Computing power resource scheduling method and device of intelligent computing center

Abstract

The invention provides a computing power resource scheduling method and device for an intelligent computing center, and belongs to the technical field of intelligent computing centers, intelligent computing centers and computing power infrastructure, and the method comprises the steps: monitoring and obtaining the current sampling time required by the sampling process in the current round group relative strategy optimization training process of a model, and obtaining the current sampling time required by the sampling process; obtaining the current evaluation time required by the evaluation process; and based on the current sampling time consumption and the current evaluation time consumption, adjusting a computing power resource scale for the next round of training process of the model. In the method, the sampling time consumption and the evaluation time consumption of the current round of the model can reflect the calculation power resource utilization condition of the current round of sampling and evaluation tasks executed by the calculation power scale resources of the current round of the model, so that the calculation power scale resources of the current round of the model can be calculated according to the calculation power resource utilization condition of the current round of the model. And the computing power resource scale of the next round of the model is adjusted, so that the utilization rate of the computing power resources can be improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent computing centers, intelligent computing centers, and computing power infrastructure, and particularly relates to a method and device for scheduling computing power resources of an intelligent computing center. Background Art

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "intelligent computing centers" have emerged as the times require.

[0003] An "intelligent computing center" refers to a facility that mainly provides the required computing power, data, and algorithms for artificial intelligence applications (such as scenarios of artificial intelligence deep learning model development, model training, and model inference) by using large-scale heterogeneous computing power resources, including general computing power and intelligent computing power. The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enabling.

[0004] The "intelligent computing center" includes, but is not limited to, the "intelligent computing center".

[0005] An "intelligent computing center", that is, an artificial intelligence computing center, is a type of computing power infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications based on artificial intelligence theory and using an artificial intelligence computing architecture.

[0006] "Computing power" is the core of "intelligent computing centers" and "intelligent computing centers", which is the ability of computer devices or computing / data centers to process information, the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement, the computing ability to achieve the output of target results by processing information data, and a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, and mainly provides services to society through computing power infrastructure.

[0007] Currently, during the training process of Group Relative Policy Optimization (GRPO) of a model in an intelligent computing center, it is necessary to sample the model results at a high sampling frequency, and the sampling results need to be evaluated through a reward method or a reward model. The above sampling and evaluation processes need to occupy most of the entire training process, and during the training process, the time consumption of both will change. In order to improve the training speed, in the prior art, generally, an excessive amount of computing power resources will be configured for the sampling and evaluation processes at the beginning of the task. If the problem complexity surges during the training process, there will also be a situation of insufficient computing power, which also leads to low utilization rate of computing power resources. Summary of the Invention

[0008] The present invention provides a method and device for scheduling computing power resources in an intelligent computing center, which are used to solve the problem of low utilization rate of computing power resources during model training in the intelligent computing center.

[0009] To solve the above technical problems, the present invention is implemented as follows:

[0010] In a first aspect, the present invention provides a method for scheduling computing power resources in an intelligent computing center, including:

[0011] Step S1: During the population relative policy optimization training process of the current round of the model, monitor and obtain the current sampling time required for the sampling process, and obtain the current evaluation time required for the evaluation process;

[0012] Step S2: Based on the current sampling time and the current evaluation time, adjust the scale of computing power resources for the next round of training process of the model.

[0013] Optionally, step S2 includes:

[0014] Step S21: Determine the current time by adding the current sampling time and the current evaluation time;

[0015] Step S22: Compare the current time with a preset time, and adjust the scale of computing power resources for the next round of training process of the model according to the obtained comparison result.

[0016] Optionally, step S2 includes:

[0017] Step S23: Obtain the current scale of computing power resources corresponding to the current round of training process of the model, and obtain the historical scale of computing power resources, the historical sampling time corresponding to the sampling process, and the historical evaluation time corresponding to the evaluation process of the previous round of training process of the model;

[0018] Step S24: Based on the current scale of computing power resources, the historical scale of computing power resources, the historical sampling time, and the historical evaluation time, determine the theoretical time, where the theoretical time is the time required to process the sampling and evaluation of the previous round of training process of the model under the condition of the highest utilization rate of the current scale of computing power resources;

[0019] Step S25: Based on the theoretical time, the current sampling time, and the current evaluation time, adjust the scale of computing power resources for the next round of training process of the model.

[0020] Optionally, step S25 includes:

[0021] Step S251: Determine the current time by adding the current sampling time and the current evaluation time;

[0022] Step S252: Determine a target time-consuming range based on the theoretical time-consuming and preset parameters;

[0023] Step S253: Adjust the scale of computing power resources for the next round of training process of the model based on the current time-consuming and the target time-consuming range.

[0024] Optionally, step S2 includes:

[0025] Step S26: Obtain the historical sampling time-consuming corresponding to the sampling process and the historical evaluation time-consuming corresponding to the evaluation process in the previous round of training process of the model;

[0026] Step S27: Adjust the scale of computing power resources for the next round of training process of the model based on the historical sampling time-consuming, the historical evaluation time-consuming, the current sampling time-consuming, and the current evaluation time-consuming.

[0027] Optionally, step S27 includes:

[0028] Step S271: Determine the sum of the historical sampling time-consuming and the historical evaluation time-consuming as the historical time-consuming;

[0029] Step S272: Determine the sum of the current sampling time-consuming and the current evaluation time-consuming as the current time-consuming;

[0030] Step S273: Increase the scale of computing power resources for the next round of training process of the model when the historical time-consuming is less than the current time-consuming; or decrease the scale of computing power resources for the next round of training process of the model when the current time-consuming is less than the historical time-consuming.

[0031] In a second aspect, the present invention provides a computing power resource scheduling device for an intelligent computing center, including:

[0032] A monitoring module, configured to monitor and obtain the current sampling time-consuming required for the sampling process and obtain the current evaluation time-consuming required for the evaluation process during the population relative policy optimization training process of the current round of the model;

[0033] A scheduling module, configured to adjust the scale of computing power resources for the next round of training process of the model based on the current sampling time-consuming and the current evaluation time-consuming.

[0034] In a third aspect, the present invention provides a server, including: a processor, a memory, and a program stored on the memory and executable on the processor, and when the program is executed by the processor, the steps of the computing power resource scheduling method for the intelligent computing center as described in the first aspect above are implemented.

[0035] Fourthly, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the computing power resource scheduling method of the intelligent computing center as described in the first aspect above are implemented.

[0036] Fifthly, the present invention provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, the steps of the computing power resource scheduling method of the intelligent computing center as described in the first aspect above are implemented.

[0037] In the present invention, during the population relative policy optimization training process of the current round of the model, the current sampling time required for the sampling process is monitored and obtained, and the current evaluation time required for the evaluation process is obtained; based on the current sampling time and the current evaluation time, the computing power resource scale for the next round of training process of the model is adjusted. Through the above method, the sampling time and evaluation time of the current round of the model can reflect the utilization of the computing power resources for the sampling and evaluation tasks of the current round of the computing power scale resources of the model. Therefore, according to the utilization of the computing power resources of the current round of the model, adjusting the computing power resource scale of the next round of the model can improve the utilization rate of the computing power resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0039] Figure 1 is a schematic flowchart of the computing power resource scheduling method of the intelligent computing center provided by the embodiment of the present invention;

[0040] Figure 2 is a schematic flowchart of GPRO provided by the embodiment of the present invention;

[0041] Figure 3 is a schematic structural diagram of the computing power resource scheduling device of the intelligent computing center provided by the embodiment of the present invention;

[0042] Figure 4 is a schematic structural diagram of the electronic device provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts fall within the protection scope of the present invention.

[0044] First, the technical terms related to the present invention will be briefly described below.

[0045] The "computing power" referred to in the present invention means: the ability of a computer device or a computing / data center to process information, the ability of computer hardware and software to cooperate to jointly execute a certain computing requirement, the computing ability to achieve the output of a target result by processing information data, a new type of productive force integrating information computing power, network carrying capacity, and data storage capacity, and mainly providing services to society through computing power infrastructure.

[0046] The "computational power" (Computational Power, CP) referred to in the present invention means: the ability of a data center server to process data and achieve result output, a comprehensive index to measure the computing ability of a data center, including general computing ability, supercomputing ability, and intelligent computing ability. The commonly used measurement unit is the number of floating-point operations per second (FLOPS, 1EFLOPS = 10^18 FLOPS), and the larger the value, the stronger the comprehensive computing ability. It is estimated that 1EFLOPS is approximately the computing power output of 5 Tianhe 2A or 500,000 mainstream server CPUs or 2 million mainstream laptops. The calculation formula is: CP = CP_general + CP_intelligent + CP_super.

[0047] The "network power" (Network Power, NP) referred to in the present invention means: the manifestation of the data transmission ability of computing power facilities, a comprehensive ability including network architecture, network bandwidth, transmission delay, intelligent management and scheduling, etc., involving network transmission inside and between data centers, and a comprehensive index to measure network transmission scheduling ability.

[0048] The "storage power" (Storage Power, SP) referred to in the present invention means: the comprehensive ability of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon, a comprehensive index to measure the data storage ability of a data center, including external storage devices such as storage arrays and server internal storage devices. The commonly used measurement unit for storage capacity is exabyte (EB, 1EB = 2^60 bytes), the commonly used measurement unit for performance is the number of read and write operations per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB), and the disaster recovery ratio is an important manifestation of security and reliability.

[0049] The "computing power infrastructure" described in the present invention refers to a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity, and can realize the centralized computing, storage, transmission, and application of information.

[0050] The "new type of information infrastructure" described in the present invention mainly includes network infrastructures such as 5G networks, fiber broadband networks, backbone networks, international communication networks, and satellite Internet, computing power infrastructures such as data centers, general computing power centers, intelligent computing centers, and supercomputing centers, and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0051] The "computing power" described in the present invention includes general computing power, intelligent computing power, and super computing power.

[0052] The "general computing power" described 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.

[0053] The "intelligent computing power" described in the present invention refers to a computing platform that is scaled up and deployed for various artificial intelligence innovation applications based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit), such as natural language processing and machine vision.

[0054] The "super computing power" described in the present invention mainly refers to 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 processes extremely complex or data-intensive problems through a dedicated operating system. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0055] The "intelligent computing center" described in the present invention refers to a facility that mainly provides the required computing power, data, and algorithms for artificial intelligence applications (such as scenarios of artificial intelligence deep learning model development, model training, and model inference) by using large-scale heterogeneous computing power resources, including general computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enabling.

[0056] The "intelligent computing center" described in the present invention includes but is not limited to the "intelligent computing center".

[0057] The "Intelligent Computing Center" described in the present invention, namely the artificial intelligence computing center, is a type of computing infrastructure that is based on artificial intelligence theory, adopts an artificial intelligence computing architecture, and provides computing power services, data services, and algorithm services required for artificial intelligence applications.

[0058] The "Computing Power Center" described in the present invention refers to: a facility mainly composed of infrastructure such as wind, fire, water, and electricity, and IT software and hardware devices, with computing power, transportation capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0059] The "Supercomputing Center" described in the present invention refers to: namely the supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters, capable of providing functions such as large-scale computing, storage, and network services, and is widely used in application scenarios such as aerospace, national defense, oil exploration, climate modeling, and genome sequencing.

[0060] The "Computing Power Resources" described in the present invention refers to: technologies and facilities required for the development of the digital society with information computing, transmission, storage, and application capabilities, including but not limited to computing resources such as CPUs and GPUs, 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, and support and guarantee resources such as wind, fire, water, and electricity.

[0061] The "Group Relative Policy Optimization" described in the present invention is a swarm intelligence optimization method based on reinforcement learning, aiming to achieve efficient policy optimization for multi-agent systems or group decision-making tasks through relative comparison of intra-group policies rather than absolute reward signals.

[0062] The present invention provides a method for scheduling computing power resources of an intelligent computing center, as Figure 1 shown, the method includes the following steps:

[0063] Step S1: During the group relative policy optimization training process of the current round of the model, monitor and obtain the current sampling time required for the sampling process, and obtain the current evaluation time required for the evaluation process.

[0064] Combined with Figure 2 For the group relative policy optimization (Group Relative Policy Optimization, GRPO) in this step, the training data q is input into the policy model (Policy Model), and the Policy Model generates multiple output results O1, O2,..., O based on multiple operation methods G(The process of generating multiple output results is called the sampling process). The multiple generated output results are input into a preset Reference Model and Reward Model for evaluation, generating multiple evaluation results r1, r2, … r G (The process of generating multiple evaluation results is called the evaluation process). Based on the multiple evaluation results, a Group Computation is performed to obtain correction parameters A1, A2, … A for multiple operation methods G , and the corresponding operation method of the Policy Model can be adjusted. The above process is called one round of training of the model.

[0065] It should be noted that during the training process, situations such as data changes, problem difficulty changes, and evaluation difficulty changes may occur, resulting in changes in the time consumption of the sampling process and the evaluation process.

[0066] Step S2: Based on the current sampling time consumption and the current evaluation time consumption, adjust the scale of computing power resources for the next round of training process of the model.

[0067] In this step, the current sampling time consumption and the current evaluation time consumption can reflect the utilization of computing power resources when the current scale of computing power resources executes the current sampling and evaluation tasks. When the utilization rate of computing power resources is lower than the preset utilization rate, the scale of computing power resources can be reduced to improve the utilization rate and avoid waste; when the computing power resources are already overloaded, the scale of computing power resources can be increased to ensure the running speed of the task and improve the task processing efficiency.

[0068] In the computing power resource scheduling method of the intelligent computing center provided by the present invention, the sampling time consumption and the evaluation time consumption of the current round of the model can reflect the utilization of computing power resources when the computing power scale resources of the current round of the model execute the sampling and evaluation tasks of the current round. Therefore, by adjusting the scale of computing power resources for the next round of the model according to the utilization of computing power resources of the current round of the model, the utilization rate of computing power resources can be improved.

[0069] Optionally, step S2 includes:

[0070] Step S21: Determine the sum of the current sampling time consumption and the current evaluation time consumption as the current time consumption;

[0071] Step S22: Compare the current time consumption with the preset time consumption, and adjust the scale of computing power resources for the next round of training process of the model according to the obtained comparison result.

[0072] In this embodiment, the preset time consumption is the time consumption for the sampling and evaluation processes preset according to historical data or industry standards. Calculate the sum of the current sampling time consumption and the current evaluation time consumption, that is, the current time consumption, and compare the current time consumption with the preset time consumption to obtain a comparison result. Exemplarily, a time consumption range can be obtained according to the preset time consumption. In the case where the current time consumption is not within the time consumption range, adjust the scale of the computing power resources. When it is less than the preset time consumption, the scale of the computing power resources can be reduced, and when it is greater than the preset time consumption, the scale of the computing power resources can be increased.

[0073] Optionally, step S2 includes:

[0074] Step S23: Obtain the current computing power resource scale corresponding to the current round of training process of the model, and obtain the historical computing power resource scale of the previous round of training process of the model, the historical sampling time consumption corresponding to the sampling process, and the historical evaluation time consumption corresponding to the evaluation process;

[0075] Step S24: Determine the theoretical time consumption based on the current computing power resource scale, the historical computing power resource scale, the historical sampling time consumption, and the historical evaluation time consumption, where the theoretical time consumption is the time consumption required to process the sampling and evaluation of the previous round of training process of the model under the condition of the highest utilization rate of the current computing power resource scale;

[0076] Step S25: Adjust the computing power resource scale for the next round of training process of the model based on the theoretical time consumption, the current sampling time consumption, and the current evaluation time consumption.

[0077] In this embodiment, based on the current computing power resource scale and the historical computing power resource scale, a scale change coefficient indicating the degree of change of the current computing power resource scale relative to the historical computing power resource scale can be determined. When the current computing power resource scale can be determined according to the scale change coefficient, the historical sampling time consumption, and the historical evaluation time consumption, the theoretical time consumption for processing the sampling and evaluation tasks of the previous round under the condition of the highest utilization rate of the computing power resources can be determined.

[0078] Compare the theoretical time consumption with the current sampling time consumption and the current evaluation time consumption, and the computing power resource scale for the next round of training process of the model can be adjusted according to the comparison result.

[0079] In this embodiment, based on the current computing power resource scale, the historical computing power resource scale, the historical sampling time consumption, and the historical evaluation time consumption, the theoretical time consumption is determined. Since the theoretical time consumption can more accurately reflect the usage status of the abandoned computing power scale resources compared to the preset time consumption, by comparing and analyzing the theoretical time consumption with the current sampling time consumption and the current evaluation time consumption and then adjusting the computing power resource scale, the accuracy of the computing power resource scale can be improved, and the computing power resource utilization rate can be increased while ensuring that the task processing time does not exceed the range.

[0080] Optionally, step S25 includes:

[0081] Step S251: Determine the current time consumption by adding the current sampling time consumption and the current evaluation time consumption;

[0082] Step S252: Determine the target time consumption range based on the theoretical time consumption and the preset parameters;

[0083] Step S253: Adjust the computing power resource scale for the next round of training process of the model based on the current time consumption and the target time consumption range.

[0084] In this embodiment, when the current time consumption is not within the target time consumption range, the computing power resource scale is adjusted. When it is less than the theoretical time consumption, the computing power resource scale can be reduced, and when it is greater than the theoretical time consumption, the computing power resource scale can be increased. It should be noted that the scale of the reduction or increase of the computing power resources can be determined by the degree of change of the current time consumption compared to the theoretical time consumption. The greater the degree of change, the greater the scale of the scheduled computing power resources.

[0085] Optionally, step S2 includes:

[0086] Step S26: Obtain the historical sampling time consumption corresponding to the sampling process and the historical evaluation time consumption corresponding to the evaluation process in the previous round of training process of the model;

[0087] Step S27: Adjust the computing power resource scale for the next round of training process of the model based on the historical sampling time consumption, the historical evaluation time consumption, the current sampling time consumption, and the current evaluation time consumption.

[0088] In this embodiment, by directly comparing the current sampling time consumption and the current evaluation time consumption with the historical sampling time consumption and the historical evaluation time consumption, the change in the computing power resource utilization situation in the current round compared to that in the previous round is judged. When the computing power resource utilization rate decreases, the computing power resource scale can be reduced, and when the computing power resources are overloaded, the computing power resource scale can be increased. Compared with scheduling according to the historical time consumption or the theoretical time consumption, the method in this embodiment is more intuitive in terms of the performance of the resulting time consumption after scheduling.

[0089] Optionally, step S27 includes:

[0090] Step S271: Determine the sum of the historical sampling time consumption and the historical evaluation time consumption as the historical time consumption;

[0091] Step S272: Determine the sum of the current sampling time consumption and the current evaluation time consumption as the current time consumption;

[0092] Step S273: When the historical time consumption is less than the current time consumption, increase the scale of computing power resources for the next round of training process of the model; or, when the current time consumption is less than the historical time consumption, reduce the scale of computing power resources for the next round of training process of the model.

[0093] In this embodiment, by comparing the sum of the historical sampling time consumption and the historical evaluation time consumption with the sum of the current sampling time consumption and the current evaluation time consumption, it is possible to intuitively understand the change in the utilization of computing power resources under the scale of computing power resources in the current round compared with the previous round, and adjust the scale of computing power resources according to this change, which is simpler than other methods.

[0094] As Figure 3 shown, the present application embodiment also provides a computing power resource scheduling device 300 for an intelligent computing center, including:

[0095] A monitoring module 301, configured to monitor and obtain the current sampling time consumption required for the sampling process, and obtain the current evaluation time consumption required for the evaluation process during the population relative strategy optimization training process of the current round of the model;

[0096] A scheduling module 302, configured to adjust the scale of computing power resources for the next round of training process of the model based on the current sampling time consumption and the current evaluation time consumption.

[0097] Optionally, the scheduling module 302 is further configured to:

[0098] Determine the sum of the current sampling time consumption and the current evaluation time consumption as the current time consumption;

[0099] Compare the current time consumption with a preset time consumption, and adjust the scale of computing power resources for the next round of training process of the model according to the obtained comparison result.

[0100] Optionally, the scheduling module 302 is further configured to:

[0101] Obtain the current computing power resource scale corresponding to the current round of training process of the model, and obtain the historical computing power resource scale of the previous round of training process of the model, the historical sampling time consumption corresponding to the sampling process, and the historical evaluation time consumption corresponding to the evaluation process;

[0102] Based on the current computing power resource scale, the historical computing power resource scale, the historical sampling time consumption, and the historical evaluation time consumption, determine the theoretical time consumption, where the theoretical time consumption is the time consumption required for sampling and evaluation in the previous round of training of the model when the utilization rate of the current computing power resource scale is the highest;

[0103] Based on the theoretical time consumption, the current sampling time consumption, and the current evaluation time consumption, adjust the computing power resource scale for the next round of training of the model.

[0104] Optionally, the scheduling module 302 is further configured to:

[0105] Determine the sum of the current sampling time consumption and the current evaluation time consumption as the current time consumption;

[0106] Based on the theoretical time consumption and preset parameters, determine the target time consumption range;

[0107] Based on the current time consumption and the target time consumption range, adjust the computing power resource scale for the next round of training of the model.

[0108] Optionally, the scheduling module 302 is further configured to:

[0109] Obtain the historical sampling time consumption corresponding to the sampling process and the historical evaluation time consumption corresponding to the evaluation process in the previous round of training of the model;

[0110] Based on the historical sampling time consumption, the historical evaluation time consumption, the current sampling time consumption, and the current evaluation time consumption, adjust the computing power resource scale for the next round of training of the model.

[0111] Optionally, the scheduling module 302 is further configured to:

[0112] Determine the sum of the historical sampling time consumption and the historical evaluation time consumption as the historical time consumption;

[0113] Determine the sum of the current sampling time consumption and the current evaluation time consumption as the current time consumption;

[0114] In the case where the historical time consumption is less than the current time consumption, increase the computing power resource scale for the next round of training of the model; or, in the case where the current time consumption is less than the historical time consumption, decrease the computing power resource scale for the next round of training of the model.

[0115] The computing power resource scheduling device 300 of the intelligent computing center according to the embodiments of the present invention can implement each process of the computing power resource scheduling method of the intelligent computing center according to the embodiments of the present invention and achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0116] Please refer to Figure 4 Furthermore, the present invention also provides an electronic device, including a processor 402, a memory 401, and a computer program stored on the memory 501 and executable on the processor 402. When the computer program is executed by the processor 402, it implements each process of the foregoing method embodiment for scheduling computing power resources of the intelligent computing center and can achieve the same technical effects. To avoid repetition, details are not described herein again.

[0117] The present invention also provides a computer-readable storage medium with a computer program stored thereon. When the computer program is executed by a processor, it implements each process of the foregoing method embodiment for scheduling computing power resources of the intelligent computing center and can achieve the same technical effects. To avoid repetition, details are not described herein again. Among them, the computer-readable storage medium may be, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0118] The embodiment of the present application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement each process of the foregoing Figure 1 method embodiment for scheduling computing power resources of the intelligent computing center as shown and can achieve the same technical effects. To avoid repetition, details are not described herein again.

[0119] It should be noted that in this document, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including that element.

[0120] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc) and includes several instructions for causing a terminal (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0121] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative rather than restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the purpose of the present invention and the scope protected by the claims, and all of them fall within the protection scope of the present invention.

Claims

1. A computing power resource scheduling method for an intelligent computing center, characterized in that, Including: Step S1: During the population relative policy optimization training process of the current round of the model, monitor and obtain the current sampling time required for the sampling process, and obtain the current evaluation time required for the evaluation process; Step S2: Based on the current sampling time and the current evaluation time, adjust the scale of computing power resources for the next round of training process of the model.

2. The method according to claim 1, wherein The said Step S2 includes: Step S21: Determine the current time as the sum of the current sampling time and the current evaluation time; Step S22: Compare the current time with a preset time, and adjust the scale of computing power resources for the next round of training process of the model according to the obtained comparison result.

3. The method according to claim 1, characterized in that, The said Step S2 includes: Step S23: Obtain the current scale of computing power resources corresponding to the current round of training process of the model, and obtain the historical scale of computing power resources of the previous round of training process of the model, the historical sampling time corresponding to the sampling process, and the historical evaluation time corresponding to the evaluation process; Step S24: Based on the current scale of computing power resources, the historical scale of computing power resources, the historical sampling time, and the historical evaluation time, determine the theoretical time, where the theoretical time is the time required to process the sampling and evaluation of the previous round of training process of the model under the condition of the highest utilization rate of the current scale of computing power resources; Step S25: Based on the theoretical time, the current sampling time, and the current evaluation time, adjust the scale of computing power resources for the next round of training process of the model.

4. The method according to claim 3, wherein The said Step S25 includes: Step S251: Determine the current time as the sum of the current sampling time and the current evaluation time; Step S252: Based on the theoretical time and preset parameters, determine the target time range; Step S253: Based on the current time and the target time range, adjust the scale of computing power resources for the next round of training process of the model.

5. The method according to claim 1, wherein The said Step S2 includes: Step S26: Obtain the historical sampling time corresponding to the sampling process in the previous round of training process of the model, and the historical evaluation time corresponding to the evaluation process; Step S27: Based on the historical sampling time, the historical evaluation time, the current sampling time, and the current evaluation time, adjust the scale of computing power resources for the next round of training process of the model.

6. The method according to claim 5, characterized in that, The said Step S27 includes: Step S271: Determine the historical time as the sum of the historical sampling time and the historical evaluation time; Step S272: Determine the current time as the sum of the current sampling time and the current evaluation time; Step S273: In the case where the historical time is less than the current time, increase the scale of computing power resources for the next round of training process of the model; or, in the case where the current time is less than the historical time, decrease the scale of computing power resources for the next round of training process of the model.

7. A computing power resource scheduling device for an intelligent computing center, characterized in that, The said device includes: A monitoring module, configured to monitor and obtain the current sampling time required for the sampling process, and obtain the current evaluation time required for the evaluation process during the population relative policy optimization training process of the current round of the model; A scheduling module, configured to adjust the scale of computing power resources for the next round of training process of the model based on the current sampling time consumption and the current evaluation time consumption.

8. A server, characterized in that, Comprising: A processor, a memory, and a program stored on the memory and executable on the processor, the program implementing the steps of the computing power resource scheduling method of the intelligent computing center according to any one of claims 1 to 6 when executed by the processor.

9. A computer-readable storage medium, characterized in that A computer program is stored on the computer-readable storage medium, and the computer program implements the steps of the computing power resource scheduling method of the intelligent computing center according to any one of claims 1 to 6 when executed by a processor.

10. A computer program product, characterized in that, Including computer instructions, the computer instructions implementing the steps of the computing power resource scheduling method of the intelligent computing center according to any one of claims 1 to 6 when executed by a processor.

Images