invention can quickly identify peak demand in this case, dynamically expand resources and perform priority management, ensure the availability and reliability of key applications under high load, and provide technical support for disaster response.

[0137] F) Autonomous driving and intelligent transportation

[0138] In autonomous driving and intelligent traffic management, computing load and resource requirements may fluctuate dramatically with road conditions, traffic flow and traffic events. This system can allocate more resources to key tasks during peak hours and congested roads, thereby ensuring the real-time performance of the system and the safety of vehicles. For example, in an intelligent traffic signal system, the system of the present invention can dynamically allocate resources to process sudden traffic accident data or adjust the traffic light schedule to achieve intelligent control of traffic flow.

[0139] G) Medical health monitoring system

[0140] In medical health monitoring and diagnosis systems, real-time and accuracy are key. For example, in telemedicine and emergency scenarios, medical equipment needs to continuously monitor the patient's health data, and this data load may increase dramatically as the patient's condition changes. The system of the present invention can dynamically adjust resources according to the patient's condition to ensure the timeliness of medical data processing, thereby helping doctors make decisions in a short time and improving the efficiency and accuracy of diagnosis.

[0141] The embodiment of the present invention provides a computing resource allocation device for an intelligent computing center, see Figure 4 As shown, Figure 4 This is a principle block diagram of a computing resource allocation device for an intelligent computing center according to an embodiment of the present invention. The computing resource allocation device 40 for an intelligent computing center includes:

[0142] The acquisition module 41 is used to acquire the operation data of all the computing tasks running on the intelligent computing center at the current moment, and extract the operation features from the operation data to obtain the operation feature vector;

[0143] An allocation module 42 is used to input the operation feature vector into a computing resource allocation language model deployed in the intelligent computing center to obtain a computing resource allocation strategy; the computing resource allocation strategy represents the type and amount of computing resources allocated by the intelligent computing center to each computing task;

[0144] The execution module 43 is used to instruct the intelligent computing center to execute the computing resource allocation strategy.

[0145] In some embodiments of the present invention, optionally, the execution module 43 is also used to monitor actual performance data of the intelligent computing center; optimize the training feature set used to train the computing power resource allocation large language model according to the actual performance data, perform training iterations on the computing power resource allocation large language model with the optimized training feature set, and replace the computing power resource allocation large language model with the computing power resource allocation large language model after the training iteration on the intelligent computing center.

[0146] In some embodiments of the present invention, optionally, the operating characteristics include at least one of the following: CPU usage, memory occupancy, I / O rate, network bandwidth, request arrival rate, and system delay.

[0147] The computing resource allocation device of the intelligent computing center provided in the embodiment of the present application can achieve Figure 1 The various processes implemented in the method embodiment achieve the same technical effect and will not be described again here to avoid repetition.

[0148] The embodiment of the present invention provides a training device for a large language model by allocating computing resources. Figure 5As shown, Figure 5 This is a principle block diagram of a training device for a large language model for computing resource allocation according to an embodiment of the present invention. The training device 50 for a large language model for computing resource allocation includes:

[0149] The data acquisition module 51 is used to acquire historical operation data of multiple historical computing tasks that have been run on the intelligent computing center, extract operation features from the historical operation data to obtain historical operation feature vectors; and acquire historical computing resource allocation strategies of the intelligent computing center corresponding to the multiple historical computing tasks.

[0150] The training set generation module 52 is used to label the historical computing resource allocation strategy of the same historical computing task for each of the historical operation feature vectors until all the historical operation feature vectors are labeled; and to combine the labeled historical operation feature vectors to obtain a training feature set;

[0151] The training module 53 is used to train the initial large language model using the training feature set to obtain a computing resource allocation large language model.

[0152] In some embodiments of the present invention, optionally,

[0153] The training module 53 is further used to verify whether the value of the loss function of the current initial large language model is less than a preset loss threshold after each round of training iteration, and obtain a verification result;

[0154] The training module 53 is further configured to continue the next round of training iteration if the verification result indicates that the value of the loss function is greater than or equal to the loss threshold;

[0155] The training module 53 is further configured to terminate the next round of training iterations if the verification result indicates that the value of the loss function is less than the loss threshold, and use the current initial large language model as the computing resource allocation large language model.

[0156] The computing resource allocation large language model training device provided in the embodiment of the present application can achieve Figure 2 The various processes implemented in the method embodiment achieve the same technical effect and will not be described again here to avoid repetition.

[0157] The embodiment of the present invention provides an electronic device 60, see Figure 6 As shown, Figure 6It is a principle block diagram of an electronic device 60 according to an embodiment of the present invention, including a processor 61, a memory 62, and a program or instruction stored in the memory 62 and executable on the processor 61. When the program or instruction is executed by the processor, the steps of the method for allocating computing power resources of an intelligent computing center described in any one of the embodiments of the present invention are implemented, or the steps of the method for training a large language model for allocating computing power resources described in any one of the embodiments of the present invention are implemented.

[0158] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of an embodiment of a method for allocating computing power resources for an intelligent computing center such as any of the above-mentioned items are implemented, or the various processes of an embodiment of a method for allocating computing power resources for training a large language model such as any of the above-mentioned items are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0159] The readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc. In some examples, the readable storage medium may be a non-transitory readable storage medium.

[0160] An embodiment of the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the various processes of the embodiments of the computing power resource allocation method for an intelligent computing center described above, or implement the various processes of the embodiments of the computing power resource allocation method for training a large language model described above, and can achieve the same technical effect. To avoid repetition, they will not be described here.

[0161] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A computing resource allocation method for an intelligent computing center, characterized in that: include: Step S1: obtaining the operation data of all computing tasks running on the intelligent computing center at the current moment, and extracting operation features from the operation data to obtain an operation feature vector; Step S2: inputting the operation feature vector into a large language model for computing resource allocation deployed in the intelligent computing center to obtain a computing resource allocation strategy; The computing power resource allocation strategy represents the type and amount of computing power resources allocated by the intelligent computing center to each computing task; Step S3: instruct the intelligent computing center to execute the computing resource allocation strategy.

2. The computing resource allocation method of the intelligent computing center according to claim 1, characterized in that: After step S3, the following steps are included: Step S4: monitor the actual performance data of the intelligent computing center; optimize the training feature set used to train the large language model for computing resource allocation according to the actual performance data, perform training iterations on the large language model for computing resource allocation with the optimized training feature set, and replace the large language model for computing resource allocation with the large language model for computing resource allocation after the training iteration on the intelligent computing center.

3. The computing resource allocation method of the intelligent computing center according to claim 1, characterized in that: The operation characteristics include at least one of the following: CPU usage, memory occupancy, I / O rate, network bandwidth, request arrival rate, and system delay.

4. A method for training a large language model by allocating computing resources, characterized in that: include: Step S01: acquiring historical operation data of multiple historical computing tasks that have been run on the intelligent computing center, extracting operation features from the historical operation data to obtain historical operation feature vectors; and acquiring historical computing power resource allocation strategies of the intelligent computing center corresponding to the multiple historical computing tasks one by one; Step S02: for each of the historical operation feature vectors, label the historical computing resource allocation strategy of the same historical computing task until all the historical operation feature vectors are labeled; and the labeled historical operation feature vectors are combined to obtain a training feature set; Step S03: using the training feature set to train the initial large language model to obtain a computing resource allocation large language model.

5. The method for training a large language model for computing resource allocation according to claim 4, characterized in that: The step S03 comprises: Step S031: After each round of training iteration, check whether the value of the loss function of the current initial large language model is less than a preset loss threshold to obtain a verification result; Step S032: If the verification result indicates that the value of the loss function is greater than or equal to the loss threshold, proceed to the next round of training iteration; Step S033: If the verification result indicates that the value of the loss function is less than the loss threshold, the next round of training iteration is terminated, and the current initial large language model is used as the computing resource allocation large language model.

6. A computing resource allocation device for an intelligent computing center, characterized in that: include: An acquisition module is used to acquire the operation data of all computing tasks running on the intelligent computing center at the current moment, and to extract operation features from the operation data to obtain an operation feature vector; An allocation module, configured to input the operation feature vector into a computing resource allocation large language model deployed in the intelligent computing center to obtain a computing resource allocation strategy; The computing power resource allocation strategy represents the type and amount of computing power resources allocated by the intelligent computing center to each computing task; An execution module is used to instruct the intelligent computing center to execute the computing power resource allocation strategy.

7. A training device for a large language model with computing power resource allocation, characterized in that: include: A data acquisition module is used to acquire historical operation data of multiple historical computing tasks that have been run on the intelligent computing center, extract operation features from the historical operation data to obtain historical operation feature vectors; and acquire historical computing power resource allocation strategies of the intelligent computing center corresponding to the multiple historical computing tasks one by one; A training set generation module is used to label the historical computing resource allocation strategy of the same historical computing task for each of the historical operation feature vectors until all the historical operation feature vectors are labeled; and to combine the labeled historical operation feature vectors to obtain a training feature set; The training module is used to train the initial large language model using the training feature set to obtain a computing resource allocation large language model.

8. An electronic device, characterized in that: It includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, it implements the steps in the method for allocating computing power resources of an intelligent computing center as described in any one of claims 1 to 3, or implements the steps in the method for training a large language model for allocating computing power resources as described in any one of claims 4 to 5.

9. A readable storage medium, characterized in that: The readable storage medium stores programs or instructions, which, when executed by the processor, implement the steps in the method for allocating computing power resources of an intelligent computing center as described in any one of claims 1 to 3, or implement the steps in the method for training a large language model for allocating computing power resources as described in any one of claims 4 to 5.

10. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps in the method for allocating computing resources of an intelligent computing center as described in any one of claims 1 to 3, or implement the steps in the method for training a large language model for allocating computing resources as described in any one of claims 4 to 5.

Images