WO2023216500A1 - 智算中心的算力资源部署方法、装置、设备及存储介质 - Google Patents

智算中心的算力资源部署方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2023216500A1
WO2023216500A1 PCT/CN2022/123402 CN2022123402W WO2023216500A1 WO 2023216500 A1 WO2023216500 A1 WO 2023216500A1 CN 2022123402 W CN2022123402 W CN 2022123402W WO 2023216500 A1 WO2023216500 A1 WO 2023216500A1
Authority
WO
WIPO (PCT)
Prior art keywords
production
stage
computing power
benefit
computing
Prior art date
Application number
PCT/CN2022/123402
Other languages
English (en)
French (fr)
Inventor
崔坤磊
Original Assignee
苏州元脑智能科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 苏州元脑智能科技有限公司 filed Critical 苏州元脑智能科技有限公司
Publication of WO2023216500A1 publication Critical patent/WO2023216500A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06313—Resource planning in a project environment
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06315—Needs-based resource requirements planning or analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G06Q10/087—Inventory or stock management, e.g. order filling, procurement or balancing against orders
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the present application relates to the field of computer technology, and in particular to a computing resource deployment method, device, electronic equipment and non-volatile readable storage medium for an intelligent computing center.
  • the Intelligent Computing Center is the most important computing power production center in the smart era. It uses a converged architecture computing system as a platform and data as a resource. It can use powerful computing power to drive AI (Artificial Intelligence) models to deeply process data.
  • AI Artificial Intelligence
  • Various smart computing services are continuously generated and provided to organizations and individuals in the form of cloud services through the network.
  • computing has gradually evolved from initial numerical calculation to scientific calculation, critical calculation and intelligent calculation. Each calculation has a corresponding computing power center to support it.
  • the computing power center that carries scientific calculations is the supercomputing center.
  • the computing power centers that carry current enterprise applications, government applications, and personal applications are a large number of various data centers.
  • the current computing demand for artificial intelligence is growing exponentially, and will account for more than 80% of society's total computing demand in the future.
  • the AI computing power center or intelligent computing center, carries this demand.
  • the intelligent computing center has three basic requirements: open standards, intensive efficiency and universal benefit. More than 80% of enterprises have applied open source software technologies in their data centers, such as OpenStack, K8S (Kubernetes), Hadoop (Haidupu), TensorFlow, etc., open source basic software for cloud computing, big data, artificial intelligence and other scenarios. It has become the de facto standard for intelligent computing center software platforms. Open computing technology can save power, reduce system failure rates, improve operation and maintenance efficiency and investment return rate, and can deliver computing resources at a speed of up to 10,000 units per day.
  • This application provides a computing resource deployment method, device, electronic equipment and non-volatile readable storage medium for an intelligent computing center to achieve optimal deployment of computing resources in the intelligent computing center.
  • embodiments of the present application provide a computing power resource deployment method for an intelligent computing center, which is applied to the intelligent computing center and includes:
  • computing power inventory fixed production computing power, computing power production planning and benefit characterization factors in the target production stage, the relationship between benefits and computing power production quantity is generated;
  • the computing power production plan for each production stage is determined based on the relationship between benefits and computing power production quantity; among them, the fixed production computing power of each production stage is less than or equal to the corresponding computing power inventory and calculation The sum of force production planning.
  • the stage benefits starting from each production stage to the end of the entire production process are determined;
  • the corresponding computing power inventory and quantity of each production stage are determined in sequence from the second production stage to the last production stage.
  • each production stage includes the computing power inventory and the end-of-stage computing power.
  • the end-of-stage computing power of each production stage is the computing power inventory of the adjacent next production stage; the end-of-stage computing power of each production stage is based on the corresponding The computing power inventory and computing power production planning in the production stage are determined.
  • the method also includes: receiving demand information, the demand information includes computing power demand, the computing power demand is the total amount of computing power resources required in all to-be-planned production cycles; wherein, the computing power inventory in the first production stage The computing power output of the last production stage is the same as the computing power demand for the inventory level before the start of all planned production periods.
  • the demand information also includes computing power delivery time and delivery volume
  • computing power inventory Based on the computing power inventory, fixed production computing power, computing power production planning and benefit characterization factors of the target production stage, before generating the relationship between benefits and computing power production quantity, it also includes: determining each production stage based on the computing power delivery time and delivery volume. of fixed production computing power.
  • the target production stage before generating the relationship between benefit and computing power production quantity based on the computing power inventory, fixed production computing power, computing power production planning and benefit characterization factors of the target production stage, it also includes:
  • the benefit information includes benefit representation factors, the functional relationship between the stage benefit of each production stage and the stage computing power output.
  • the target production stage before generating the relationship between benefit and computing power production quantity based on the computing power inventory, fixed production computing power, computing power production planning and benefit characterization factors of the target production stage, it also includes:
  • the order delivery time and total number of delivery computing nodes are obtained by parsing the production order request;
  • the fixed production computing power for each production stage is determined based on the order delivery time and the total number of delivery computing nodes.
  • computing power production planning and benefit characterization factors of the target production stage Based on the computing power inventory, computing power production planning and benefit characterization factors of the target production stage, the target stage benefits of the target production stage are generated;
  • the computing power inventory of the target production stage is determined to be used to generate the adjacent production stage of the target production stage.
  • a benefit representation function is generated to express the relationship between benefits and computing power production quantity.
  • a benefit representation function to express the relationship between benefits and computing power production quantity, including:
  • the first benefit representation function relationship is called to calculate the benefit representation function of the target production stage; the first benefit representation function relationship is:
  • x k is the computing power inventory of the target production stage
  • f k (x k ) is the benefit representation function of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) are the target stage benefits
  • f k+1 (x k+1 ) is the benefit representation function of the next production stage after the target production stage
  • N is the total number of production stages.
  • a benefit representation function to express the relationship between benefits and computing power production quantity, including:
  • the second benefit representation function relationship is called to calculate the benefit representation function of the target production stage; the second benefit representation function relationship is:
  • x k is the computing power inventory of the target production stage
  • f k (x k ) is the benefit representation function of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) are the target stage benefits
  • f k+1 (x k+1 ) is the benefit representation function of the next production stage after the target production stage
  • N is the total number of production stages.
  • calculate the first computing power production plan of the first production stage including:
  • the first computing power production plan of the first production stage is determined.
  • reduced benefits include one or more of the following: production costs, warehousing costs, and resource consumption.
  • growth benefits include one or more of the following: profit, distance, and output.
  • stage benefits of each production stage including:
  • the functional relationship between the stage benefit of each production stage and the stage's computing power output is the functional relationship between the production cost of each production stage and the square of the stage's computing power output;
  • the functional relationship between the production cost of each production stage and the square of the computational power output of the stage, the stage benefit of each production stage is determined.
  • the stage benefit of each production stage is the production cost of the corresponding stage.
  • the target stage benefits of the target production stage based on the computing power inventory, computing power production planning and benefit characterization factors of the target production stage, including:
  • the target stage benefit of the target production stage is the production cost
  • the target stage benefit of the target production stage is determined according to the following formula:
  • x k is the computing power inventory of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) is the target stage benefit
  • ⁇ is the production cost coefficient
  • determine the computing power inventory of the next production stage after the target production stage including:
  • x k+1 is the computing power inventory of the next production stage after the target production stage
  • a k is the fixed production computing power of the target production stage.
  • embodiments of the present application provide a computing resource deployment device for an intelligent computing center, which is applied to the intelligent computing center and includes:
  • the production segmentation module is used to respond to phase division instructions and divide all production cycles to be planned to generate multiple production phases that meet the preset correlation relationships;
  • Benefit and output relationship determination module is used to generate the relationship between benefit and computing power production quantity based on the computing power inventory, fixed production computing power, computing power production planning and benefit characterization factors of the target production stage; fixed production calculation for each production stage The force is less than or equal to the sum of the corresponding computing power inventory and computing power production plan;
  • the production planning generation module is used to determine the computing power production plan for each production stage based on the relationship between efficiency and computing power production quantity when the total benefit optimal conditions are met.
  • An embodiment of the present application also provides an electronic device, including a processor, and the processor is configured to implement the steps of the computing resource deployment method of the intelligent computing center as described above when executing a computer program stored in the memory.
  • the embodiment of the present application also provides a non-volatile readable storage medium.
  • a computer program is stored on the non-volatile readable storage medium.
  • the computer program is executed by the processor, the calculation of the intelligent computing center as mentioned above is realized. steps in the human resources deployment method.
  • the advantage of the technical solution provided by this application is that the future production cycle of the required computing power resources is divided into stages according to specified standards, so that the entire computing power production plan can be deployed in multiple production stages, and each production stage is connected and affects each other. , on the basis of meeting the computing volume requirements of each production stage, determine the computing power production plan under the optimal benefit of the corresponding stage, and finally determine the computing power resource production plan under the optimal benefit, thereby minimizing costs. Waste and achieve optimal deployment of computing power resources in intelligent computing centers.
  • the embodiments of this application also provide corresponding implementation devices, electronic equipment and non-volatile readable storage media for the computing resource deployment method of the intelligent computing center, further making the method more practical, and the devices, electronic equipment and non-volatile readable storage media Volatile readable storage media has corresponding advantages.
  • Figure 1 is a schematic flowchart of a method for deploying computing resources in an intelligent computing center provided by an embodiment of the present application
  • Figure 2 is a schematic diagram of the relationship between various production stages in an exemplary application scenario provided by the embodiment of the present application;
  • Figure 3 is a structural diagram of a specific implementation of the computing resource deployment device of the intelligent computing center provided by the embodiment of the present application;
  • Figure 4 is a structural diagram of a specific implementation of the electronic device provided by the embodiment of the present application.
  • Figure 1 is a schematic flow chart of a computing resource deployment method for an intelligent computing center provided by an embodiment of the present application. It is applied to an intelligent computing center.
  • the embodiment of the present application may include the following:
  • the phase division instruction can be issued through the human-computer interaction page.
  • the phase division instruction includes the standard unit of division of the stage and the period that needs to be divided.
  • the period that needs to be divided is all the production cycles to be planned.
  • All production cycles to be planned are all time periods in which computing resource deployment operations are performed, that is, all time periods in which computing resources are produced.
  • This embodiment treats all production cycles to be planned as a whole time period.
  • Computing resources can include multiple types of resources. Computing resources can be, for example, computing nodes, that is, the number of computing nodes can be used to represent the measurement of computing power.
  • the phase division instruction can also carry a computing resource type field. After receiving the phase division instruction, the system extracts the division standard unit and the period to be divided from the instruction.
  • the division standard unit can be based on the actual demand, that is, the production stage that needs to be divided, the computing resource demand and the length of the production cycle to be planned. Make flexible choices.
  • the standard unit of division can be month, week, quarter, etc.
  • the overall time period is divided into multiple time stages according to the division standard unit, and each time stage is a production stage.
  • the computing power production plan for the next few production cycles can be divided into phases on a monthly basis. That is, by analyzing the stage division instructions, the division standard unit is obtained; according to the division standard unit, all the production cycles to be planned are divided to obtain multiple production stages. In this embodiment, computing power is produced in each production stage, and each production stage together constitutes all the production cycles to be planned.
  • each production stage has an initial state and an end state.
  • the state represents the position or state of the system in a certain production stage, and its end state is the initial state of the next stage; a certain production After the state of the stage is determined, the decision to evolve from this state to the next production stage is called decision-making, which is also called computing resource planning or computing production planning.
  • each production stage can include computing power inventory and end-of-stage computing power.
  • the computing power inventory is the computing power inventory before each production stage starts to execute computing resource production.
  • the computing resources already available are the inventory at each production stage.
  • the computing power at the end of the stage refers to all the computing resources obtained after the computing power production operation in this production stage.
  • the computing power inventory in the first production stage is the inventory before the start of all planned production cycles, and the computing power output in the last production stage is the same as the computing power demand.
  • users can pre-deliver the total amount of computing power resources required in all planned production cycles through the human-computer interaction interface, that is, computing power requirements.
  • the required computing resources are deployed for production in all to-be-planned production cycles. That is to say, users can put forward their own requirements for computing power scale according to their own needs, which can be simply expressed as how many computing nodes are needed, and submit the requirements such as delivery time and number of delivery nodes to the intelligent computing center; the intelligent computing center based on user needs Plan and specify computing power production schedules.
  • the end-of-stage computing power of each production stage is the computing power inventory of the adjacent next production stage; the end-of-stage computing power of each production stage is determined based on the computing power inventory of the corresponding production stage and the computing power production plan.
  • S102 Generate a relationship between benefits and computing power production quantity based on the computing power inventory, fixed production computing power, computing power production planning and benefit characterization factors in the target production stage.
  • the target production stage can be any one of the multiple production stages divided in the previous step.
  • the computing power inventory is the inventory of the target production stage, and the fixed production computing power is the user in the target production stage.
  • the computing power resources are required, and the computing power production planning is the computing power resources planned to be produced in the target production stage.
  • the computing power production planning is not exactly the same as the pre-acquired computing power demand, that is, the fixed production computing power. It is understandable that in order to ensure the best overall efficiency, each production stage may produce more than the initial user demand of the production stage. If there are more computing resources, they may be less than the initial user demand for the production stage. However, the overall required computing resources must be met.
  • This embodiment is to select an optimal decision within the allowed range so that the entire system can achieve the best effect under predetermined standards.
  • the predetermined standards can usually be determined based on benefits, that is, to achieve the best overall benefits based on meeting the user's computing resource needs. Therefore, it is necessary to determine the relationship between the benefits of each production stage and the computational power production quantity when the sum of the benefits of all production stages is optimal.
  • the benefits obtained from executing the decisions of this production stage are the stage benefits.
  • the stage benefits are part of the benefits of the entire system.
  • the stage benefits can be calculated based on the preset benefit information, and the benefit information can be stored to the specified path, or the user inputs benefit information.
  • the benefit information may include benefit characterization factors, the functional relationship between the stage benefit of each production stage and the stage computing power output.
  • the benefit characterization factor is the parameter that characterizes the benefit. For example, if the benefit is the production cost, then the benefit characterization factor is the production cost coefficient. If the benefit is the warehousing cost, then the benefit characterization factor is the warehousing cost coefficient.
  • the functional relationship between the stage benefit and the stage computing power output can be preset. Taking the benefit as the production cost as an example, if the production cost of each production stage is set to be proportional to the square of the output, it is also the stage benefit of the production stage. as a function of stage computational power yield.
  • the computing resource production process of the intelligent computing center's computing power can be order-based production.
  • the production of computing power can be planned according to user needs, that is, the fixed production computing power can be informed to the intelligent computing center through orders.
  • the computing power resource is a computing power node
  • the computing power inventory, fixed production computing power, Computing power production planning and benefit characterization factors, before generating the relationship between benefits and computing power production quantity when receiving a production order request, by parsing the production order request, the order delivery time and the total number of delivery computing nodes are obtained; based on the order delivery time and delivery calculation
  • the total number of nodes determines the fixed production computing power of each production stage, that is, the computing power resources required by users in that production stage.
  • the correlation between each production stage can be known based on S101.
  • the benefits of the remaining production stages can be calculated sequentially based on the relationship between the benefits of the target production stage and the computational power production quantity.
  • the production plan will be gradually decomposed, not only to meet the computing power needs of each production stage, but also to ensure that the total benefit of the overall production plan is the highest , through basic mathematical knowledge, the computing power production plan of each production stage is obtained.
  • the intelligent computing center deploys all computing power resources required by users in each production stage based on the computing power production planning.
  • the future production cycle of the required computing power resources is divided into stages according to specified standards, so that the entire computing power production plan can be deployed in multiple production stages, and each production stage is interconnected and interconnected. Impact, on the basis of meeting the computing volume requirements of each production stage, determine the computing power production plan under the optimal benefit of the corresponding stage, and finally determine the computing power resource production plan under the optimal benefit, thus minimizing the Eliminate cost waste and achieve the optimal deployment of computing power resources in the intelligent computing center.
  • step S103 provides an optional method of determining the computing power production plan for each production stage, which may include the following steps:
  • the stage benefits starting from each production stage to the end of the entire production process are determined;
  • the corresponding computing power inventory and quantity of each production stage are determined in sequence from the second production stage to the last production stage.
  • the target production stage can be the last production stage. Based on the relationship between the efficiency and computing power of the last production stage and the production quantity, it can be deduced from the end point one production stage one production stage to meet the production needs of the user. Under the premise of planning, the optimal output and efficiency from each production stage to the end point can be obtained; after calculating the optimal production plan of the first production stage, the production plan of each production stage can be calculated sequentially, and the overall Optimum efficiency. The whole process is simple to calculate and easy to implement.
  • step S102 there is no limitation on how to perform step S102.
  • This embodiment provides an optional determination method of the relationship between benefit and computing power production quantity, which may include the following steps:
  • computing power production planning and benefit characterization factors of the target production stage Based on the computing power inventory, computing power production planning and benefit characterization factors of the target production stage, the target stage benefits of the target production stage are generated;
  • the computing power inventory of the target production stage is determined to be used to generate the adjacent production stage of the target production stage.
  • a benefit representation function is generated to express the relationship between benefits and computing power production quantity.
  • the benefit representation function can be defined on the entire process or on the subsequent sub-processes.
  • the benefit representation function is often a certain sum of benefits at each stage. After executing the optimal strategy, the benefit representation function value should also be the highest or lowest.
  • the benefit representation function is a quantitative representation of the benefits generated by the execution of a certain strategy by the system. According to different actual situations, benefits can be an increase in benefits, such as profit, distance, output, etc.; benefits can also be a decrease in benefits, Such as production costs, warehousing costs, resource consumption, etc. Based on this, this embodiment provides different benefit characterization functions based on different benefit types, which may include the following:
  • the reduced benefit is a type in which the smaller the index value, the better the benefit, such as production cost, warehousing cost, resource consumption, etc.
  • the first benefit characterization function relation is called to calculate the benefit characterization function of the target production stage; the first benefit characterization function relation can be expressed as:
  • x k is the computing power inventory of the target production stage
  • f k (x k ) is the benefit representation function of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) are the target stage benefits
  • f k+1 (x k+1 ) is the benefit representation function of the next production stage after the target production stage
  • N is the total number of production stages.
  • the growth benefit is a type with a larger index value and better benefits, such as profit, distance, output, etc.
  • the second benefit characterization function relation is called to calculate the benefit characterization function of the target production stage; the second benefit characterization function relation can be expressed as:
  • x k is the computing power inventory of the target production stage
  • f k (x k ) is the benefit representation function of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) are the target stage benefits
  • f k+1 (x k+1 ) is the benefit representation function of the next production stage after the target production stage
  • N is the total number of production stages.
  • this embodiment uses a benefit representation function to quantitatively express the relationship between benefits and computing power production quantity, which is beneficial to improving the efficiency of the entire production planning and deployment, and provides benefit representation functions corresponding to different benefit types, which is more practical.
  • this application also provides a schematic example, which may include the following content:
  • all the production cycles to be planned are the next four months
  • the computing power resources are the number of computing nodes
  • the user’s fixed computing power production order requirements in the next four months are 600 computing nodes, 700 computing nodes, 500 computing nodes, 1200 computing nodes.
  • the benefit of this embodiment is cost, and the computing power cost is simply classified into two categories: warehousing cost and production cost. Without making too detailed a classification, all costs can be classified into the above two categories.
  • the actual warehousing cost and production cost may be more complicated. In actual operation, the warehousing cost function and the production cost function can be defined, or they can be simply described by coefficients, that is, the warehousing cost coefficient and the production cost coefficient.
  • x k is the computing power inventory of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) is the target stage benefit
  • ⁇ is the production cost coefficient
  • the future computing power production plan can be divided into stages.
  • this embodiment divides the total production plan into stages on a monthly basis, and divides the next four months into multiple production stages on a monthly basis.
  • the monthly computing power requirements can be easily obtained based on the delivery time and delivery node number requirements submitted by the user.
  • the existing computing power at the beginning of the k-th month is the state variable x k
  • the number of computing nodes that need to be produced in the k-th month is the computing power production plan or the computing node production plan is the decision variable u k ; then, take the decision variable u k from the state x k
  • the stage benefits of this embodiment are reflected in the production cost.
  • the target production stage is the last production stage, and the benefit representation function of the target production stage can be:
  • f k (x k ) is defined as the cost from the starting point of state x k to the end of the entire production process, and it must satisfy x k + uk ⁇ A k .
  • the technical problem to be solved in this embodiment is how to formulate a production plan for computing nodes and reduce costs to a minimum on the premise of meeting the monthly order quantity, that is, based on the computing power needs of the intelligent computing center in the next few production cycles. .
  • This embodiment starts backtracking from the last production stage, calculates the computing node production plan of the first production stage, and then sequentially calculates the computing node production plan of each production stage.
  • the calculation process can be as follows:
  • the computing node production plan u k for each production stage is determined, that is, how many computing nodes should be produced in each production stage.
  • the initial state x k of each production stage that is, the current number of computing nodes in inventory is regarded as a known quantity.
  • the computing power production planning of this production stage is u 3
  • f 3 (x 3 ) min ⁇ x 3 +0.005(u 3 ) 2 +7200 -11(x 3 +u 3 -500)+0.005(x 3 +u 3 -500) 2 ⁇ .
  • f 3 (x 3 ) is minimum, the derivative function of f 3 (x 3 ) is 0.
  • u 3 800-0.5x 3.
  • f 3 (x 3 ) 7550 -7x 3 +0.0025(x 3 ) 2 .
  • f 2 (x 2 ) 10000 -6x 2 +0.005(x 2 ) 2 /3.
  • f 1 (x 1 ) min ⁇ x 1 +0.005u 1 2 +10000-6(x 1 +u 1 -600)+0.005(x 1 +u 1 -600) 2 /3 ⁇
  • the total production cost is the lowest, 11,800.
  • the computer program on which the above method relies can be formed into the computing resource deployment program of the intelligent computing center and embedded in the processor.
  • the deployment of computing resources can be determined through programmed operation, which not only satisfies production needs of users, and can ensure optimal benefits.
  • the embodiments of this application also provide corresponding devices for the computing resource deployment method of the intelligent computing center, further making the method more practical.
  • the device can be described separately from the perspective of functional modules and the perspective of hardware.
  • the following is an introduction to the computing power resource deployment device of the intelligent computing center provided by the embodiment of the present application.
  • the computing power resource deployment device of the intelligent computing center described below and the computing power resource deployment method of the intelligent computing center described above can correspond to each other.
  • Figure 3 is a structural diagram of a computing power resource deployment device of an intelligent computing center provided by an embodiment of the present application in a specific implementation.
  • the device may include:
  • the production segmentation module 301 is used to respond to the phase division instructions and divide all the production cycles to be planned to generate multiple production phases that meet the preset association relationships;
  • the benefit and output relationship determination module 302 is used to generate the relationship between benefit and computational power production quantity based on the computing power inventory, fixed production computing power, computing power production planning and benefit characterization factors of the target production stage; the fixed production of each production stage The computing power is less than or equal to the sum of the corresponding computing power inventory and computing power production plan;
  • the production plan generation module 303 is used to determine the computing power production plan for each production stage based on the relationship between efficiency and computing power production quantity when the total benefit optimal conditions are met.
  • the above-mentioned production plan generation module 303 can also be used to: based on the relationship between efficiency and computing power production quantity, in order from the last production stage to the first production stage, determine sequentially Taking each production stage as the starting point to the end of the entire production process, the stage benefit; when the stage benefit of the first production stage is the largest, calculate the first computing power production plan of the first production stage; according to the first computing power production plan,
  • the relationship between the computing power inventory and the computing power inventory of adjacent production stages is to determine the corresponding computing power inventory and computing power production planning of each production stage in sequence from the second production stage to the last production stage.
  • the above-mentioned production segmentation module 301 can be further used to: obtain the division standard unit by parsing the stage division instructions; divide all the production cycles to be planned according to the division standard unit to obtain multiple production cycles.
  • the stage end computing power of each production stage is the computing power inventory of the adjacent next production stage; the stage end of each production stage The computing power is determined based on the computing power inventory and the computing power production plan of the corresponding production stage.
  • the above device also includes a demand acquisition module, configured to: receive demand information, the demand information includes computing power requirements, and the computing power requirements are all computing power resources required within the planned production cycle.
  • the demand information also includes computing power delivery time and delivery amount
  • the above-mentioned device also includes a fixed computing power acquisition module, which is used to determine the fixed production computing power of each production stage based on the computing power delivery time and delivery volume.
  • the above device may also include a benefit information acquisition module for acquiring benefit information to determine the stage benefits of each production stage based on the benefit information; wherein the benefit information includes benefits The functional relationship between the characterization factor, the stage benefit of each production stage and the stage computing power output.
  • the above device may also include a fixed production computing power determination module, for when receiving a production order request, obtain the order delivery time and delivery computing node by parsing the production order request. Total number; determine the fixed production computing power for each production stage based on the order delivery time and the total number of delivery computing nodes.
  • a fixed production computing power determination module for when receiving a production order request, obtain the order delivery time and delivery computing node by parsing the production order request. Total number; determine the fixed production computing power for each production stage based on the order delivery time and the total number of delivery computing nodes.
  • the above-mentioned benefit and output relationship determination module 302 can also be used to: generate the target production stage based on the computing power inventory, computing power production planning and benefit characterization factors of the target production stage.
  • Target stage benefit based on the computing power inventory, fixed production computing power and computing power production planning of the target production stage, determine the computing power inventory of the next production stage of the target production stage to be used to generate the next production of the target production stage.
  • the benefits of adjacent stages of the stage based on the benefits of the target stage and the benefits of adjacent stages, a benefit representation function is generated to express the relationship between benefits and computing power production quantity.
  • the benefit and output relationship determination module 302 can be further used to: if the benefit type to which the benefit characterization factor belongs is reduced benefit, call the first benefit representation function relationship to calculate the target production stage The benefit representation function of ; the first benefit representation function relationship formula is:
  • x k is the computing power inventory of the target production stage
  • f k (x k ) is the benefit representation function of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) are the target stage benefits
  • f k+1 (x k+1 ) is the benefit representation function of the next production stage after the target production stage
  • N is the total number of production stages.
  • the benefit and output relationship determination module 302 can further be used to: if the benefit type to which the benefit representation factor belongs is growth benefit, call the second benefit representation function relationship to calculate the target Benefit representation function in the production stage; the second benefit representation function relationship is:
  • x k is the computing power inventory of the target production stage
  • f k (x k ) is the benefit representation function of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) are the target stage benefits
  • f k+1 (x k+1 ) is the benefit representation function of the next production stage after the target production stage
  • N is the total number of production stages.
  • the above-mentioned production plan generation module is also used to: calculate the stage benefit of the first production stage according to the benefit representation function, derive the stage benefit of the first production stage, and determine the first production stage The functional relationship between the computing power production plan and the computing power inventory; based on the functional relationship between the computing power production planning of the first production stage and the computing power inventory, determine the first computing power production plan of the first production stage.
  • reduced benefits include one or more of the following: production costs, warehousing costs, and resource consumption.
  • growth benefits include one or more of the following: profit, distance, and output.
  • the above-mentioned benefit and output relationship determination module is also used to: when the benefit characterization factor is the production cost coefficient, the functional relationship between the stage benefit of each production stage and the stage calculation power output It is the functional relationship that the production cost of each production stage is proportional to the square of the stage's computing power output; according to the production cost coefficient, the functional relationship that the production cost of each production stage is proportional to the square of the stage's computing power output, determine each production The stage benefit of each stage is the production cost of the corresponding stage.
  • the above-mentioned benefit and output relationship determination module is also used to: when the target stage benefit of the target production stage is the production cost, the target stage benefit of the target production stage is determined according to the following formula:
  • x k is the computing power inventory of the target production stage
  • u k is the computing power production planning of the target production stage
  • d k (x k , u k ) is the target stage benefit
  • ⁇ is the production cost coefficient
  • the above-mentioned benefit and output relationship determination module is also used to determine the computing power inventory of the next production stage of the target production stage according to the following formula:
  • x k+1 is the computing power inventory of the next production stage after the target production stage
  • a k is the fixed production computing power of the target production stage.
  • each functional module of the computing resource deployment device of the intelligent computing center in the embodiment of this application can be specifically implemented according to the method in the above method embodiment.
  • the specific implementation process can be referred to the relevant description of the above method embodiment, and will not be described again here. .
  • FIG. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present application in an implementation manner. As shown in Figure 4, the electronic device includes a memory 40 for storing a computer program; a processor 41 for executing the computer program to implement the steps of the computing resource deployment method of the intelligent computing center mentioned in any of the above embodiments. .
  • the processor 41 may include one or more processing cores, such as a 4-core processor or an 8-core processor.
  • the processor 41 may also be a controller, a microcontroller, a microprocessor or other data processing chips.
  • the processor 41 can adopt at least one hardware form among DSP (Digital Signal Processing, digital signal processing), FPGA (Field-Programmable Gate Array, field programmable gate array), and PLA (Programmable Logic Array, programmable logic array). accomplish.
  • the processor 41 may also include a main processor and a co-processor.
  • the main processor is a processor used to process data in the wake-up state, also called CPU (Central Processing Unit, central processing unit); the co-processor is A low-power processor used to process data in standby mode.
  • the processor 41 may be integrated with a GPU (Graphics Processing Unit, image processor), and the GPU is responsible for rendering and drawing content that needs to be displayed on the display screen.
  • the processor 41 may also include an AI (Artificial Intelligence, artificial intelligence) processor, which is used to process computing operations related to machine learning.
  • AI Artificial Intelligence, artificial intelligence
  • Memory 40 may include one or more computer non-volatile readable storage media, which may be non-transitory.
  • the memory 40 may also include high-speed random access memory and non-volatile memory, such as one or more magnetic disk storage devices and flash memory storage devices.
  • the memory 40 in some embodiments may be an internal storage unit of the electronic device, such as a hard drive of a server.
  • the memory 40 may also be an external storage device of an electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, or a flash memory equipped on a server. Flash Card, etc.
  • the memory 40 may also include both an internal storage unit of the electronic device and an external storage device.
  • the memory 40 can not only be used to store application software installed on the electronic device and various types of data, such as codes for executing vulnerability processing methods, etc., but can also be used to temporarily store data that has been output or is to be output.
  • the memory 40 is at least used to store the following computer program 401. After the computer program is loaded and executed by the processor 41, it can implement the computing resource deployment method of the intelligent computing center disclosed in any of the foregoing embodiments. step.
  • the resources stored in the memory 40 may also include the operating system 402, data 403, etc., and the storage method may be short-term storage or permanent storage.
  • the operating system 402 may include Windows, Unix, Linux, etc.
  • Data 403 may include but is not limited to data corresponding to the computing resource deployment results of the intelligent computing center, etc.
  • the above-mentioned electronic device may also include a display screen 42 , an input-output interface 43 , a communication interface 44 or also called a network interface, a power supply 45 and a communication bus 46 .
  • the display screen 42 and the input and output interface 43 such as a keyboard belong to the user interface, and optional user interfaces may also include standard wired interfaces, wireless interfaces, etc.
  • the display may be an LED display, a liquid crystal display, a touch-controlled liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc.
  • a display which may also appropriately be called a display screen or display unit, is used for displaying information processed in the electronic device and for displaying a visual user interface.
  • the communication interface 44 may optionally include a wired interface and/or a wireless interface, such as a WI-FI interface, a Bluetooth interface, etc., and is generally used to establish communication connections between electronic devices and other electronic devices.
  • the communication bus 46 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc.
  • PCI peripheral component interconnect
  • EISA extended industry standard architecture
  • the bus can be divided into address bus, data bus, control bus, etc. For ease of presentation, only one thick line is used in Figure 4, but it does not mean that there is only one bus or one type of bus.
  • FIG. 4 does not limit the electronic device, and may include more or fewer components than shown, for example, it may also include sensors 47 that implement various functions.
  • the computing resource deployment method of the intelligent computing center in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the technical solution of the present application is essentially or contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , execute all or part of the steps of the methods of various embodiments of this application.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), electrically erasable programmable ROM, register, hard disk, multimedia Cards, card-type memories (such as SD or DX memories, etc.), magnetic memories, removable disks, CD-ROMs, magnetic disks or optical disks and other media that can store program codes.
  • embodiments of the present application also provide a non-volatile readable storage medium that stores a computer program.
  • the computer program is executed by the processor, the steps of the computing resource deployment method of the intelligent computing center in any of the above embodiments are performed.

Landscapes

  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Engineering & Computer Science (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Strategic Management (AREA)
  • Quality & Reliability (AREA)
  • Theoretical Computer Science (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Game Theory and Decision Science (AREA)
  • Accounting & Taxation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Finance (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请公开了一种智算中心的算力资源部署方法、装置、电子设备及非易失性可读存储介质，应用于计算机技术领域中的智算中心。其中，方法包括响应阶段划分指令，对所有待计划生产周期进行划分，生成多个满足预设关联关系的生产阶段。根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系；其中，每个生产阶段的固定生产计算力小于等于相应的计算力库存量和计算力生产规划之和。当满足总效益最优条件时，根据效益与计算力生产数量关系确定每个生产阶段的计算力生产规划，进而可基于各生产阶段的计算力生产规划进行算力资源部署，实现智算中心计算力资源的最优部署。

Description

智算中心的算力资源部署方法、装置、设备及存储介质
相关申请的交叉引用
本申请要求于2022年05月09日提交中国专利局，申请号为202210495846.1，申请名称为“智算中心的算力资源部署方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，特别是涉及一种智算中心的算力资源部署方法、装置、电子设备及非易失性可读存储介质。
背景技术
智算中心是智慧时代最主要的计算力生产中心，它以融合架构计算系统为平台，以数据为资源，能够以强大算力驱动AI(Artificial Intelligence，人工智能)模型对数据进行深度加工，源源不断地产生各种智慧计算服务，并通过网络以云服务形式供应给组织及个人。计算在发展过程中从最初的数值计算逐渐演变为科学计算、关键计算和智慧计算。每种计算都有相应的算力中心去支撑。承载科学计算的算力中心是超算中心。承载当前企业应用、政府应用和个人应用的算力中心是数量众多的各类数据中心。当前人工智能计算需求正呈指数级增长，未来在社会总计算需求中将占据80％以上，承载这种需求的就是AI算力中心，即智算中心。智算中心具有开放标准、集约高效和普适普惠三个基本要求。超过80％的企业都在其数据中心中应用了开源软件技术，如OpenStack、K8S(Kubernetes)、Hadoop(海杜普)、TensorFlow等，面向云计算、大数据、人工智能等场景的开源基础软件已经成为了智算中心软件平台的事实标准。开放的计算技术可以节省电力、降低系统故障率、提高运维效率和投资收益率，对计算资源的交付速度可达到每天1万台。
相关技术在对智算中心的算力资源进行部署时，通常是在已知未来生产周期计划的前提下，严格按照当前计算力需求量进行生产调度与部署。但是，这种方式并不是最优的算力资源部署策略，尤其是在面对智算中心大规模生产的应用场景中，采用相关技术会造成较大的成本浪费。
发明内容
本申请提供了一种智算中心的算力资源部署方法、装置、电子设备及非易失性可读存储介质，实现智算中心计算力资源的最优部署。
为解决上述技术问题，本申请实施例提供以下技术方案：
本申请实施例一方面提供了一种智算中心的算力资源部署方法，应用于智算中心，包括：
响应阶段划分指令，将所有待计划生产周期进行划分，以生成多个满足预设关联关系的生产阶段；
根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系；
当满足总效益最优条件时，基于效益与计算力生产数量关系，确定每个生产阶段的计算力生产规划；其中，每个生产阶段的固定生产计算力小于等于相应的计算力库存量和计算力生产规划之和。
可选的，当满足总效益最优条件时，基于效益与计算力生产数量关系，确定每个生产阶段的计算力生产规划，包括：
基于效益与计算力生产数量关系，按照从最后一个生产阶段到第一个生产阶段的顺序，依次确定以每个生产阶段为起点到整个生产过程结束的阶段效益；
当第一个生产阶段的阶段效益最大时，计算第一个生产阶段的第一计算力生产规划；
根据第一计算力生产规划、和相邻生产阶段的计算力库存量之间的关系，按照从第二个生产阶段到最后一个生产阶段的顺序，依次确定各生产阶段相应的计算力库存量和计算力生产规划。
可选的，将所有待计划生产周期进行划分，以生成多个满足预设关联关系的生产阶段，包括：
通过解析阶段划分指令，得到划分标准单位；
按照划分标准单位，对所有待计划生产周期进行划分，得到多个生产阶段；
其中，各生产阶段包括计算力库存量和阶段结束计算力，每个生产阶段的阶段结束计算力为相邻的下一个生产阶段的计算力库存量；每个生产阶段的阶段结束计算力根据相应生产阶段的计算力库存量和计算力生产规划确定。
可选的，方法还包括：接收需求信息，需求信息包括计算力需求，计算力需求为所有待计划生产周期内所需求的计算力资源总量；其中，第一个生产阶段的计算力库存量为所有待计划生产周期开始之前的库存量，最后一个生产阶段的计算力产量与计算力需求相同。
可选的，需求信息还包括计算力交付时间和交付量；
根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系之前，还包括：根据计算力交付时间和交付量，确定各生产阶段的固定生产计算力。
可选的，根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系之前，还包括：
获取效益信息，以根据效益信息，确定每个生产阶段的阶段效益；
其中，效益信息包括效益表征因子、每个生产阶段的阶段效益与阶段计算力产量的函数关系。
可选的，根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系之前，还包括：
当接收到生产订单请求，通过解析生产订单请求，得到订单交付时间和交付计算节点总数；
根据订单交付时间和交付计算节点总数，确定每个生产阶段的固定生产计算力。
可选的，根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系，包括：
基于目标生产阶段的计算力库存量、计算力生产规划和效益表征因子，生成目标生产阶段的目标阶段效益；
根据目标生产阶段的计算力库存量、固定生产计算力和计算力生产规划，确定目标生产阶段的下一个生产阶段的计算力库存量，以用于生成目标生产阶段的下一个生产阶段的相邻阶段效益；
基于目标阶段效益和相邻阶段效益，生成用于表示效益与计算力生产数量关系的效益表征函数。
可选的，基于目标阶段效益和相邻阶段效益，生成用于表示效益与计算力生产数量关系的效益表征函数，包括：
若效益表征因子所属效益类型为降低型效益，调用第一效益表征函数关系式计算目标生产阶段的效益表征函数；第一效益表征函数关系式为：
Figure PCTCN2022123402-appb-000001
式中，x k为目标生产阶段的计算力库存量，f k(x k)为目标生产阶段的目标生产阶段的效 益表征函数，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，f k+1(x k+1)为目标生产阶段的下一个生产阶段的效益表征函数，N为生产阶段总数。
可选的，基于目标阶段效益和相邻阶段效益，生成用于表示效益与计算力生产数量关系的效益表征函数，包括：
若效益表征因子所属效益类型为增长型效益，调用第二效益表征函数关系式计算目标生产阶段的效益表征函数；第二效益表征函数关系式为：
Figure PCTCN2022123402-appb-000002
式中，x k为目标生产阶段的计算力库存量，f k(x k)为目标生产阶段的目标生产阶段的效益表征函数，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，f k+1(x k+1)为目标生产阶段的下一个生产阶段的效益表征函数，N为生产阶段总数。
可选的，当第一个生产阶段的阶段效益最大时，计算第一个生产阶段的第一计算力生产规划，包括：
根据效益表征函数计算第一生产阶段的阶段效益，对第一生产阶段的阶段效益求导，确定第一生产阶段的计算力生产规划与计算力库存量之间的函数关系；
基于第一生产阶段的计算力生产规划与计算力库存量之间的函数关系，确定第一生产阶段的第一计算力生产规划。
可选的，降低型效益包括以下一种或多种：生产成本、仓储成本、资源消耗。
可选的，增长性效益包括以下一种或多种：利润、距离、产量。
可选的，根据效益信息，确定每个生产阶段的阶段效益，包括：
在效益表征因子为生产成本系数的情况下，每个生产阶段的阶段效益与阶段计算力产量的函数关系为每个生产阶段的生产成本与阶段计算力产量的平方成正比的函数关系；
根据生产成本系数、每个生产阶段的生产成本与阶段计算力产量的平方成正比的函数关系，确定每个生产阶段的阶段效益，每个生产阶段的阶段效益为相应阶段的生产成本。
可选的，基于目标生产阶段的计算力库存量、计算力生产规划和效益表征因子，生成目标生产阶段的目标阶段效益，包括：
在目标生产阶段的目标阶段效益为生产成本的情况下，目标生产阶段的目标阶段效益根据下式确定：
d k(x k，u k)＝x k+β(u k) 2
式中，x k为目标生产阶段的计算力库存量，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，β为生产成本系数。
可选的，根据目标生产阶段的计算力库存量、固定生产计算力和计算力生产规划，确定目标生产阶段的下一个生产阶段的计算力库存量，包括：
根据下式确定目标生产阶段的下一个生产阶段的计算力库存量：
x k+1＝x k+u k-A k
式中，x k+1为目标生产阶段的下一个生产阶段的计算力库存量，A k为目标生产阶段的固定生产计算力。
本申请实施例另一方面提供了一种智算中心的算力资源部署装置，应用于智算中心，包括：
生产分段模块，用于响应阶段划分指令，将所有待计划生产周期进行划分，以生成多个满足预设关联关系的生产阶段；
效益与产量关系确定模块，用于根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系；每个生产阶段的固定生产计算力小于等于相应的计算力库存量和计算力生产规划之和；
生产规划生成模块，用于当满足总效益最优条件时，基于效益与计算力生产数量关系，确定每个生产阶段的计算力生产规划。
本申请实施例还提供了一种电子设备，包括处理器，处理器用于执行存储器中存储的计算机程序时实现如前任一项智算中心的算力资源部署方法的步骤。
本申请实施例最后还提供了一种非易失性可读存储介质，非易失性可读存储介质上存储有计算机程序，计算机程序被处理器执行时实现如前任一项智算中心的算力资源部署方法的步骤。
本申请提供的技术方案的优点在于，将所需计算力资源的未来生产周期按照指定标准进行分阶段，从而可将整个计算力生产计划部署在多个生产阶段，各生产阶段彼此联系又互相影响，在满足每个生产阶段的计算量需求的基础上，确定相应阶段的最优效益下的计算力生产规划，最终可确定最优效益下的算力资源生产计划，从而可最大程度地降低成本浪费，实现对智算中心算力资源的最优部署。
此外，本申请实施例还针对智算中心的算力资源部署方法提供了相应的实现装置、电子设备及非易失性可读存储介质，进一步使得方法更具有实用性，装置、电子设备及非易失性可读存储介质具有相应的优点。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性的，并不能限制本公开。
附图说明
为了更清楚的说明本申请实施例或相关技术的技术方案，下面将对实施例或相关技术描述中所需要使用的附图作简单的介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的一种智算中心的算力资源部署方法的流程示意图；
图2为本申请实施例提供的一个示例性应用场景中的各生产阶段之间的关系示意图；
图3为本申请实施例提供的智算中心的算力资源部署装置的一种具体实施方式结构图；
图4为本申请实施例提供的电子设备的一种具体实施方式结构图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案，下面结合附图和具体实施方式对本申请作进一步的详细说明。显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等是用于区别不同的对象，而不是用于描述特定的顺序。此外术语“包括”和“具有”以及他们任何变形，意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、系统、产品或设备没有限定于已列出的步骤或单元，而是可包括没有列出的步骤或单元。
在介绍了本申请实施例的技术方案后，下面详细的说明本申请的各种非限制性实施方式。
首先参见图1，图1为本申请实施例提供的一种智算中心的算力资源部署方法的流程示意图，应用于智算中心，本申请实施例可包括以下内容：
S101：响应阶段划分指令，将所有待计划生产周期进行划分，以生成多个满足预设关联关系的生产阶段。
在本实施例中，阶段划分指令可通过人机交互页面进行下发，阶段划分指令包括阶段的划分标准单位和需要划分的周期，需要划分的周期也即所有待计划生产周期。所有待计划生产周期为执行算力资源部署操作的所有时间段，也就是生产计算力资源的所有时间段，本实施例将所有待计划生产周期作为一个整体时间段。算力资源可包括多种类型资源，计算力资 源例如可为计算节点，也即可采用计算节点的数量表示对计算力的衡量。进一步的，阶段划分指令还可携带算力资源类型字段。系统在接收到阶段划分指令之后，从该指令中提取划分标准单位和需要划分的周期，划分标准单位可根据实际需求也即需要划分得到的生产阶段、计算力资源需求量和待计划生产周期时长进行灵活选择。划分标准单位可为月、周、季度等等。按照划分标准单位将该整体时间段划分为多个时间阶段，每个时间阶段为一个生产阶段。举例来说，可将未来几个生产周期的计算力生产计划按月为单位进行分阶段划分。也即通过解析阶段划分指令，得到划分标准单位；按照划分标准单位，对所有待计划生产周期进行划分，得到多个生产阶段。本实施例在每个生产阶段均会生产计算力，各生产阶段共同组成所有待计划生产周期，所以各生产阶段的计算力产量在整个时间轴周期上具有一定的数值关系以及各生产阶段在整个生产周期上具有互相联系，上述各生产阶段之间的关系也即本步骤所称的预设关联关系。结合图2所示，每个生产阶段都有初始状态和结束状态，状态表示系统在某一生产阶段所处的位置或所处的状态，其结束状态为下一个阶段的初始状态；某一生产阶段的状态确定以后，从该状态演变到下一生产阶段状态所做的决定称为决策也即算力资源规划或者是称为计算力生产规划。某一个生产阶段的初始状态和决策确定后，下一个阶段的初始状态就随之而定。也就是说，每一个生产阶段都需要做出决策，每一个生产阶段的结果依赖于其初始状态和该生产阶段的决策。第一个生产阶段的初始状态为整个生产周期的初始状态，最后一个生产阶段的结束状态是整个生产周期的结束状态。对于各生产阶段均生产计算力资源的应用场景，各生产阶段均可包括计算力库存量和阶段结束计算力，计算力库存量即为每个生产阶段还未开始执行计算力资源生产之前，就已经具备的计算力资源，是每个生产阶段的库存量。阶段结束计算力是指经过该生产阶段进行计算力生产操作之后，所得到的所有计算力资源。第一个生产阶段的计算力库存量为所有待计划生产周期开始之前的库存量，最后一个生产阶段的计算力产量与计算力需求相同。当然，用户可通过人机交互界面预先下发在所有待计划生产周期内所需求的计算力资源总量，也即计算力需求。所需求的计算力资源部署在所有待计划生产周期内进行生产。也就是说，用户可根据自己的需要自行提出对计算力规模的需求，可以简单表述为需要多少计算节点，并将需求如交付时间和交付节点数量提交给智算中心；智算中心根据用户需求规划和指定计算力生产计划。这样所有待计划生产周期如未来几个月的计算力需求是已知的、可信的。每个生产阶段的阶段结束计算力为相邻的下一个生产阶段的计算力库存量；每个生产阶段的阶段结束计算力根据相应生产阶段的计算力库存量和计算力生产规划确定。
S102：根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表 征因子，生成效益与计算力生产数量关系。
在本步骤中，目标生产阶段可为上个步骤所划分的多个生产阶段中的任意一个生产阶段，计算力库存量为该目标生产阶段的库存量，固定生产计算力为目标生产阶段的用户需求计算力资源，计算力生产规划为该目标生产阶段的计划要生产的计算力资源。计算力生产规划与预先获取的计算力需求也即固定生产计算力并不完全相同，可以理解的是，为了保证整体效益最佳，每个生产阶段可能会生产比该生产阶段最初的用户需求的计算力资源多，也可能会比该生产阶段最初的用户需求计算力资源少，但是整体所需求的计算力资源必须要满足，这就需要每个生产阶段的固定生产计算力小于等于相应的计算力库存量和计算力生产规划之和。基于S101可知，将作为一个整体的所有待计划生产周期划分为多个不同的生产阶段，通过确定每个生产阶段相应的计算力生产规划，使得各生产阶段的计算力生产规划合起来为所有待计划生产周期的最优计算力生产规划。也就是说，将所有待计划生产周期作为一个整体再划分为若干个相互联系的过程，在每个生产阶段都需要做出决策，并且一个生产阶段的决策确定以后，会影响下一个生产阶段的决策，从而影响整个决策的结果。本实施例就是要在允许的范围内选择一个最优决策，使整个系统在预定标准下达到最佳效果。预定标准通常可基于效益来确定，也即在满足用户的计算力资源需求基础上实现整体效益最佳。所以需要使得在所有生产阶段的效益之和最佳时，确定每个生产阶段的效益与计算力生产数量关系。系统某个生产阶段的状态一经确定，执行该生产阶段的决策所得的效益即为阶段效益，阶段效益是整个系统效益的一部分，阶段效益可根据预先设置的效益信息来计算，可将效益信息存储至指定路径下，或者是用户输入效益信息。效益信息可包括效益表征因子、每个生产阶段的阶段效益与阶段计算力产量的函数关系。效益表征因子即为表征效益的参数，举例来说，效益为生产成本，则效益表征因子为生产成本系数，效益为仓储成本，则效益表征因子为仓储成本系数。阶段效益与阶段计算力产量的函数关系可为预先设置好的，以效益为生产成本为例，若设置每个生产阶段的生产成本与产量的平方成正比，也即为该生产阶段的阶段效益与阶段计算力产量的函数关系。
为了更加便于实施，智算中心计算力的算力资源生产过程可以为订单式生产。可根据用户需求规划计算力的生产，也即固定生产计算力可通过订单方式告知智算中心，当计算力资源为计算力节点，在根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系之前，当接收到生产订单请求，通过解析该生产订单请求，得到订单交付时间和交付计算节点总数；根据订单交付时间和交付计算节点总数，确定每个生产阶段的固定生产计算力，也即该生产阶段的用户需求计算力资源。
S103：当满足总效益最优条件时，基于效益与计算力生产数量关系，确定每个生产阶段的计算力生产规划。
在上个步骤确定目标生产阶段的效益与计算力生产数量关系之后，基于S101可知各生产阶段的关联关系，可基于目标生产阶段的效益与计算力生产数量关系依次计算得到其余各生产阶段的效益与计算力生产数量关系，在满足总效益最优条件如成本降低到最小时，逐步将生产计划分解，既要满足每个生产阶段的计算力需求，也要保证总的生产计划的总效益最高，通过基础数学知识依次计算得到每个生产阶段的计算力生产规划。在确定每个生产阶段的计算力成产规划之后，智算中心将用户所需求的所有计算力资源基于计算力生产规划部署在各生产阶段。
在本申请实施例提供的技术方案中，将所需计算力资源的未来生产周期按照指定标准进行分阶段，从而可将整个计算力生产计划部署在多个生产阶段，各生产阶段彼此联系又互相影响，在满足每个生产阶段的计算量需求的基础上，确定相应阶段的最优效益下的计算力生产规划，最终可确定最优效益下的算力资源生产计划，从而可最大程度地降低成本浪费，实现对智算中心算力资源的最优部署。
需要说明的是，本申请中各步骤间没有严格的先后执行顺序，只要符合逻辑上的顺序，则这些步骤可以同时执行，也可按照某种预设顺序执行，图1只是一种示意方式，并不代表只能是这样的执行顺序。
在上述实施例中，对于如何执行步骤S103并不做限定，本实施例中给出每个生产阶段的计算力生产规划的一种可选的确定方式，可包括如下步骤：
基于效益与计算力生产数量关系，按照从最后一个生产阶段到第一个生产阶段的顺序，依次确定以每个生产阶段为起点到整个生产过程结束的阶段效益；
当第一个生产阶段的阶段效益最大时，计算第一个生产阶段的第一计算力生产规划；
根据第一计算力生产规划、和相邻生产阶段的计算力库存量之间的关系，按照从第二个生产阶段到最后一个生产阶段的顺序，依次确定各生产阶段相应的计算力库存量和计算力生产规划。
在本实施例中，目标生产阶段可为最后一个生产阶段，基于最后一个生产阶段的效益与计算力生产数量关系，可以从终点一个生产阶段一个生产阶段地反推，在满足用户所需求的生产计划的前提下，得到每个生产阶段到终点的最优化的产量与效益；在计算出第一个生产阶段的最优生产计划后，可以顺序推算出每一生产阶段的生产计划，且总的效益最优。整个过程计算简单，易于实施。
在上述实施例中，对于如何执行步骤S102并不做限定，本实施例中给出效益与计算力生产数量关系的一种可选的确定方式，可包括如下步骤：
基于目标生产阶段的计算力库存量、计算力生产规划和效益表征因子，生成目标生产阶段的目标阶段效益；
根据目标生产阶段的计算力库存量、固定生产计算力和计算力生产规划，确定目标生产阶段的下一个生产阶段的计算力库存量，以用于生成目标生产阶段的下一个生产阶段的相邻阶段效益；
基于目标阶段效益和相邻阶段效益，生成用于表示效益与计算力生产数量关系的效益表征函数。
在本实施例中，效益表征函数可以定义在全过程上，也可以定义在后部分的子过程上，效益表征函数往往是各阶段效益的某种和式。执行最优策略后效益表征函数值也应当是最高或最低。可以理解的是，效益表征函数是系统执行某一策略所产生效益的数量表示，根据不同的实际情况，效益可以是效益的提升，如利润、距离、产量等；效益也可以是效益的降低，如生产成本、仓储成本、资源消耗等。基于此，本实施例基于不同效益类型给出了不同的效益表征函数，可包括下述内容：
若效益表征因子所属效益类型为降低型效益，降低型效益是指数值越小效益反而更优的类型，如生产成本、仓储成本、资源消耗等。调用第一效益表征函数关系式计算目标生产阶段的效益表征函数；第一效益表征函数关系式可表述为：
Figure PCTCN2022123402-appb-000003
式中，x k为目标生产阶段的计算力库存量，f k(x k)为目标生产阶段的目标生产阶段的效益表征函数，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，f k+1(x k+1)为目标生产阶段的下一个生产阶段的效益表征函数，N为生产阶段总数。
若效益表征因子所属效益类型为增长型效益，增长型效益是指数值越大效益更优的类型，如利润、距离、产量等。调用第二效益表征函数关系式计算目标生产阶段的效益表征函数；第二效益表征函数关系式可表示为：
Figure PCTCN2022123402-appb-000004
式中，x k为目标生产阶段的计算力库存量，f k(x k)为目标生产阶段的目标生产阶段的效益表征函数，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，f k+1(x k+1)为目标生产阶段的下一个生产阶段的效益表征函数，N为生产阶段总数。
由上可知，本实施例采用效益表征函数来定量表示效益与计算力生产数量关系，有利于提升整个生产规划部署效率，且给出了不同效益类型对应的效益表征函数，实用性更好。
为了使所属领域技术人员更加清楚明白本申请的技术方案，本申请还提供了一个示意性的例子，可包括下述内容：
在本示意性例子中，所有待计划生产周期为未来四个月，计算力资源为计算节点数量，用户在未来四个月的固定计算力生产订单需求为600台计算节点、700台计算节点、500台计算节点、1200台计算节点。本实施例的效益为成本，并将计算力成本简单归类为两类：仓储成本和生产成本，不做过于细化的划分，所有成本可以归为以上两类当中。实际的仓储成本和生产成本可能较复杂，在实际操作中可以定义仓储成本函数、生产成本函数，也可以简单地用系数描述，即仓储成本系数、生产成本系数。在目标生产阶段的目标阶段效益为生产成本的情况下，目标生产阶段的目标阶段效益根据下式确定：
d k(x k，u k)＝x k+β(u k) 2
式中，x k为目标生产阶段的计算力库存量，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，β为生产成本系数。
作为示例，若每台计算节点的仓储成本系数为1，生产成本系数为0.005，每个月的生产成本与产量的平方成正比，则每个生产阶段的成本可表示为：d k(x k,u k)＝x k+0.005(u k) 2。
基于上述实施例，未来计算力的生产计划是可以分阶段的，为便于理解和实施，本实施例以月度为单位对总生产计划分阶段，将未来四个月按月度划分为多个生产阶段，根据用户提交的交付时间和交付节点数量需求可以很容易得到每个月的计算力需求。第k个月月初已有的计算力为状态变量x k，第k个月需要生产的计算节点数量即计算力生产规划或者是说计算节点生产规划为决策变量u k；则从状态x k采取策略u k后的状态转移方程也即从x k到x k+1的状态转移方程为：x k+1＝x k+u k-A k；其中A k为每个月的固定计算节点的订单数量，是已知的，即A 1＝600，A 2＝700，A 3＝500，A 4＝1200。实际生产中已经下单进入生产计划的计算力必须满足，因此在采取计算节点生产规划u k后需要确保x k+1不能为负。本实施例的阶段效益体现在生产成本上，每个生产阶段的生产成本可表示为：d k(x k,u k)＝x k+0.005(u k) 2。目标生产阶段为最后一个生产阶段，目标生产阶段的效益表征函数可为：
Figure PCTCN2022123402-appb-000005
其中，f k(x k)定义为从状态x k出发到整个生产过程结束后的成本，且要满足x k+u k≥A k。那么本实施例要解决的技术问题即为在满足每月订单数量的前提下，也即根据智算中心未来几个生产周期的计算力需求，如何制定计算节点的生产计划，将成本降低到最小。本实施例从最后一个生产阶段开始回溯，计算得到第一个生产阶段的计算节点生产规划，再顺序计算得到每一个生产阶段的计算节点生产规划，计算过程可如下所示：
从最后一个生产阶段也即k＝4计算，因最后一个生产阶段计算力的产量即计算节点总数刚好满足当前生产阶段的需求，对f 4(x 4)＝min{x 4+0.005(u 4) 2}求极值，则可得到u 4＝A 4-x 4＝1200-x 4，进而可计算得到f 4(x 4)＝7200-11x 4+0.005(x 4) 2。本实施例要确定每个生产阶段的计算节点生产规划u k，即每个生产阶段应该生产多少计算节点，每个生产阶段的初始状态x k，即当前库存计算节点数量视为已知量。基于效益表征函数和状态转移函数计算k＝3对应生产阶段的数据：该生产阶段的计算力生产规划为u 3，f 3(x 3)＝min{x 3+0.005(u 3) 2+7200-11(x 3+u 3-500)+0.005(x 3+u 3-500) 2}。当f 3(x 3)最小时，f 3(x 3)的导函数为0，据此可对f 3(x 3)求导，得到x 3和u 3的关系为0.01u 3-11+0.01(x 3+u 3-500)＝0，也即u 3＝800-0.5x 3，通过将u 3代入f 3(x 3)消掉u 3可计算得到f 3(x 3)＝7550-7x 3+0.0025(x 3) 2。依此类推，再计算k＝2时的f 2(x 2)，对f 2(x 2)求导可得u 2＝700-x 2/3，代入可得f 2(x 2)＝10000-6x 2+0.005(x 2) 2/3。为了使说明书更加简明扼要，具体计算过程请参阅k＝3时计算过程，此处便不再赘述。最后计算k＝1时的情况。整个问题的效益表征函数即f 1(x 1)，那么最优的生产调度决策下f 1(x 1)一定是收益最高的也即对f 1(x 1)＝min{x 1+0.005u 1 2+10000-6(x 1+u 1-600)+0.005(x 1+u 1-600) 2/3}求导可以得到0.01u 1-6+0.01(x 1+u 1-600)/3＝0。其中，x 1是已知量，即智算中心当前已有的库存计算节点是已知的，这是整个问题的初始状态。假设x 1＝0，根据上述求导后关系式可以计算得到u 1＝600，即在整个生产部署过程中，第一生产阶段的计算力产量应当为600台；f 1(x 1)＝11800，即最优生产调度决策的成本为11800，这就是最低成本。得到u 1后，根据每个生产阶段的状态转移方程、求导得到的u k和x k的关系式，便可得到每一生产阶段的库存量和最优生产策略：x 1＝0，x 2＝0，x 3＝0，x 4＝300，u 1＝600，u 2＝700，u 3＝800，u 4＝900。应用本实施例的技术方案，总的生产成本最少，为11800。
在实际应用场景中，可将上述方法所依赖的计算机程序形成为智算中心的算力资源部署 程序内嵌于处理器中，通过程序化运行便可确定算力资源的部署情况，不仅可满足用户的生产需求，而且可保证效益最优。
本申请实施例还针对智算中心的算力资源部署方法提供了相应的装置，进一步使得方法更具有实用性。其中，装置可从功能模块的角度和硬件的角度分别说明。下面对本申请实施例提供的智算中心的算力资源部署装置进行介绍，下文描述的智算中心的算力资源部署装置与上文描述的智算中心的算力资源部署方法可相互对应参照。
基于功能模块的角度，参见图3，图3为本申请实施例提供的智算中心的算力资源部署装置在一种具体实施方式下的结构图，应用于智算中心，该装置可包括：
生产分段模块301，用于响应阶段划分指令，将所有待计划生产周期进行划分，以生成多个满足预设关联关系的生产阶段；
效益与产量关系确定模块302，用于根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系；每个生产阶段的固定生产计算力小于等于相应的计算力库存量和计算力生产规划之和；
生产规划生成模块303，用于当满足总效益最优条件时，基于效益与计算力生产数量关系，确定每个生产阶段的计算力生产规划。
可选的，在本实施例的一些实施方式中，上述生产规划生成模块303还可用于：基于效益与计算力生产数量关系，按照从最后一个生产阶段到第一个生产阶段的顺序，依次确定以每个生产阶段为起点到整个生产过程结束的阶段效益；当第一个生产阶段的阶段效益最大时，计算第一个生产阶段的第一计算力生产规划；根据第一计算力生产规划、和相邻生产阶段的计算力库存量之间的关系，按照从第二个生产阶段到最后一个生产阶段的顺序，依次确定各生产阶段相应的计算力库存量和计算力生产规划。
作为本实施例的一个可选的实施方式，上述生产分段模块301可进一步用于：通过解析阶段划分指令，得到划分标准单位；按照划分标准单位，对所有待计划生产周期进行划分，得到多个生产阶段；其中，各生产阶段包括计算力库存量和阶段结束计算力，每个生产阶段的阶段结束计算力为相邻的下一个生产阶段的计算力库存量；每个生产阶段的阶段结束计算力根据相应生产阶段的计算力库存量和计算力生产规划确定。
作为本实施例的一个可选的实施方式，上述装置还包括需求获取模块，用于：接收需求信息，需求信息包括计算力需求，计算力需求为所有待计划生产周期内所需求的计算力资源总量；其中，第一个生产阶段的计算力库存量为所有待计划生产周期开始之前的库存量，最后一个生产阶段的计算力产量与计算力需求相同。
作为本实施例的一个可选的实施方式，需求信息还包括计算力交付时间和交付量；
上述装置还包括固定计算力获取模块，用于：根据计算力交付时间和交付量，确定各生产阶段的固定生产计算力。
作为本实施例的另外一种可选的实施方式，上述装置还可包括效益信息获取模块，用于获取效益信息，以根据效益信息，确定每个生产阶段的阶段效益；其中，效益信息包括效益表征因子、每个生产阶段的阶段效益与阶段计算力产量的函数关系。
作为本实施例的再一种可选的实施方式，上述装置例如还可包括固定生产计算力确定模块，用于当接收到生产订单请求，通过解析生产订单请求，得到订单交付时间和交付计算节点总数；根据订单交付时间和交付计算节点总数，确定每个生产阶段的固定生产计算力。
可选的，在本实施例的一些实施方式中，上述效益与产量关系确定模块302还可用于：基于目标生产阶段的计算力库存量、计算力生产规划和效益表征因子，生成目标生产阶段的目标阶段效益；根据目标生产阶段的计算力库存量、固定生产计算力和计算力生产规划，确定目标生产阶段的下一个生产阶段的计算力库存量，以用于生成目标生产阶段的下一个生产阶段的相邻阶段效益；基于目标阶段效益和相邻阶段效益，生成用于表示效益与计算力生产数量关系的效益表征函数。
作为上述实施例的一种可选的实施方式，该效益与产量关系确定模块302可进一步用于：若效益表征因子所属效益类型为降低型效益，调用第一效益表征函数关系式计算目标生产阶段的效益表征函数；第一效益表征函数关系式为：
Figure PCTCN2022123402-appb-000006
式中，x k为目标生产阶段的计算力库存量，f k(x k)为目标生产阶段的目标生产阶段的效益表征函数，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，f k+1(x k+1)为目标生产阶段的下一个生产阶段的效益表征函数，N为生产阶段总数。
作为上述实施例的另一种可选的实施方式，该效益与产量关系确定模块302还可进一步用于：若效益表征因子所属效益类型为增长型效益，调用第二效益表征函数关系式计算目标生产阶段的效益表征函数；第二效益表征函数关系式为：
Figure PCTCN2022123402-appb-000007
式中，x k为目标生产阶段的计算力库存量，f k(x k)为目标生产阶段的目标生产阶段的效益表征函数，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，f k+1(x k+1)为目标生产阶段的下一个生产阶段的效益表征函数，N为生产阶段总数。
作为本实施例的一个可选的实施方式，上述生产规划生成模块还用于：根据效益表征函数计算第一生产阶段的阶段效益，对第一生产阶段的阶段效益求导，确定第一生产阶段的计算力生产规划与计算力库存量之间的函数关系；基于第一生产阶段的计算力生产规划与计算力库存量之间的函数关系，确定第一生产阶段的第一计算力生产规划。
作为本实施例的一个可选的实施方式，降低型效益包括以下一种或多种：生产成本、仓储成本、资源消耗。
作为本实施例的一个可选的实施方式，增长性效益包括以下一种或多种：利润、距离、产量。
作为本实施例的一个可选的实施方式，上述效益与产量关系确定模块还用于：在效益表征因子为生产成本系数的情况下，每个生产阶段的阶段效益与阶段计算力产量的函数关系为每个生产阶段的生产成本与阶段计算力产量的平方成正比的函数关系；根据生产成本系数、每个生产阶段的生产成本与阶段计算力产量的平方成正比的函数关系，确定每个生产阶段的阶段效益，每个生产阶段的阶段效益为相应阶段的生产成本。
作为本实施例的一个可选的实施方式，上述效益与产量关系确定模块还用于：在目标生产阶段的目标阶段效益为生产成本的情况下，目标生产阶段的目标阶段效益根据下式确定：
d k(x k，u k)＝x k+β(u k) 2
式中，x k为目标生产阶段的计算力库存量，u k为目标生产阶段的计算力生产规划，d k(x k，u k)为目标阶段效益，β为生产成本系数。
作为本实施例的一个可选的实施方式，上述效益与产量关系确定模块还用于：根据下式确定目标生产阶段的下一个生产阶段的计算力库存量：
x k+1＝x k+u k-A k
式中，x k+1为目标生产阶段的下一个生产阶段的计算力库存量，A k为目标生产阶段的固定生产计算力。
本申请实施例智算中心的算力资源部署装置的各功能模块的功能可根据上述方法实施例中的方法具体实现，其具体实现过程可以参照上述方法实施例的相关描述，此处不再赘述。
由上可知，本申请实施例可实现对智算中心计算力资源的最优部署。
上文中提到的智算中心的算力资源部署装置是从功能模块的角度描述，进一步的，本申 请还提供一种电子设备，是从硬件角度描述。图4为本申请实施例提供的电子设备在一种实施方式下的结构示意图。如图4所示，该电子设备包括存储器40，用于存储计算机程序；处理器41，用于执行计算机程序时实现如上述任一实施例提到的智算中心的算力资源部署方法的步骤。
其中，处理器41可以包括一个或多个处理核心，比如4核心处理器、8核心处理器，处理器41还可为控制器、微控制器、微处理器或其他数据处理芯片等。处理器41可以采用DSP(Digital Signal Processing，数字信号处理)、FPGA(Field－Programmable Gate Array，现场可编程门阵列)、PLA(Programmable Logic Array，可编程逻辑阵列)中的至少一种硬件形式来实现。处理器41也可以包括主处理器和协处理器，主处理器是用于对在唤醒状态下的数据进行处理的处理器，也称CPU(Central Processing Unit，中央处理器)；协处理器是用于对在待机状态下的数据进行处理的低功耗处理器。在一些实施例中，处理器41可以集成有GPU(Graphics Processing Unit，图像处理器)，GPU用于负责显示屏所需要显示的内容的渲染和绘制。一些实施例中，处理器41还可以包括AI(Artificial Intelligence，人工智能)处理器，该AI处理器用于处理有关机器学习的计算操作。
存储器40可以包括一个或多个计算机非易失性可读存储介质，该计算机非易失性可读存储介质可以是非暂态的。存储器40还可包括高速随机存取存储器以及非易失性存储器，比如一个或多个磁盘存储设备、闪存存储设备。存储器40在一些实施例中可以是电子设备的内部存储单元，例如服务器的硬盘。存储器40在另一些实施例中也可以是电子设备的外部存储设备，例如服务器上配备的插接式硬盘，智能存储卡(Smart Media Card，SMC)，安全数字(Secure Digital，SD)卡，闪存卡(Flash Card)等。进一步地，存储器40还可以既包括电子设备的内部存储单元也包括外部存储设备。存储器40不仅可以用于存储安装于电子设备的应用软件及各类数据，例如：执行漏洞处理方法的程序的代码等，还可以用于暂时地存储已经输出或者将要输出的数据。本实施例中，存储器40至少用于存储以下计算机程序401，其中，该计算机程序被处理器41加载并执行之后，能够实现前述任一实施例公开的智算中心的算力资源部署方法的相关步骤。另外，存储器40所存储的资源还可以包括操作系统402和数据403等，存储方式可以是短暂存储或者永久存储。其中，操作系统402可以包括Windows、Unix、Linux等。数据403可以包括但不限于智算中心的算力资源部署结果对应的数据等。
在一些实施例中，上述电子设备还可包括有显示屏42、输入输出接口43、通信接口44或者称为网络接口、电源45以及通信总线46。其中，显示屏42、输入输出接口43比如键 盘(Keyboard)属于用户接口，可选的用户接口还可以包括标准的有线接口、无线接口等。可选地，在一些实施例中，显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode，有机发光二极管)触摸器等。显示器也可以适当的称为显示屏或显示单元，用于显示在电子设备中处理的信息以及用于显示可视化的用户界面。通信接口44可选的可以包括有线接口和/或无线接口，如WI-FI接口、蓝牙接口等，通常用于在电子设备与其他电子设备之间建立通信连接。通信总线46可以是外设部件互连标准(peripheral component interconnect，简称PCI)总线或扩展工业标准结构(extended industry standard architecture，简称EISA)总线等。该总线可以分为地址总线、数据总线、控制总线等。为便于表示，图4中仅用一条粗线表示，但并不表示仅有一根总线或一种类型的总线。
本领域技术人员可以理解，图4中示出的结构并不构成对该电子设备的限定，可以包括比图示更多或更少的组件，例如还可包括实现各类功能的传感器47。
本申请实施例电子设备的各功能模块的功能可根据上述方法实施例中的方法具体实现，其具体实现过程可以参照上述方法实施例的相关描述，此处不再赘述。
由上可知，本申请实施例可实现对智算中心计算力资源的最优部署。
可以理解的是，如果上述实施例中的智算中心的算力资源部署方法以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，执行本申请各个实施例方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read-Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、电可擦除可编程ROM、寄存器、硬盘、多媒体卡、卡型存储器(例如SD或DX存储器等)、磁性存储器、可移动磁盘、CD-ROM、磁碟或者光盘等各种可以存储程序代码的介质。
基于此，本申请实施例还提供了一种非易失性可读存储介质，存储有计算机程序，计算机程序被处理器执行时如上任意一实施例智算中心的算力资源部署方法的步骤。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其它实施例的不同之处，各个实施例之间相同或相似部分互相参见即可。对于实施例公开的硬件包括装置及电子设备而言，由于其与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
专业人员还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法 步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
以上对本申请所提供的一种智算中心的算力资源部署方法、装置、电子设备及非易失性可读存储介质进行了详细介绍。本文中应用了具体个例对本申请的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本申请的方法及其核心思想。应当指出，对于本技术领域的普通技术人员来说，在不脱离本申请原理的前提下，还可以对本申请进行若干改进和修饰，这些改进和修饰也落入本申请权利要求的保护范围内。

Claims (20)

  1. 一种智算中心的算力资源部署方法，其中，应用于智算中心，包括：
    响应阶段划分指令，将所有待计划生产周期进行划分，以生成多个满足预设关联关系的生产阶段；
    根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系；
    当满足总效益最优条件时，基于所述效益与计算力生产数量关系，确定每个生产阶段的计算力生产规划；
    其中，每个生产阶段的固定生产计算力小于等于相应的计算力库存量和计算力生产规划之和。
  2. 根据权利要求1所述的智算中心的算力资源部署方法，其中，所述当满足总效益最优条件时，基于所述效益与计算力生产数量关系，确定每个生产阶段的计算力生产规划，包括：
    基于所述效益与计算力生产数量关系，按照从最后一个生产阶段到第一个生产阶段的顺序，依次确定以每个生产阶段为起点到整个生产过程结束的阶段效益；
    当所述第一个生产阶段的阶段效益最大时，计算所述第一个生产阶段的第一计算力生产规划；
    根据所述第一计算力生产规划和相邻生产阶段的计算力库存量之间的关系，按照从第二个生产阶段到最后一个生产阶段的顺序，依次确定各生产阶段相应的计算力库存量和计算力生产规划。
  3. 根据权利要求2所述的智算中心的算力资源部署方法，其中，所述将所有待计划生产周期进行划分，以生成多个满足预设关联关系的生产阶段，包括：
    通过解析所述阶段划分指令，得到划分标准单位；
    按照所述划分标准单位，对所有待计划生产周期进行划分，得到多个生产阶段；
    其中，各生产阶段包括计算力库存量和阶段结束计算力，每个生产阶段的阶段结束计算力为相邻的下一个生产阶段的计算力库存量；每个生产阶段的阶段结束计算力根据相应生产阶段的计算力库存量和计算力生产规划确定。
  4. 根据权利要求3所述的智算中心的算力资源部署方法，其中，所述方法还包括：
    接收需求信息，所述需求信息包括计算力需求，所述计算力需求为所述所有待计划生产周期内所需求的计算力资源总量；
    其中，所述第一个生产阶段的计算力库存量为所有待计划生产周期开始之前的库存量，所述最后一个生产阶段的计算力产量与所述计算力需求相同。
  5. 根据权利要求4所述的智算中心的算力资源部署方法，其中，所述需求信息还包括计算力交付时间和交付量；
    所述根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系之前，还包括：
    根据所述计算力交付时间和所述交付量，确定各生产阶段的固定生产计算力。
  6. 根据权利要求1所述的智算中心的算力资源部署方法，其中，所述根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系之前，还包括：
    获取效益信息，以根据所述效益信息，确定每个生产阶段的阶段效益；
    其中，所述效益信息包括效益表征因子、每个生产阶段的阶段效益与阶段计算力产量的函数关系。
  7. 根据权利要求5所述的智算中心的算力资源部署方法，其中，所述根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系之前，还包括：
    当接收到生产订单请求，通过解析所述生产订单请求，得到订单交付时间和交付计算节点总数；
    根据所述订单交付时间和所述交付计算节点总数，确定每个生产阶段的固定生产计算力。
  8. 根据权利要求1至7任意一项所述的智算中心的算力资源部署方法，其中，所述根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系，包括：
    基于目标生产阶段的计算力库存量、计算力生产规划和效益表征因子，生成所述目标生产阶段的目标阶段效益；
    根据目标生产阶段的计算力库存量、固定生产计算力和计算力生产规划，确定所述目标生产阶段的下一个生产阶段的计算力库存量，以用于生成所述目标生产阶段的下一个生产阶段的相邻阶段效益；
    基于所述目标阶段效益和所述相邻阶段效益，生成用于表示效益与计算力生产数量关系的效益表征函数。
  9. 根据权利要求8所述的智算中心的算力资源部署方法，其中，所述基于所述目标 阶段效益和所述相邻阶段效益，生成用于表示效益与计算力生产数量关系的效益表征函数，包括：
    若所述效益表征因子所属效益类型为降低型效益，调用第一效益表征函数关系式计算所述目标生产阶段的效益表征函数；所述第一效益表征函数关系式为：
    Figure PCTCN2022123402-appb-100001
    式中，x k为所述目标生产阶段的计算力库存量，f k(x k)为所述目标生产阶段的目标生产阶段的效益表征函数，u k为所述目标生产阶段的计算力生产规划，d k(x k，u k)为所述目标阶段效益，f k+1(x k+1)为所述目标生产阶段的下一个生产阶段的效益表征函数，N为生产阶段总数。
  10. 根据权利要求8所述的智算中心的算力资源部署方法，其中，所述基于所述目标阶段效益和所述相邻阶段效益，生成用于表示效益与计算力生产数量关系的效益表征函数，包括：
    若所述效益表征因子所属效益类型为增长型效益，调用第二效益表征函数关系式计算所述目标生产阶段的效益表征函数；所述第二效益表征函数关系式为：
    Figure PCTCN2022123402-appb-100002
    式中，x k为所述目标生产阶段的计算力库存量，f k(x k)为所述目标生产阶段的目标生产阶段的效益表征函数，u k为所述目标生产阶段的计算力生产规划，d k(x k，u k)为所述目标阶段效益，f k+1(x k+1)为所述目标生产阶段的下一个生产阶段的效益表征函数，N为生产阶段总数。
  11. 根据权利要求9或10所述的智算中心的算力资源部署方法，其中，所述当所述第一个生产阶段的阶段效益最大时，计算所述第一个生产阶段的第一计算力生产规划，包括：
    根据所述效益表征函数计算所述第一生产阶段的阶段效益，对所述第一生产阶段的阶段效益求导，确定所述第一生产阶段的计算力生产规划与计算力库存量之间的函数关系；
    基于所述第一生产阶段的计算力生产规划与计算力库存量之间的函数关系，确定所 述第一生产阶段的第一计算力生产规划。
  12. 根据权利要求9所述的智算中心的算力资源部署方法，其中，所述降低型效益包括以下一种或多种：生产成本、仓储成本、资源消耗。
  13. 根据权利要求10所述的智算中心的算力资源部署方法，其中，所述增长性效益包括以下一种或多种：利润、距离、产量。
  14. 根据权利要求12所述的智算中心的算力资源部署方法，其中，所述根据所述效益信息，确定每个生产阶段的阶段效益，包括：
    在所述效益表征因子为生产成本系数的情况下，所述每个生产阶段的阶段效益与阶段计算力产量的函数关系为每个生产阶段的生产成本与阶段计算力产量的平方成正比的函数关系；
    根据所述生产成本系数、所述每个生产阶段的生产成本与阶段计算力产量的平方成正比的函数关系，确定每个生产阶段的阶段效益，所述每个生产阶段的阶段效益为相应阶段的生产成本。
  15. 根据权利要求14所述的智算中心的算力资源部署方法，其中，基于目标生产阶段的计算力库存量、计算力生产规划和效益表征因子，生成所述目标生产阶段的目标阶段效益，包括：
    在所述目标生产阶段的目标阶段效益为生产成本的情况下，所述目标生产阶段的目标阶段效益根据下式确定：
    d k(x k，u k)＝x k+β(u k) 2
    式中，x k为所述目标生产阶段的计算力库存量，u k为所述目标生产阶段的计算力生产规划，d k(x k，u k)为所述目标阶段效益，β为生产成本系数。
  16. 根据权利要求15所述的智算中心的算力资源部署方法，其中，所述根据目标生产阶段的计算力库存量、固定生产计算力和计算力生产规划，确定所述目标生产阶段的下一个生产阶段的计算力库存量，包括：
    根据下式确定所述目标生产阶段的下一个生产阶段的计算力库存量：
    x k+1＝x k+u k-A k
    式中，x k+1为所述目标生产阶段的下一个生产阶段的计算力库存量，A k为所述目标生产阶段的固定生产计算力。
  17. 一种智算中心的算力资源部署装置，其中，应用于智算中心，包括：
    生产分段模块，用于响应阶段划分指令，将所有待计划生产周期进行划分，以生成多个满足预设关联关系的生产阶段；
    效益与产量关系确定模块，用于根据目标生产阶段的计算力库存量、固定生产计算力、计算力生产规划和效益表征因子，生成效益与计算力生产数量关系；每个生产阶段的固定生产计算力小于等于相应的计算力库存量和计算力生产规划之和；
    生产规划生成模块，用于当满足总效益最优条件时，基于所述效益与计算力生产数量关系，确定每个生产阶段的计算力生产规划。
  18. 根据权利要求17所述的智算中心的算力资源部署装置，其中，所述生产规划生成模块还用于：
    基于所述效益与计算力生产数量关系，按照从最后一个生产阶段到第一个生产阶段的顺序，依次确定以每个生产阶段为起点到整个生产过程结束的阶段效益；
    当所述第一个生产阶段的阶段效益最大时，计算所述第一个生产阶段的第一计算力生产规划；
    根据所述第一计算力生产规划和相邻生产阶段的计算力库存量之间的关系，按照从第二个生产阶段到最后一个生产阶段的顺序，依次确定各生产阶段相应的计算力库存量和计算力生产规划。
  19. 一种电子设备，其中，包括处理器和存储器，所述处理器用于执行所述存储器中存储的计算机程序时实现如权利要求1至16任一项所述智算中心的算力资源部署方法的步骤。
  20. 一种非易失性可读存储介质，其中，所述非易失性可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时实现如权利要求1至16任一项所述智算中心的算力资源部署方法的步骤。
PCT/CN2022/123402 2022-05-09 2022-09-30 智算中心的算力资源部署方法、装置、设备及存储介质 WO2023216500A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210495846.1 2022-05-09
CN202210495846.1A CN114596009B (zh) 2022-05-09 2022-05-09 智算中心的算力资源部署方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2023216500A1 true WO2023216500A1 (zh) 2023-11-16

Family

ID=81812603

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/123402 WO2023216500A1 (zh) 2022-05-09 2022-09-30 智算中心的算力资源部署方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN114596009B (zh)
WO (1) WO2023216500A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114596009B (zh) * 2022-05-09 2022-07-22 苏州浪潮智能科技有限公司 智算中心的算力资源部署方法、装置、设备及存储介质

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170371636A1 (en) * 2016-06-27 2017-12-28 Vmware, Inc. Methods and systems to optimize cost and automate a development and operations deployment pipeline
CN109583749A (zh) * 2018-11-27 2019-04-05 李伟 一种基于动态规划的软件研制成本智能控制方法及系统
CN111680877A (zh) * 2020-05-06 2020-09-18 杭州传化智能制造科技有限公司 产线调度方法、装置、计算机设备和计算机可读存储介质
CN112650590A (zh) * 2020-12-29 2021-04-13 北京奇艺世纪科技有限公司 任务的处理方法、装置及系统、分配方法和装置
CN112769641A (zh) * 2020-12-24 2021-05-07 电子科技大学长三角研究院(衢州) 面向智能数据处理的区块链算力优化调度方法
CN113986535A (zh) * 2021-10-19 2022-01-28 北京三快在线科技有限公司 算力资源的调节方法、装置、存储介质和电子设备
CN114416352A (zh) * 2021-12-29 2022-04-29 中国电信股份有限公司 算力资源分配方法、装置、电子设备及储存介质
CN114596009A (zh) * 2022-05-09 2022-06-07 苏州浪潮智能科技有限公司 智算中心的算力资源部署方法、装置、设备及存储介质

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170371636A1 (en) * 2016-06-27 2017-12-28 Vmware, Inc. Methods and systems to optimize cost and automate a development and operations deployment pipeline
CN109583749A (zh) * 2018-11-27 2019-04-05 李伟 一种基于动态规划的软件研制成本智能控制方法及系统
CN111680877A (zh) * 2020-05-06 2020-09-18 杭州传化智能制造科技有限公司 产线调度方法、装置、计算机设备和计算机可读存储介质
CN112769641A (zh) * 2020-12-24 2021-05-07 电子科技大学长三角研究院(衢州) 面向智能数据处理的区块链算力优化调度方法
CN112650590A (zh) * 2020-12-29 2021-04-13 北京奇艺世纪科技有限公司 任务的处理方法、装置及系统、分配方法和装置
CN113986535A (zh) * 2021-10-19 2022-01-28 北京三快在线科技有限公司 算力资源的调节方法、装置、存储介质和电子设备
CN114416352A (zh) * 2021-12-29 2022-04-29 中国电信股份有限公司 算力资源分配方法、装置、电子设备及储存介质
CN114596009A (zh) * 2022-05-09 2022-06-07 苏州浪潮智能科技有限公司 智算中心的算力资源部署方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN114596009A (zh) 2022-06-07
CN114596009B (zh) 2022-07-22

Similar Documents

Publication Publication Date Title
US20230035451A1 (en) Resource usage prediction for deep learning model
US20170286864A1 (en) Batching inputs to a machine learning model
WO2022048557A1 (zh) Ai模型的训练方法、装置、计算设备和存储介质
US10394302B2 (en) Self-service equipment energy saving control method and device
Yoo et al. Integrated scheduling of real-time and interactive tasks for configurable industrial systems
CN104142909A (zh) 一种汉字注音方法及装置
CN110795238B (zh) 负载计算方法、装置、存储介质及电子设备
CN113886080B (zh) 高性能集群任务调度方法、装置、电子设备及存储介质
WO2023216500A1 (zh) 智算中心的算力资源部署方法、装置、设备及存储介质
CN113515382A (zh) 云资源的分配方法、装置、电子设备及程序产品
CN110134215A (zh) 数据处理方法、装置、电子设备及可读存储介质
CN114968520B (zh) 任务搜索方法及装置、服务器和存储介质
CN114860412B (zh) 任务处理方法、装置、电子设备和介质
EP4222598A1 (en) Optimizing job runtimes via prediction-based token allocation
CN113791882B (zh) 多任务的部署方法、装置、电子设备及存储介质
CN114091589B (zh) 模型训练方法、装置、电子设备及介质
CN118364918A (zh) 一种大语言模型的推理方法、装置、设备及存储介质
US12033215B2 (en) Portfolio performance prediction
CN111274667B (zh) 一种跨尺度材料计算软件集成计算系统及方法
CN116993861A (zh) 图案生成方法、装置和电子设备
CN113112311B (zh) 训练因果推断模型的方法、信息提示方法以装置
CN113408641B (zh) 资源生成模型的训练与服务资源的生成方法、装置
CN115470900A (zh) 一种神经网络模型的剪枝方法、装置及设备
CN114153592A (zh) 云平台的物理主机负载调度方法、装置、电子设备及介质
CN117519996B (zh) 一种数据处理方法、装置、设备以及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22941439

Country of ref document: EP

Kind code of ref document: A1