Method and device for measuring computing power of intelligent computing center by 1 degree computing power

Abstract

The present invention provides a method and device for measuring the computing power of an intelligent computing center through 1 degree computing power, which relates to the field of computing power technology. The method comprises: step S1, obtaining the initial half-precision floating-point computing power number, initial video memory bandwidth and initial video memory capacity of a processor of a first intelligent computing center; step S2, calculating the first ratio of the initial half-precision floating-point computing power number to the preset half-precision floating-point computing power number, calculating the second ratio of the initial video memory bandwidth to the preset video memory bandwidth, and calculating the third ratio of the initial video memory capacity to the preset video memory capacity; step S3, setting the weighted sum of the first ratio, the second ratio and the third ratio, the preset floating-point operation number unit and the preset time unit as the initial 1 degree computing power value of the first intelligent computing center. The present invention can accurately measure the computing power of the intelligent computing center through the 1 degree computing power value.

Description

Technical Field

[0001] The present invention relates to the field of computing power technology, and in particular to a method and device for measuring the computing power of an intelligent computing center by using 1 degree of computing power. Background Art

[0002] Computing power is the ability of computer equipment or computing / data centers to process information. It is the ability of computer hardware and software to work together to perform certain computing needs. It is the computing power to achieve the output of target results by processing information data. It is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to the society through computing power infrastructure. Computing power infrastructure is an important part of the new information infrastructure. It presents the characteristics of diversity, ubiquity, intelligence, agility, security, reliability, green and low-carbon. It is of great significance to promote industrial transformation and upgrading, enable scientific and technological innovation and progress, meet people's needs for a better life, and achieve high-efficiency social governance.

[0003] Currently, in scenarios where model training is required, the intelligent computing center usually allocates processor resources to the model, and then processes and calculates the model based on the processor resources to achieve model training.

[0004] However, in the prior art, there is no perfect unit of measurement for the computing power of the intelligent computing center, and the computing power of the intelligent computing center cannot be conveniently measured. The computing power of the intelligent computing center can only be roughly estimated, resulting in redundant or insufficient computing power allocated to the intelligent computing center during model training in the prior art.

[0005] It can be seen that the existing technology has the problem of low accuracy in allocating processor resources for model training and poor accuracy in measuring the computing power of the intelligent computing center. Summary of the invention

[0006] The embodiments of the present invention provide a method and device for measuring the computing power of an intelligent computing center by 1 degree of computing power, so as to solve the problems in the prior art of low accuracy of processor resource allocation for model training and poor accuracy of measuring the computing power of an intelligent computing center.

[0007] To solve the above problems, the present invention is achieved as follows:

[0008] In a first aspect, an embodiment of the present invention provides a method for measuring the computing power of an intelligent computing center by using 1 degree of computing power, which is applied to a first intelligent computing center and includes:

[0009] Step S1, obtaining an initial half-precision floating-point computing power, an initial video memory bandwidth, and an initial video memory capacity of a processor of the first intelligent computing center;

[0010] Step S2, calculating a first ratio of the initial half-precision floating-point computing power to a preset half-precision floating-point computing power, calculating a second ratio of the initial video memory bandwidth to a preset video memory bandwidth, and calculating a third ratio of the initial video memory capacity to a preset video memory capacity, wherein the preset half-precision floating-point computing power is the half-precision floating-point computing power of a preset processor, the preset video memory bandwidth is the video memory bandwidth of the preset processor, and the preset video memory capacity is the video memory capacity of the preset processor;

[0011] Step S3: set the product of the weighted sum of the first ratio, the second ratio and the third ratio, the preset floating-point operation number unit and the preset time unit as the initial 1-degree computing power value of the first intelligent computing center.

[0012] In one embodiment, after step S3, the method further includes:

[0013] Step S4, obtaining model parameters of the initial model;

[0014] Step S5: Calculate the required 1-degree computing power value of the initial model based on the model parameters, where the required 1-degree computing power value is used to characterize the computing power required for training the initial model;

[0015] Step S6: Perform model training on the initial model based on the required 1-degree computing power value and the initial 1-degree computing power value.

[0016] In one embodiment, the model parameters include the size of the initial model and the size of the training data set;

[0017] The step S5 comprises:

[0018] Step S51: Calculate the computing power required for the initial model based on the size of the initial model and the size of the training data set.

[0019] In one embodiment, step S6 includes:

[0020] Step S61: Calculate the remaining 1-degree computing power value of the first intelligent computing center based on the initial 1-degree computing power;

[0021] Step S62: when the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model, allocating processor resources based on the required 1-degree computing power value;

[0022] Step S63: Perform model training on the initial model based on the processor resources.

[0023] In one embodiment, the step S6 further includes:

[0024] Step S64: When the remaining 1-degree computing power value is less than the required 1-degree computing power value of the initial model, the initial model, the model parameters and the required 1-degree computing power value are sent to the second intelligent computing center;

[0025] Step S65, receiving the target model sent by the second intelligent computing center, wherein the target model is obtained by allocating the processor resources by the second intelligent computing center based on the required 1-degree computing power value, and training the initial model based on the processor resources and the model parameters.

[0026] In one embodiment, the step S61 includes:

[0027] Step S611, obtaining the training 1 degree computing power value of the first intelligent computing center for training other models;

[0028] Step S312: Set the difference between the initial 1-degree computing power value and the training 1-degree computing power value as the remaining 1-degree computing power value.

[0029] In one embodiment, the processor resources include the target half-precision floating-point computing power corresponding to the required 1-degree computing power value, the target video memory bandwidth, and the target video memory capacity; the step S62 includes:

[0030] Step S621: When the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model, calculate the target half-precision floating-point computing power number, the target video memory bandwidth, and the target video memory capacity corresponding to the required 1-degree computing power value;

[0031] Step S622: Allocate the target half-precision floating-point computing power, the target video memory bandwidth, and the target video memory capacity.

[0032] In a second aspect, an embodiment of the present invention further provides a device for measuring the computing power of an intelligent computing center by 1 degree of computing power, including:

[0033] A first acquisition module is used to acquire an initial half-precision floating-point computing power, an initial video memory bandwidth, and an initial video memory capacity of a processor of a first intelligent computing center;

[0034] a first calculation module, configured to calculate a first ratio of the initial half-precision floating-point computing power to a preset half-precision floating-point computing power, calculate a second ratio of the initial video memory bandwidth to a preset video memory bandwidth, and calculate a third ratio of the initial video memory capacity to a preset video memory capacity, wherein the preset half-precision floating-point computing power is the half-precision floating-point computing power of a preset processor, the preset video memory bandwidth is the video memory bandwidth of the preset processor, and the preset video memory capacity is the video memory capacity of the preset processor;

[0035] A setting module is used to set the product of the weighted sum of the first ratio, the second ratio and the third ratio, a preset floating-point operation number unit and a preset time unit as the initial 1-degree computing power value of the first intelligent computing center.

[0036] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps in the method for measuring the computing power of an intelligent computing center by using 1 degree of computing power as described in the first aspect above.

[0037] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium for storing a program, which, when executed by a processor, implements the steps in the method for measuring the computing power of an intelligent computing center by using 1 degree of computing power as described in the first aspect above.

[0038] In a fifth aspect, an embodiment of the present invention further provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps in the method for measuring the computing power of an intelligent computing center by using 1 degree of computing power as described in the first aspect above.

[0039] In an embodiment of the present invention, an initial half-precision floating-point computing power, an initial video memory bandwidth, and an initial video memory capacity of a processor of the first intelligent computing center are obtained; a first ratio of the initial half-precision floating-point computing power to a preset half-precision floating-point computing power is calculated, a second ratio of the initial video memory bandwidth to a preset video memory bandwidth is calculated, and a third ratio of the initial video memory capacity to a preset video memory capacity is calculated, wherein the preset half-precision floating-point computing power is the half-precision floating-point computing power of a preset processor, the preset video memory bandwidth is the video memory bandwidth of the preset processor, and the preset video memory capacity is the video memory capacity of the preset processor; a weighted sum of the first ratio, the second ratio, and the third ratio, a preset floating-point operation number unit, and a preset time unit are multiplied as the initial 1-degree computing power value of the first intelligent computing center. In this way, the initial 1-degree computing power value of the first intelligent computing center is calculated through the initial half-precision floating-point computing power, initial memory bandwidth and initial memory capacity of the first intelligent computing center. The initial 1-degree computing power value can be used to measure the computing power of the intelligent computing center, thereby improving the accuracy of processor resource allocation for model training and improving the accuracy of measuring the computing power of the intelligent computing center. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0041] Figure 1 It is a flow chart of a method for measuring the computing power of an intelligent computing center by 1 degree of computing power provided by an embodiment of the present invention;

[0042] Figure 2 1 is a schematic diagram of floating-point operation analysis of the BERT model provided in an embodiment of the present invention;

[0043] Figure 3 It is a structural diagram of a device for measuring the computing power of an intelligent computing center by 1 degree of computing power provided by an embodiment of the present invention;

[0044] Figure 4 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The "computing power" mentioned in the present invention is the ability of computer equipment or computing / data center to process information. It is the ability of computer hardware and software to work together to execute certain computing requirements. It is the computing power to achieve target result output by processing information data. It is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to the society through computing power infrastructure.

[0047] The "computing power (CP)" mentioned in the present invention is the ability of a data center server to process data and output results. It is a comprehensive indicator to measure the computing power of a data center, including general computing power, super computing power and intelligent computing power. The commonly used unit of measurement is the number of floating point operations per second (Floating Point Operations Per Second, FLOPS, where 1EFLOPS=10^18 FLOPS). The larger the value, the stronger the comprehensive computing power. According to calculations, 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs or 2 million mainstream notebooks. The calculation formula is: CP=CP 通用 +CP 智能 +CP 超级 In the embodiment of the present invention, the computing power adopts the half-precision floating point computing power (FP16) of the graphics card.

[0048] The "carrying capacity" (Network Power, NP) described in the present invention is a manifestation of the data transmission capability of computing facilities, including comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling. Carrying capacity involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling capabilities. In the embodiment of the present invention, the carrying capacity uses the video memory bandwidth. In the embodiment of the present invention, the storage capacity uses the video memory bandwidth.

[0049] The "Storage Power" (SP) described in the present invention refers to the comprehensive capabilities of a data center in terms of data storage capacity, performance, safety and reliability, and green and low-carbon. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and server built-in storage devices. The commonly used unit of measurement for storage capacity is exabyte (EB, 1EB=2^60bytes), and the commonly used unit of measurement for performance is the number of reads and writes per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB). The disaster recovery ratio is an important manifestation of safety and reliability. In the embodiment of the present invention, the storage power adopts the video memory capacity.

[0050] The "computing power infrastructure" described in the present invention is a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity. It can realize centralized computing, storage, transmission and application of information, and presents characteristics such as diversity and ubiquity, intelligence and agility, safety and reliability, green and low-carbon. It is of great significance to promote industrial transformation and upgrading, enable my country's scientific and technological innovation, meet people's better life and realize high-efficiency social governance.

[0051] The "degree" mentioned in the present invention is a unit of measurement for computing power, and its properties and uses are equivalent to "meter", "second", "kilogram", "ampere", "candela", "mole", "Kelvin", etc. The "degree" mentioned in the present invention is illustrative rather than restrictive, and other units can also be used to measure computing power. Under the enlightenment of the present invention, ordinary technicians in this field can make many forms of expansion without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

[0052] The "1 degree computing power" described in the present invention refers to 1 unit of computing power, which refers to the ability of the intelligent computing center to meet the model training within a preset time range. The larger the value, the stronger the ability to meet the model training within the preset time range; or, it refers to the computing power provided by the intelligent computing center during model training. The larger the value, the stronger the computing power within the preset time range, or the larger the value, the less time required for the intelligent computing center to complete the same model training effect, that is, the higher the efficiency of model training. The nature and use of the "1 degree computing power" described in the present invention are equivalent to "1 degree of electricity". For example, the unit of "1 degree computing power" can be the number of floating-point operations within a preset time. For example, "1 degree computing power" can refer to "1TFLOPS·h", that is: one trillion floating-point operations per second × hour. The "1 degree computing power" described in the present invention is illustrative, not restrictive, and other units can also be used to measure the computing power of the intelligent computing center. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms of expansion without departing from the scope of protection of the purpose of the present invention and the claims, all of which belong to the protection of the present invention.

[0053] The “computing power” mentioned in the present invention includes general computing power CP 通用 、Intelligent computing power CP 智能 And super computing power CP 超级 .

[0054] The “general computing power” mentioned in the present invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0055] The "intelligent computing power" mentioned in the present invention refers to the large-scale deployment of intelligent computing centers based on special chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit) for various innovative artificial intelligence applications, such as natural language processing, machine vision, etc.

[0056] The "super computing power" mentioned in the present invention is mainly the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for calculations in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0057] The "intelligent computing center" mentioned in the present invention refers to a facility that provides the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning scenarios) by using large-scale heterogeneous computing power resources, including general computing power (CPU: Central Processing Unit) and intelligent computing power (GPU: Graphics Processing Unit, FPGA: Field Programmable Gate Array, ASIC: Application Specific Integrated Circuit, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0058] The "computing resources" mentioned in the present invention refer to technologies and facilities with information computing, transmission, storage and application capabilities required for the development of a digital society, including but not limited to computing resources such as CPU (Central Processing Unit), GPU (Graphics Processing Unit), network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and supporting and guarantee resources such as wind, fire, water and electricity.

[0059] See also Figure 1 , Figure 1 1 is a flowchart of a method for measuring the computing power of an intelligent computing center by 1 degree of computing power provided by an embodiment of the present invention, which is applied to a first intelligent computing center, such as Figure 1 As shown, the method comprises the following steps:

[0060] Step S1, obtaining an initial half-precision floating-point computing power, an initial video memory bandwidth, and an initial video memory capacity of a processor of the first intelligent computing center;

[0061] Step S2, calculating a first ratio of the initial half-precision floating-point computing power to a preset half-precision floating-point computing power, calculating a second ratio of the initial video memory bandwidth to a preset video memory bandwidth, and calculating a third ratio of the initial video memory capacity to a preset video memory capacity, wherein the preset half-precision floating-point computing power is the half-precision floating-point computing power of a preset processor, the preset video memory bandwidth is the video memory bandwidth of the preset processor, and the preset video memory capacity is the video memory capacity of the preset processor;

[0062] Step S3: set the product of the weighted sum of the first ratio, the second ratio and the third ratio, the preset floating-point operation number unit and the preset time unit as the initial 1-degree computing power value of the first intelligent computing center.

[0063] In an embodiment of the present invention, an initial half-precision floating-point computing power, an initial video memory bandwidth, and an initial video memory capacity of a processor of the first intelligent computing center are obtained; a first ratio of the initial half-precision floating-point computing power to a preset half-precision floating-point computing power is calculated, a second ratio of the initial video memory bandwidth to a preset video memory bandwidth is calculated, and a third ratio of the initial video memory capacity to a preset video memory capacity is calculated, wherein the preset half-precision floating-point computing power is the half-precision floating-point computing power of a preset processor, the preset video memory bandwidth is the video memory bandwidth of the preset processor, and the preset video memory capacity is the video memory capacity of the preset processor; a weighted sum of the first ratio, the second ratio, and the third ratio, a preset floating-point operation number unit, and a preset time unit are multiplied as the initial 1-degree computing power value of the first intelligent computing center. In this way, the initial 1-degree computing power value of the first intelligent computing center is calculated through the initial half-precision floating-point computing power, initial memory bandwidth and initial memory capacity of the first intelligent computing center. The initial 1-degree computing power value can be used to measure the computing power of the intelligent computing center, thereby improving the accuracy of processor resource allocation for model training and improving the accuracy of measuring the computing power of the intelligent computing center.

[0064] In the embodiment of the present invention, the preset half-precision floating point computing power number is the half-precision floating point computing power number of the preset processor, and the preset video memory bandwidth is the preset processor video memory bandwidth. In the process of calculating the initial 1 degree computing power value of different intelligent computing centers, the same processor needs to be used as the preset processor, so that the calculated 1 degree computing power value can accurately measure the data processing status of different processors.

[0065] In some implementations, the computing power of the processor is directly used as the unit for measuring the computing power of the intelligent computing center. However, the unit of measurement for computing power is the number of floating-point operations performed per second. In the actual process of the processor processing data, the speed at which the processor processes data is also limited by the bandwidth of the video memory. Figure 2The floating-point operation analysis of the bidirectional encoder representations from transformer (BERT) model shown in the figure shows that the speed at which the processor processes data is also limited by the memory bandwidth and memory capacity.

[0066] It should be noted that the computing power of computing equipment is mainly based on the market distribution of computing power equipment in various regions in the past six years, and measures the basic computing power, intelligent computing power, and supercomputing computing power from three major categories of products: general servers, artificial intelligence (AI) servers, and supercomputers. Among them, the basic computing power mainly focuses on the computing power scale of servers in various regions, and uses single-precision floating-point (FP32) computing power to measure computing power performance; intelligent computing power mainly focuses on the computing power scale of AI servers in various regions, and uses the mainstream half-precision floating-point computing power (FP16) computing power to measure computing power performance; supercomputing computing power is mainly based on the internationally renowned TOP500 rankings and refers to the relevant data of supercomputing manufacturers, and uses double-precision floating-point (FP64) computing power to measure the computing power performance of supercomputing. Therefore, in the embodiment of the present invention, for different intelligent computing centers, the corresponding initial 1-degree computing power value is calculated using half-precision floating-point computing power.

[0067] Based on the above reasons, the initial 1-degree computing power value of the first intelligent computing center in the present invention is calculated by the initial half-precision floating-point computing power, initial video memory bandwidth and initial video memory capacity of the processor of the first intelligent computing center, so that the initial 1-degree computing power value can more accurately measure the computing power of the intelligent computing center.

[0068] Using a typical BERT (transformer) model training scenario for measurement, according to the total time consumption test content of the model training process in Ivanov A, DrydenN, Ben-Nun T, et al. Data movement is all you need: A case study on optimizing transformers[J]. Proceedings of Machine Learning and Systems, 2021, 3: 711-732, the time consumption of the process mainly limited by the GPU computing power and the time consumption of the process mainly limited by the GPU video memory are counted respectively, and the time consumption of each process accounts for about 6:4.

[0069] For example, the processor NVIDIA A800 80GB SXM is used as the preset processor. The computing power generated by running for 1 hour is 1 1-degree computing power value, and its unit is TFLOPS·h (trillion floating-point operations per second × hour). With the half-precision floating-point computing power, video memory capacity, and video memory bandwidth ratio of 6:1:3 [that is, the initial 1-degree computing power value = (first ratio × 0.6 + second ratio × 0.1 + third ratio × 0.3) × preset floating-point operation number unit × preset time unit)], the initial 1-degree computing power value of processors of different models is calculated as shown in the following table:

[0070]

[0071] In the table, "TFLOPS" stands for trillion floating-point operations per second.

[0072] The initial 1-degree computing power value of the processors in different intelligent computing centers can be calculated in the above manner.

[0073] In the embodiment of the present invention, a metering conversion formula for heterogeneous processors (such as GPUs) and accelerator cards is provided, which takes into account the computing power of the processor, the memory bandwidth of the processor, and the weight of the typical workload, so that different intelligent computing centers can adopt a unified computing power settlement standard (i.e., 1 degree computing power value). This solves the trouble of complex negotiations and calculations between intelligent computing centers in the prior art, promotes the unified scheduling of computing power resources, and realizes the connection and integration of computing power resources from different sources and different geographical locations through network technology, so as to realize the sharing and optimal utilization of computing power.

[0074] Furthermore, through the intelligent computing center and artificial intelligence basic software, users can flexibly use various heterogeneous computing devices without having to worry about the conversion of specific computing units, which effectively enhances the flexibility of resource allocation and usage optimization, allowing various intelligent computing resources to be used in the most efficient way, thereby improving the resource utilization of the entire intelligent computing center.

[0075] Furthermore, the use of 1 kWh computing power as a standard unit of measurement will help the extensive collaborative construction and operation of intelligent computing centers, reducing the construction and operation and maintenance costs between intelligent computing centers. Each intelligent computing center can not only build and share computing power resources, but also achieve accurate matching of computing power demand and supply based on the 1 kWh computing power standard, giving full play to the advantages of intelligent computing resources in each region.

[0076] Furthermore, through the computing power service determined by 1 degree computing power value, users can quickly obtain the corresponding computing power resources without complicated evaluation of the detailed specifications of various hardware devices. This significantly improves the user experience, while also lowering the threshold for users to use computing power services and promoting the efficient development of artificial intelligence (AI) related research and development work.

[0077] Furthermore, by calculating the initial 1-degree computing power value, users can simply and intuitively evaluate and compare the actual effects of using different processors (such as GPUs) and accelerator cards in large model training and inference, eliminating the need for complex analysis of each specific performance indicator of various types of processor hardware and accelerator cards, making users more efficient and convenient when choosing.

[0078] In an embodiment of the present invention, the preset floating-point operation number unit can be set according to user needs, for example, it can be set to teraflops (TFLOPS), petaflops (PFLOPS) or exaflops (EFLOPS), etc.

[0079] In the embodiment of the present invention, the preset time unit can be set according to user needs, for example, it can be set to every second, every minute, or every hour.

[0080] In one embodiment, after step S3, the method further includes:

[0081] Step S4, obtaining model parameters of the initial model;

[0082] Step S5: Calculate the required 1-degree computing power value of the initial model based on the model parameters, where the required 1-degree computing power value is used to characterize the computing power required for training the initial model;

[0083] Step S6: Perform model training on the initial model based on the required 1-degree computing power value and the initial 1-degree computing power value.

[0084] In an embodiment of the present invention, the initial model is a model that needs to be trained. It can be a model uploaded by a user to the first intelligent computing center, or a model designed according to user needs, or a model sent to the first intelligent computing center by other intelligent computing centers, requesting the first intelligent computing center to train it.

[0085] In some embodiments, when the initial model is a model uploaded by the user to the first intelligent computing center, or a model designed according to user needs, the first intelligent computing center directly sends the trained model to the terminal corresponding to the user after training the initial model; or, the first intelligent computing center retains the trained model and sends the trained model to the terminal corresponding to the user after receiving a request instruction from the terminal corresponding to the user.

[0086] In some embodiments, when the initial model is a model sent to the first intelligent computing center by other intelligent computing centers, the first intelligent computing center directly sends the trained model to other intelligent computing centers after training the initial model.

[0087] In the embodiment of the present invention, the model parameters are parameters of the initial model, which are used to characterize relevant information of the initial model, such as the size of the initial model, the structure of the initial model, and the training data set required for training the initial model.

[0088] In an embodiment of the present invention, the required 1-degree computing power value of the initial model is calculated based on the model parameters. The required 1-degree computing power value can represent the computing power required for training the initial model. The required 1-degree computing power value obtained by calculating the model parameters is used to train the initial model based on the required 1-degree computing power value and the initial 1-degree computing power value, so that the first intelligent computing center does not allocate too many processor resources while completing the initial model training, thereby avoiding waste of processor resources.

[0089] In one embodiment, the model parameters include the size of the initial model and the size of the training data set;

[0090] The step S5 comprises:

[0091] Step S51: Calculate the required 1-degree computing power value of the initial model based on the size of the initial model and the size of the training data set.

[0092] In an embodiment of the present invention, the required 1-degree computing power value of the initial model is calculated based on the size of the initial model and the size of the training data set, so that the processor resources allocated based on the required 1-degree computing power value can meet the size of the initial model and the size of the training data set.

[0093] In some embodiments, the required 1-degree computing power value may be the product of the size of the initial model, the size of the training data set, and a preset coefficient. The preset coefficient may be determined according to the computing performance of the intelligent computing center, so that different intelligent computing centers may be allocated the same processor resources according to the required 1-degree computing power value.

[0094] In some embodiments, the model parameters may also include the number of training iterations, and the required 1-degree computing power value is calculated based on the size of the initial model, the size of the training data set, and the number of training iterations, so that the processor resources allocated based on the required 1-degree computing power value can meet the size of the initial model, the size of the training data set, and the number of training iterations.

[0095] For example, the required computing power value of 1 degree of the initial model can be calculated by the following formula:

[0096] FLOPS 模型 = k × number of model parameters × number of token training

[0097] Among them, FLOPS 模型is the required computing power value of 1 degree, k is the preset coefficient, the model parameter quantity represents the size of the initial model, and the token training number represents the number of iterations. In some examples, k=8.

[0098] It is understandable that model training includes forward propagation (forwardpass) and backward propagation (backwardpass), which are counted in units of calculation. One forward propagation is 1unit, and one backward propagation is 2unit (output gradient + parameter gradient). The activation recalculation technology is used to reduce the intermediate activation memory, and an additional forward pass is required (generally, activation recalculation technology is required, otherwise a single flagship card cannot train a 7B model). Therefore, a complete training: forward pass + backward pass + activation recalculation = 1+2+1=4unit; that is, for each token and each model parameter, 4units of calculation are required. The calculation of each unit is a matrix operation. For a matrix operation, one multiplication and one addition are required, a total of 2 floating-point operations. Therefore, a total of 8 times (number of model parameters × total number of tokens) calculation times are required, so the above calculation formula is obtained: FLOPS 模型 =8×number of model parameters×total number of tokens.

[0099] For example, the initial model is the llama2-7b model, the total number of parameters is 6.74B, activation recalculation is enabled, and the total number of tokens in the training data is 1T, then:

[0100] FLOPS 模型 = 8×6.74B×1T=8×(6.74×10^9) ×(1×10^12) = 53.92×10^21=5392×10^7 TFLOPS·h= 53,920,000 PFLOPS·h.

[0101] If an A100 card (with a computing power of 312×10^12 FLOPS) is used for training, the corresponding computing power consumption per 1 kWh is estimated to be: 8×(6.74×10^9) ×(1×10^12) / (312×10^12) ≈172.8×10^6 cards×seconds = 48,000 cards×hours. Among them, 1 card×hour = 312×10^12 FLOPS·h.

[0102] Further, according to Figure 2The total time consumption of the model training process in this article is calculated separately for the time consumption that is mainly limited by the GPU computing power and the time consumption that is mainly limited by the GPU video memory and bandwidth. The time consumption ratio of each is about 6:4. Therefore, the corresponding actual training 1 degree computing power value is 48,000 / 0.6 = 80,000 card × hours.

[0103] For example, using the NVIDIA A800 80GB SXM processor as the default processor, the computing power generated by running for 1 hour is 1 degree of computing power. Therefore, even if 1Ttoken is used to train the llama2-7b model, a total of 80,000 1 degree of computing power will be consumed.

[0104] It should be noted that after calculating the required 1-degree computing power value, the first intelligent computing center can determine whether to perform model training based on the required 1-degree computing power value. In the case where the remaining 1-degree computing power value of the first intelligent computing center is greater than or equal to the required 1-degree computing power value, the remaining computing power of the first intelligent computing center can meet the training requirements of the initial model, and the first intelligent computing center can perform model training on the initial model; and in the case where the remaining 1-degree computing power value of the first intelligent computing center is less than the required 1-degree computing power value, the remaining computing power of the first intelligent computing center cannot meet the training requirements of the initial model, and at this time, it can request scheduling computing power from other intelligent computing centers to realize the training of the initial model.

[0105] Specifically, in one embodiment, step S6 includes:

[0106] Step S61: Calculate the remaining 1-degree computing power value of the first intelligent computing center based on the initial 1-degree computing power;

[0107] Step S62: when the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model, allocating processor resources based on the required 1-degree computing power value;

[0108] Step S63: Perform model training on the initial model based on the processor resources.

[0109] In an embodiment of the present invention, by obtaining the remaining 1-degree computing power value of the first intelligent computing center, and when the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model, processor resources are allocated to perform model training on the initial model based on the required 1-degree computing power value, thereby realizing local training of the initial data through the first intelligent computing center.

[0110] At the same time, in one embodiment, the step S6 further includes:

[0111] Step S64: When the remaining 1-degree computing power value is less than the required 1-degree computing power value of the initial model, the initial model, the model parameters and the required 1-degree computing power value are sent to the second intelligent computing center;

[0112] Step S65, receiving the target model sent by the second intelligent computing center, wherein the target model is obtained by allocating processor resources by the second intelligent computing center based on the required 1-degree computing power value, and training the initial model based on the processor resources and the model parameters.

[0113] In an embodiment of the present invention, when the remaining 1 degree computing power value is less than the required 1 degree computing power value of the initial model, the initial model, model parameters and the required 1 degree computing power value are sent to the second intelligent computing center, and the target model sent by the second intelligent computing center is received. The target model is obtained by training the initial model by allocating processor resources and model parameters based on the required 1 degree computing power value by the second intelligent computing center. In this way, when the remaining 1 degree computing power value is less than the required 1 degree computing power value of the initial model, the first intelligent computing center does not directly train the initial model, but dispatches the computing power of the second intelligent computing center, and the second intelligent computing center trains the initial model, and the second intelligent computing center sends it to the first intelligent computing center, thereby realizing the training of the initial model when the computing power of the first intelligent computing center is insufficient.

[0114] In the embodiment of the present invention, the remaining 1 degree computing power value of the second intelligent computing center needs to be greater than or equal to the required 1 degree computing power value in order to train the initial model. If the remaining 1 degree computing power value of the second intelligent computing center needs to be less than the required 1 degree computing power value, the second intelligent computing center should send an indication of insufficient computing power to the first intelligent computing center to prompt the first intelligent computing center to request other intelligent computing centers to train the initial model.

[0115] For example, llama2-7b model training requires 80,000 1-degree computing power values. If the training needs to be completed within 1 month, 80,000 / (30×24)≈111×A100 cards are required, that is, the first intelligent computing center needs to provide 111 A100 cards to complete the initial model training. If the first intelligent computing center can only provide 60 A100 cards. Then in order to complete the task within the specified time. The computing power of 51 A100 cards can be rented from the second intelligent computing center, which is equivalent to 51×30×24=36,720 1-degree computing power values. The "30" in the formula refers to 30 days, and "24" refers to 24 hours.

[0116] At the same time, the 1-degree computing power value can also estimate the number of cards needed to achieve computing power scheduling between different GPUs.

[0117] For example, if the second intelligent computing center only provides H800 cards, then in order to complete the equivalent computing task within 1 month, it needs to rent 36,720 / (2.4949×30×24) ≈ 20×H800 cards, that is, the second intelligent computing center needs to allocate 20 H800 cards for model training. The "30" in the formula refers to 30 days, and the "24" refers to 24 hours.

[0118] In one embodiment, the step S61 includes:

[0119] Step S611, obtaining the training 1 degree computing power value of the first intelligent computing center for training other models;

[0120] Step S612: Set the difference between the initial 1-degree computing power value and the training 1-degree computing power value as the remaining 1-degree computing power value.

[0121] In the embodiment of the present invention, the training 1-degree computing power value of the first intelligent computing center for training other models is obtained; the difference between the initial 1-degree computing power value and the training 1-degree computing power value is set as the remaining 1-degree computing power value. In this way, the remaining 1-degree computing power value is calculated by the initial 1-degree computing power value and the training 1-degree computing power value, and the remaining 1-degree computing power value can be used to determine whether the first intelligent computing center can train the initial model.

[0122] In an embodiment of the present invention, the initial 1-degree computing power value is the maximum 1-degree computing power value corresponding to the processor of the first intelligent computing center, that is, the processor resources that can be called when the first intelligent computing center processes all training models do not exceed the processor resources corresponding to the initial 1-degree computing power value.

[0123] It should be noted that different intelligent computing centers use different processors. In order to facilitate different intelligent computing centers to allocate the same processor resources according to the 1-degree computing power value, it is necessary to first calculate the initial 1-degree computing power value corresponding to each processor, and then allocate the processor resources corresponding to the required 1-degree computing power value to different training models based on the initial 1-degree computing power value.

[0124] In one embodiment, the processor resources include the target half-precision floating-point computing power corresponding to the required 1-degree computing power value, the target video memory bandwidth, and the target video memory capacity; the step S62 includes:

[0125] Step S621: When the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model, calculate the target half-precision floating-point computing power number, the target video memory bandwidth, and the target video memory capacity corresponding to the required 1-degree computing power value;

[0126] Step S622: Allocate the target half-precision floating-point computing power, the target video memory bandwidth, and the target video memory capacity.

[0127] In an embodiment of the present invention, when the remaining 1 degree computing power value is greater than or equal to the required 1 degree computing power value of the initial model, the target half-precision floating point computing power, target video memory bandwidth and target video memory capacity corresponding to the required 1 degree computing power value are calculated; the target half-precision floating point computing power, target video memory bandwidth and target video memory capacity are allocated for model training of the initial model. In this way, the allocated target half-precision floating point computing power, target video memory bandwidth and target video memory capacity can meet the training requirements of the initial model, reducing the waste of processor resources while ensuring the training of the initial model.

[0128] In an embodiment of the present invention, a fourth ratio of the target half-precision floating-point computing power to the preset half-precision floating-point computing power of the preset processor is calculated, a fifth ratio of the target video memory bandwidth to the preset video memory bandwidth of the preset processor is calculated, and a sixth ratio of the target video memory capacity to the preset video memory capacity of the preset processor is calculated. The weighted sum of the fourth ratio, the fifth ratio, and the sixth ratio is the required 1-degree computing power value. Among them, the weighted value of the third ratio is the same as the weighted value of the first ratio, and the weighted value of the fourth ratio is the same as the weighted value of the second ratio. For example, the initial 1-degree computing power value = (first ratio × 0.6 + second ratio × 0.1 + third ratio × 0.3) × preset floating-point operation number unit × preset time unit, then the required 1-degree computing power value = (fourth ratio × 0.6 + fifth ratio × 0.1 + sixth ratio × 0.3) × preset floating-point operation number unit × preset time unit.

[0129] See also Figure 3 , Figure 3 is a structural diagram of a device for measuring the computing power of an intelligent computing center by 1 degree of computing power provided by an embodiment of the present invention, such as Figure 3 As shown, the device 300 for measuring the computing power of the intelligent computing center by 1 degree of computing power includes:

[0130] The first acquisition module 301 is used to acquire the initial half-precision floating-point computing power, initial video memory bandwidth and initial video memory capacity of the processor of the first intelligent computing center;

[0131] The first calculation module 302 is used to calculate a first ratio of the initial half-precision floating-point computing power to a preset half-precision floating-point computing power, calculate a second ratio of the initial video memory bandwidth to a preset video memory bandwidth, and calculate a third ratio of the initial video memory capacity to a preset video memory capacity, wherein the preset half-precision floating-point computing power is the half-precision floating-point computing power of a preset processor, the preset video memory bandwidth is the video memory bandwidth of the preset processor, and the preset video memory capacity is the video memory capacity of the preset processor;

[0132] The setting module 303 is used to set the product of the weighted sum of the first ratio, the second ratio and the third ratio, the preset floating point operation number unit and the preset time unit as the initial 1 degree computing power value of the first intelligent computing center.

[0133] In one embodiment, after the setting module 303, the device 300 for measuring the computing capacity of the intelligent computing center by 1 degree computing power further includes:

[0134] A second acquisition module is used to acquire model parameters of the initial model;

[0135] A second calculation module is used to calculate the required 1-degree computing power value of the initial model based on the model parameters, and the required 1-degree computing power value is used to represent the computing power required for training the initial model;

[0136] A training module is used to perform model training on the initial model based on the required 1-degree computing power value and the initial 1-degree computing power value.

[0137] In one embodiment, the model parameters include the size of the initial model and the size of the training data set;

[0138] The second calculation module includes:

[0139] The first calculation unit is used to calculate the required 1-degree computing power value of the initial model based on the size of the initial model and the size of the training data set.

[0140] In one embodiment, the training module includes:

[0141] A second calculation unit, configured to calculate a remaining 1 degree computing power value of the first intelligent computing center based on the initial 1 degree computing power;

[0142] An allocating unit, configured to allocate processor resources based on the required 1-degree computing power value when the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model;

[0143] A training unit is used to perform model training on the initial model based on the processor resources.

[0144] In one embodiment, the training module further includes:

[0145] A sending unit, configured to send the initial model, the model parameters and the required 1-degree computing power value to a second intelligent computing center when the remaining 1-degree computing power value is less than the required 1-degree computing power value of the initial model;

[0146] A receiving unit is used to receive a target model sent by the second intelligent computing center, where the target model is obtained by allocating the processor resources by the second intelligent computing center based on the required 1-degree computing power value, and training the initial model based on the processor resources and the model parameters.

[0147] In one embodiment, the second computing unit comprises:

[0148] An acquisition subunit, used to obtain a training 1 degree computing power value of the first intelligent computing center for training other models;

[0149] The first calculation subunit is used to set the difference between the initial 1-degree computing power value and the training 1-degree computing power value as the remaining 1-degree computing power value.

[0150] In one embodiment, the processor resources include the target half-precision floating-point computing power number corresponding to the required 1-degree computing power value, the target video memory bandwidth and the target video memory capacity; the training unit includes:

[0151] The second calculation subunit is used to calculate the target half-precision floating point computing power, the target video memory bandwidth and the target video memory capacity corresponding to the required 1 degree computing power value when the remaining 1 degree computing power value is greater than or equal to the required 1 degree computing power value of the initial model;

[0152] The training subunit is used to allocate the target half-precision floating-point computing power, the target video memory bandwidth and the target video memory capacity.

[0153] The device for measuring the computing power of an intelligent computing center by 1 degree of computing power provided in an embodiment of the present invention is capable of realizing the various processes of each embodiment of the above-mentioned method for measuring the computing power of an intelligent computing center by 1 degree of computing power. The technical features correspond one to one and can achieve the same technical effect. To avoid repetition, they will not be described here.

[0154] It should be noted that the device for measuring the computing power of the intelligent computing center through 1 degree computing power in the embodiment of the present invention can be a device, or a component, integrated circuit, or chip in an electronic device.

[0155] The embodiment of the present invention further provides an electronic device, see Figure 4 , Figure 4 4 is a schematic diagram of a structure of an electronic device provided by the present invention, the electronic device includes a memory 401, a processor 402 and a program or instruction stored in the memory 401 and running thereon, and when the program or instruction is executed by the processor 402, it can be realized Figure 1 Correspondingly, any steps in the method embodiment of measuring the computing power of the intelligent computing center by 1 degree of computing power and achieving the same beneficial effects will not be repeated here.

[0156] The processor 402 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or a complex programmable logic device (CPLD).

[0157] A person skilled in the art can understand that all or part of the steps of the above-mentioned method embodiment for measuring the computing power of an intelligent computing center through 1 degree computing power can be completed by hardware related to program instructions, and the program can be stored in a readable medium.

[0158] The embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above Figure 1 Correspondingly, any step in the method embodiment of measuring the computing power of the intelligent computing center by 1 degree computing power can achieve the same technical effect, and will not be repeated here to avoid repetition. The storage medium is such as a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0159] The embodiment of the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the following Figure 1 The steps in the method of measuring the computing power of the intelligent computing center by 1 degree of computing power can achieve the same technical effect. To avoid repetition, they will not be repeated here.

[0160] The terms "first", "second" etc. in the embodiments of the present invention are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. In addition, the terms "include" and "have" and any variation thereof are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment. In addition, "and / or" is used in the present invention to represent at least one of the connected objects, such as A and / or B and / or C, which means to include 7 situations including single A, single B, single C, and A and B all exist, B and C all exist, A and C all exist, and A, B and C all exist.

[0161] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0162] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a second terminal device, etc.) to execute the methods of various embodiments of the present invention.

[0163] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for measuring the computing power of an intelligent computing center by 1 degree of computing power, applied to a first intelligent computing center, characterized in that: include: Step S1, obtaining the initial half-precision floating-point computing power, initial video memory bandwidth and initial video memory capacity of the central processing unit CPU of the first intelligent computing center; Step S2, calculating a first ratio of the initial half-precision floating-point computing power to a preset half-precision floating-point computing power, calculating a second ratio of the initial video memory bandwidth to a preset video memory bandwidth, and calculating a third ratio of the initial video memory capacity to a preset video memory capacity, wherein the preset half-precision floating-point computing power is the half-precision floating-point computing power of a preset processor, the preset video memory bandwidth is the video memory bandwidth of the preset processor, the preset video memory capacity is the video memory capacity of the preset processor, and the preset processor and the CPU of the first intelligent computing center are different processors; Step S3, setting the product of the weighted sum of the first ratio, the second ratio and the third ratio, the preset floating point operation number unit and the preset time unit as the initial 1 degree computing power value of the first intelligent computing center; Step S4, obtaining model parameters of the initial model; Step S5: Calculate the required 1-degree computing power value of the initial model based on the model parameters, where the required 1-degree computing power value is used to characterize the computing power required for training the initial model; Step S6: performing model training on the initial model based on the required 1-degree computing power value and the initial 1-degree computing power value; The step S6 comprises: Step S61: Calculate the remaining 1-degree computing power value of the first intelligent computing center based on the initial 1-degree computing power; Step S62: when the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model, allocating processor resources based on the required 1-degree computing power value; Step S63: performing model training on the initial model based on the processor resources; The step S6 also includes: Step S64: When the remaining 1-degree computing power value is less than the required 1-degree computing power value of the initial model, the initial model, the model parameters and the required 1-degree computing power value are sent to the second intelligent computing center; Step S65, receiving the target model sent by the second intelligent computing center, wherein the target model is obtained by allocating the processor resources by the second intelligent computing center based on the required 1-degree computing power value, and training the initial model based on the processor resources and the model parameters.

2. The method according to claim 1, characterized in that The model parameters include the size of the initial model and the size of the training data set; The step S5 comprises: Step S51: Calculate the computing power required for the initial model based on the size of the initial model and the size of the training data set.

3. The method according to claim 1, characterized in that The step S61 comprises: Step S611, obtaining the training 1 degree computing power value of the first intelligent computing center for training other models; Step S612: Set the difference between the initial 1-degree computing power value and the training 1-degree computing power value as the remaining 1-degree computing power value.

4. The method according to claim 1, characterized in that The processor resources include the target half-precision floating point computing power number corresponding to the required 1 degree computing power value, the target video memory bandwidth and the target video memory capacity; The step S62 comprises: Step S621: When the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model, calculate the target half-precision floating-point computing power number, the target video memory bandwidth, and the target video memory capacity corresponding to the required 1-degree computing power value; Step S622: Allocate the target half-precision floating-point computing power, the target video memory bandwidth, and the target video memory capacity.

5. A device for measuring the computing power of an intelligent computing center by 1 degree of computing power, characterized in that: include: A first acquisition module is used to acquire an initial half-precision floating-point computing power, an initial video memory bandwidth, and an initial video memory capacity of a central processing unit (CPU) of a first intelligent computing center; a first calculation module, used to calculate a first ratio of the initial half-precision floating-point computing power to a preset half-precision floating-point computing power, calculate a second ratio of the initial video memory bandwidth to a preset video memory bandwidth, and calculate a third ratio of the initial video memory capacity to a preset video memory capacity, wherein the preset half-precision floating-point computing power is a half-precision floating-point computing power of a preset processor, the preset video memory bandwidth is a video memory bandwidth of the preset processor, the preset video memory capacity is a video memory capacity of the preset processor, and the preset processor and the CPU of the first intelligent computing center are different processors; A setting module, configured to set a weighted sum of the first ratio, the second ratio and the third ratio, a preset floating-point operation number unit and a preset time unit as a product of an initial 1-degree computing power value of the first intelligent computing center; A second acquisition module is used to acquire model parameters of the initial model; A second calculation module is used to calculate the required 1-degree computing power value of the initial model based on the model parameters, and the required 1-degree computing power value is used to represent the computing power required for training the initial model; A training module, used for performing model training on the initial model based on the required 1-degree computing power value and the initial 1-degree computing power value; The training module includes: A second calculation unit, configured to calculate a remaining 1 degree computing power value of the first intelligent computing center based on the initial 1 degree computing power; An allocating unit, configured to allocate processor resources based on the required 1-degree computing power value when the remaining 1-degree computing power value is greater than or equal to the required 1-degree computing power value of the initial model; A training unit, configured to perform model training on the initial model based on the processor resources; A sending unit, configured to send the initial model, the model parameters and the required 1-degree computing power value to a second intelligent computing center when the remaining 1-degree computing power value is less than the required 1-degree computing power value of the initial model; A receiving unit is used to receive a target model sent by the second intelligent computing center, where the target model is obtained by allocating the processor resources by the second intelligent computing center based on the required 1-degree computing power value, and training the initial model based on the processor resources and the model parameters.

6. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the method for measuring the computing power of an intelligent computing center by 1 degree of computing power as described in any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for measuring the computing power of an intelligent computing center by using 1 degree of computing power as described in any one of claims 1 to 4.

8. A computer program product, characterized in that It comprises computer instructions, which, when executed by a processor, implement the steps of the method for measuring the computing power of an intelligent computing center by using 1 degree of computing power as described in any one of claims 1 to 4.

Images