Model configuration parameter generation method and device, equipment and computer storage medium

Abstract

The embodiment of the invention provides a model configuration parameter generation method and device, equipment and a computer storage medium. The method comprises the steps of obtaining preset target model task data; and based on the target model task data, according to a corresponding relationship between preset historical model task data and a historical model predicted calculation power value, determining a target model predicted calculation power value corresponding to the preset target model task data. And under the condition that the target model predicted computing power value is equal to the historical model predicted computing power value, historical model configuration parameters corresponding to the historical model predicted computing power value are used as target model configuration parameters, and the target model configuration parameters are used for constructing a target model. According to the invention, under the condition that the predicted calculation power value of the target model is equal to the predicted calculation power value of the historical model, the historical model configuration parameter corresponding to the predicted calculation power value of the historical model is used as the target model configuration parameter, so that the time for determining the target model configuration parameter can be shortened; therefore, the time for constructing the target model based on the target model configuration parameters is shortened.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, and in particular relates to a method, apparatus, device and computer storage medium for generating model configuration parameters. Background Technology

[0002] With the development of information technology, artificial intelligence models have received widespread attention. Currently, in the process of building a model, it is usually necessary to estimate the computing power required to build the model, and then configure the corresponding parameters for the model based on the predicted computing power.

[0003] However, the methods used by related technologies to estimate the computing power required to build the model have limitations, leading to inaccurate predictions and consequently, low accuracy in model configuration parameters. Furthermore, due to the low accuracy of these parameters, reconfiguration is necessary, resulting in a prolonged time required to build the target model based on those parameters. Summary of the Invention

[0004] This invention provides a method, apparatus, device, computer storage medium, and computer program product for generating model configuration parameters, which can improve the accuracy of determining model configuration parameters and construct a target model based on the model configuration parameters, thereby reducing the construction time of the target model.

[0005] In a first aspect, embodiments of the present invention provide a method for generating model configuration parameters, the method comprising: Acquire the preset target model task data, which represents the requirement data corresponding to the target model's execution task; Based on the correspondence between the preset historical model task data and the historical model predicted computing power value, the target model predicted computing power value corresponding to the preset target model task data is determined. The preset historical model task data represents the demand data corresponding to the historical model executing the historical task. If the target model's predicted computing power value is equal to the historical model's predicted computing power value, the historical model configuration parameters corresponding to the historical model's predicted computing power value are determined as the target model configuration parameters, and the target model configuration parameters are used to construct the target model.

[0006] In one feasible implementation, before determining the target model prediction computing power value corresponding to the preset target model task data based on the correspondence between preset historical model task data and historical model prediction computing power values, the method further includes: Determine the hardware resource configuration information and scoring information corresponding to the historical model; Obtain hardware resource utilization information from historical models; The historical model's predicted computing power value is obtained by multiplying the utilization rate information of each hardware resource in the historical model with the corresponding hardware resource's score information. Obtain preset historical model task data; The correlation coefficient is obtained by performing correlation calculations on the pre-set historical model task data and the historical model prediction computing power values. Based on the correlation coefficient, the correspondence between the preset historical model task data and the historical model prediction computing power value is determined.

[0007] In one feasible implementation, the hardware resources corresponding to the historical model include at least one of a central processing unit (CPU), a graphics processing unit (GPU), memory, and hard disk; the CPU configuration information includes at least one of the following: number of servers, number of cores, single-core clock speed, and single-cycle floating-point calculation value; the GPU configuration information includes at least one of the following: number of servers, number of cores, single-core clock speed, and single-cycle floating-point calculation value; the memory configuration information includes at least one of the following: capacity and speed; the hard disk configuration information includes at least one of the following: capacity and speed. The scoring information for the hardware resources corresponding to the configuration information of the historical model includes: When the hardware resources corresponding to the historical model include CPUs, the CPU score information is obtained by multiplying at least one piece of information about the number of servers, the number of cores, the single-core clock frequency, and the single-cycle floating-point calculation value for each type of CPU. When the hardware resources corresponding to the historical model include GPUs, the GPU score information is obtained by multiplying at least one piece of information about the number of servers, the number of cores, the single-core clock frequency, and the single-cycle floating-point calculation value for each type of GPU. If the hardware resources corresponding to the historical model include memory, memory score information is obtained based on the sum of at least one capacity information and speed information corresponding to memory. If the hardware resources corresponding to the historical model include hard drives, the hard drive rating information is obtained based on the sum of at least one capacity information and speed information corresponding to the hard drive. The hardware resource rating information corresponding to the historical model is obtained by summing at least one of the CPU rating information, GPU rating information, memory rating information, and hard disk rating information.

[0008] In one feasible implementation, the preset historical model task data includes at least one of the following: historical task type data, historical dataset data volume, preset historical model inference time, and preset historical model accuracy; the correlation coefficient is obtained by performing correlation calculation on the preset historical model task data and the historical model prediction computing power value, including: Given that the historical model task data includes historical task type data, the Pearson correlation between the historical task type data and the historical model prediction computing power value is calculated to obtain the first correlation coefficient. Given that the historical model task data includes the amount of historical dataset data, the Pearson correlation between the amount of historical dataset data and the historical model prediction computing power value is calculated to obtain the second correlation coefficient. Given that the preset historical model task data includes the preset historical model inference time, the preset historical model inference time and the historical model prediction computing power value are calculated using Pearson correlation to obtain the third correlation coefficient. Given that the preset historical model task data includes the preset historical model accuracy, the preset historical model accuracy and the historical model prediction computing power value are calculated using Pearson correlation to obtain the fourth correlation coefficient. The correlation coefficient is obtained by summing at least one of the first, second, third, and fourth correlation coefficients.

[0009] In one feasible implementation, after determining the target model configuration parameters based on the predicted computing power value of the target model, the method further includes: Construct the target model based on the target model configuration parameters; Obtain the actual computing power and operational information of the target model; Determine the computing power prediction accuracy corresponding to the actual computing power value of the target model and the predicted computing power value of the target model; Determine the accuracy of computing power prediction and the corresponding operational quality information based on operational information; Based on the operational quality information, perform optimization operations on the target model that correspond to the operational quality information.

[0010] In one feasible implementation, obtaining the actual computing power value of the target model includes: Determine the configuration information and scoring information of the hardware resources corresponding to the target model; Obtain the hardware resource utilization information of the target model; The actual computing power of the target model is obtained by multiplying the utilization rate information of each hardware resource in the target model with the corresponding hardware resource score information.

[0011] In one feasible implementation, the runtime information of the target model is obtained, including: Obtain the communication parameters corresponding to the target model; Determine the network operational stability information corresponding to the communication parameters; Obtain information on the number of failures of the target model within a preset period; Determine the number of failures and the failure rate information corresponding to the preset cycle; Obtain information on the number of failures, failure repair time, and model runtime of the target model within a preset period; Determine the effective operational information of the target model corresponding to the fault count, fault repair time, and model runtime information within a preset period; Based on network stability information, failure rate information, and effective model operation information, the operation information of the target model is determined.

[0012] Secondly, embodiments of the present invention provide an apparatus for generating model configuration parameters, the apparatus comprising: The acquisition module is used to acquire the preset target model task data, which represents the requirement data corresponding to the target model's execution task. The determination module is used to determine the target model prediction computing power value corresponding to the preset target model task data based on the correspondence between the preset historical model task data and the historical model prediction computing power value. The preset historical model task data represents the demand data corresponding to the historical model executing the historical task. The determination module is also used to determine, when the target model's predicted computing power value is equal to the historical model's predicted computing power value, to use the historical model's configuration parameters as the target model's configuration parameters. The target model's configuration parameters are used to construct the target model.

[0013] Thirdly, embodiments of the present invention provide a device for generating model configuration parameters. The device includes a processor and a memory storing computer program instructions. The processor reads and executes the computer program instructions to implement the method for generating model configuration parameters as described in the first aspect.

[0014] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the method for generating model configuration parameters as described in the first aspect.

[0015] Fifthly, embodiments of the present invention provide a computer program product, including a computer program that, when executed by a processor, implements the method for generating model configuration parameters as described in the first aspect.

[0016] This invention provides a method, apparatus, device, and computer storage medium for generating model configuration parameters. The method acquires preset target model task data. Based on the target model task data, and according to the correspondence between preset historical model task data and historical model prediction computing power values, the target model prediction computing power value corresponding to the preset target model task data is determined. Since the correspondence between preset historical model task data and historical model prediction computing power values ​​is determined based on actual values ​​and has reference value, determining the target model prediction computing power value corresponding to the preset target model task data based on this correspondence can improve the accuracy of determining the target model prediction computing power value. When the target model prediction computing power value is equal to the historical model prediction computing power value, the historical model configuration parameters corresponding to the historical model prediction computing power value are used as the target model configuration parameters, which are then used to construct the target model. This invention, by using the historical model configuration parameters corresponding to the historical model prediction computing power value as the target model configuration parameters when the target model prediction computing power value is equal to the historical model prediction computing power value, can improve the accuracy of determining the target model configuration parameters and reduce the time required to determine them, thus reducing the time required to construct the target model based on the target model configuration parameters. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for generating model configuration parameters according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for determining the correspondence between preset historical model task data and historical model prediction computing power values, provided by an embodiment of the present invention. Figure 3 This is a flowchart illustrating a method for determining hardware resource scoring information according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating a method for determining correlation coefficients provided in an embodiment of the present invention; Figure 5 This is a flowchart illustrating a method for determining optimization operations of a target model according to an embodiment of the present invention; Figure 6 This is a flowchart illustrating a method for determining the actual computing power value of a target model according to an embodiment of the present invention; Figure 7 This is a flowchart illustrating a method for determining model running information provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of a system for generating model configuration parameters provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of a device for generating model configuration parameters provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of a device for generating model configuration parameters provided in an embodiment of the present invention. Detailed Implementation

[0019] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0021] Before describing the technical solutions provided by the embodiments of the present invention, in order to facilitate understanding of the embodiments of the present invention, the present invention will specifically explain the problems existing in the related technologies: Currently, artificial intelligence models are playing a significant role in various fields. Related technologies typically involve collecting data from different data sources and evaluating the model's computational power based on this data. Then, model parameters are configured according to the evaluated computational power to design the model, thereby obtaining the target model.

[0022] However, in related technologies, the computing power of a model is usually evaluated only based on the size of the dataset used by the model. This evaluation method is too limited, so the accuracy of the computing power evaluated by related technologies is not high. As a result, the accuracy of the model configuration parameters determined based on the computing power is not high, and the model parameters need to be reconfigured. Therefore, the time to build the target model is long.

[0023] The present invention provides a method, apparatus, device, computer storage medium, and computer program product for generating model configuration parameters, which can solve the above-mentioned technical problems existing in related technologies.

[0024] This invention obtains preset target model task data. Based on the correspondence between preset historical model task data and historical model prediction computing power values, it determines the target model prediction computing power value corresponding to the preset target model task data. When the target model prediction computing power value is equal to the historical model prediction computing power value, the historical model configuration parameters corresponding to the historical model prediction computing power value are determined as the target model configuration parameters. This invention determines the target model prediction computing power value corresponding to the preset target model task data based on a predetermined correspondence. Compared to related technologies that evaluate model computing power based on the size of the dataset used by the model, the model prediction computing power value determined based on historical experience data has higher accuracy. Therefore, this invention can improve the accuracy of determining the target model prediction computing power value, thereby improving the accuracy of determining the target model configuration parameters based on the target model computing power value and reducing the time required to build the target model.

[0025] The method for generating model configuration parameters provided in this embodiment of the invention will be introduced below.

[0026] Figure 1 A flowchart illustrating a method for generating model configuration parameters according to an embodiment of the present invention is shown. Figure 1 As shown, the method includes steps S110-S130.

[0027] S110: Obtain the preset target model task data.

[0028] The pre-defined target model task data represents the required data for the target model to perform the task. This can be at least one of the following: task type, dataset size, expected model inference time, and expected model application accuracy. The task type represents the type of problem the target model aims to solve. The dataset size represents the amount of input data to the target model. The expected model inference time represents the pre-set target time for the model to process the input data and generate the output. The expected model application accuracy represents the prediction accuracy at which the model achieves the expected target.

[0029] In one embodiment, preset target model task data, including task type, dataset size, expected model inference time, and expected model application accuracy, can be obtained.

[0030] The steps of this invention to obtain preset target model task data are intended to enable the determination of the target model prediction computing power value based on the feature dimensions corresponding to the preset target model task data. Compared with related technologies that determine the computing power value by the size of the dataset used by the model, this method can improve the accuracy of determining the target model prediction computing power value.

[0031] S120: Based on the correspondence between the preset historical model task data and the historical model prediction computing power value, determine the target model prediction computing power value corresponding to the preset target model task data.

[0032] The correspondence between the preset historical model task data and the historical model prediction computing power value can be a mathematical empirical formula. The preset historical model task data represents the required data for the historical model to execute historical tasks, and may include at least one of the following: task type, dataset size, expected model inference time, and expected model application accuracy.

[0033] In one embodiment, the target model prediction computing power value corresponding to the preset target model task data can be determined based on the mathematical empirical relationship between the preset historical model task data and the historical model prediction computing power value.

[0034] In one example, the expression for the correspondence between the preset historical model task data and the historical model prediction computing power value is shown in formula (1).

[0035] in, The characterization model predicts computing power values. Characterize the task type, The amount of data representing the dataset Characterizes the expected inference time of the model. Characterize the expected accuracy of the model application. It represents a mapping rule.

[0036] In the steps of this embodiment of the invention, the target model prediction computing power value corresponding to the preset target model task data is determined by the correspondence between the preset historical model task data and the historical model prediction computing power value. Since the above correspondence is based on the historical real value, it can be used to refer to the prediction computing power value, thus improving the accuracy of determining the target model prediction computing power value.

[0037] S130: If the target model's predicted computing power value is equal to the historical model's predicted computing power value, determine that the historical model's configuration parameters corresponding to the historical model's predicted computing power value shall be used as the target model's configuration parameters.

[0038] The historical model configuration parameters represent the parameter information for setting the hardware resources of the historical model, while the target model configuration parameters represent the parameter information for setting the hardware resources of the target model. The target model configuration parameters are used to construct the target model.

[0039] In one embodiment, after obtaining the target model's predicted computing power value, it can be checked whether the target model's predicted computing power value is equal to the historical model's predicted computing power value. If the target model's predicted computing power value is equal to the historical model's predicted computing power value, the historical model configuration parameters corresponding to the historical model's predicted computing power value are used as the target model's configuration parameters.

[0040] Based on the correspondence between preset historical model task data and historical model prediction computing power values, this invention can accurately determine the target model prediction computing power value corresponding to preset target model task data. Furthermore, when the target model prediction computing power value is equal to the historical model prediction computing power value, the historical model configuration parameters corresponding to the historical model prediction computing power value are used as the target model configuration parameters. This is equivalent to reusing the historical model configuration parameters, reducing the time required to determine the target model configuration parameters and the time required to construct the target model based on the target model configuration parameters.

[0041] In one embodiment, before determining the target model prediction computing power value corresponding to the preset target model task data based on the correspondence between preset historical model task data and historical model prediction computing power values, such as Figure 2 As shown, the method for generating model configuration parameters also includes steps S140-S190.

[0042] S140: Determine the configuration information and scoring information of the hardware resources corresponding to the historical model.

[0043] Among them, the hardware resource rating information represents the performance index information of the hardware resources.

[0044] In one example, the hardware resource rating information may include at least one of the following: general computing power rating, high-performance computing power rating, memory performance rating, and disk performance rating.

[0045] In one embodiment, the score information of at least one of the hardware resources, including general computing power score, high-performance computing power score, memory performance score, and disk performance score, can be determined based on the configuration information of the hardware resources corresponding to the historical model.

[0046] This invention determines the corresponding hardware resource rating information by using the configuration information of the hardware resources corresponding to the historical model. This is to subsequently determine the historical model's predicted computing power value based on the hardware resource rating information. Compared with related technologies that use the size of the dataset used by the model to determine the computing power value, this invention determines the computing power value based on the hardware resource rating information, i.e., the hardware resource performance rating. It transforms quantifiable indicators into abstract computing power values, which can accurately reflect the computing power value of the historical model, thereby improving the accuracy of determining the historical model's predicted computing power value.

[0047] S150: Obtain hardware resource utilization information from historical models.

[0048] Among them, the hardware resource utilization information of the historical model can be the hardware resource utilization information of all types of models that have been saved in advance.

[0049] In one example, hardware resource utilization information may include at least one of the following: CPU utilization, GPU utilization, memory utilization, and hard disk utilization.

[0050] In one embodiment, at least one of the following can be obtained when the hardware resources of the historical model are executing a task: CPU utilization information, GPU utilization information, memory utilization information, and hard disk utilization information.

[0051] S160: Add the product of the utilization rate information of each hardware resource in the historical model and the corresponding hardware resource score information to obtain the predicted computing power value of the historical model.

[0052] In one embodiment, after obtaining the utilization information of each hardware resource in the historical model, the value of the predicted computing power of the historical model can be obtained by multiplying it by the corresponding hardware resource's rating information and then adding them together.

[0053] In one example, the expression for determining the historical model prediction computing power value is shown in formula (2).

[0054] in, Characterize the computing power value predicted by the historical model. A score characterizing the hardware resources of the central processing unit. It represents the utilization rate of the central processing unit. A score characterizing the hardware resources of the image processor. Characterizes the utilization rate of the image processor. A score representing hardware resources that characterize memory. It represents the utilization rate of memory. A score that characterizes the hardware resources of the hard drive. It represents the utilization rate of the hard drive.

[0055] In the steps of this embodiment of the invention, the product of the utilization rate information of each hardware resource in the historical model and the score information of the corresponding hardware resource is added together because the utilization rate information of the hardware resource reflects the dynamic performance information of the historical model, while the score information of the hardware resource reflects the static performance information of the historical model. Therefore, by comprehensively evaluating the predicted computing power value of the historical model from both static and dynamic performance perspectives, the accuracy of determining the predicted computing power value of the historical model can be improved.

[0056] S170: Obtain preset historical model task data.

[0057] In one embodiment, preset historical model task data can be obtained, including at least one of the following: task type, data volume of the dataset, expected model inference time, and expected model application accuracy.

[0058] S180: Perform correlation calculations on the preset historical model task data and the historical model prediction computing power values ​​to obtain the correlation coefficient.

[0059] Correlation calculation is used to determine the degree of linear correlation between two variables. An example of a correlation calculation is the Pearson correlation calculation.

[0060] In one embodiment, Pearson correlation calculation can be performed on preset historical model task data and historical model prediction computing power values ​​to obtain the correlation coefficient.

[0061] S190: Based on the correlation coefficient, determine the correspondence between the preset historical model task data and the historical model prediction computing power value.

[0062] In one embodiment, after obtaining the correlation coefficient, the preset historical model task data can be weighted based on the correlation coefficient to obtain the correspondence between the preset historical model task data and the historical model prediction computing power value.

[0063] In one example, the expression for the correspondence between the historical model task data and the historical model prediction computing power value is shown in formula (3).

[0064] in, Characterizes the computing power value predicted by the historical model. Characterize the task type, The amount of data representing the dataset Characterizes the expected inference time of the model. K1 represents the correlation coefficient corresponding to the task type, K2 represents the correlation coefficient corresponding to the data volume of the dataset, K3 represents the correlation coefficient corresponding to the inference time of the expected model, and K4 represents the correlation coefficient corresponding to the accuracy of the expected model application.

[0065] This invention provides an embodiment of the invention that obtains historical model prediction computing power values ​​and performs correlation calculations with preset historical model task data. This allows for the determination of the degree of correlation between the historical model prediction computing power values ​​and the preset historical model task data, thereby obtaining an accurate correspondence between the preset historical model task data and the historical model prediction computing power values ​​based on the degree of correlation.

[0066] In one embodiment, the hardware resources corresponding to the historical model include at least one of a central processing unit (CPU), a graphics processing unit (GPU), memory, and a hard disk; the CPU configuration information includes at least one of the following: number of servers, number of cores, single-core clock speed, and single-cycle floating-point calculation value; the GPU configuration information includes at least one of the following: number of servers, number of cores, single-core clock speed, and single-cycle floating-point calculation value; the memory configuration information includes at least one of the following: capacity and speed; and the hard disk configuration information includes at least one of the following: capacity and speed. The hardware resource scoring information corresponding to the configuration information of the hardware resources corresponding to the historical model is determined, such as... Figure 3 As shown, steps S141-S145 may be included.

[0067] S141: When the hardware resources corresponding to the historical model include CPUs, CPU score information is obtained by multiplying at least one piece of information on the number of servers, the number of cores, the single-core clock frequency, and the single-cycle floating-point calculation value for each type of CPU.

[0068] In one embodiment, the initial CPU score can be obtained by summing the products of the number of servers, cores, single-core frequency, and single-cycle floating-point calculation value for each CPU in the historical model. This initial CPU score is then multiplied by a general computing power score conversion factor to obtain the final CPU score. This general computing power score conversion factor, also known as the CPU performance score conversion factor, is used to convert parameters into a score on a unified scale. The general computing power score conversion factor is obtained through standardized calculations based on industry benchmarks.

[0069] In one example, the expression for obtaining the initial CPU score information is shown in formula (4).

[0070] in, Characterizing CPU score, Characterizing the conversion coefficient of general computing power score, Represents the number of servers with the i-th type of CPU. The number of cores in the i-th CPU model. Characterizing the single-core clock frequency, Characterizes the floating-point calculation value for a single cycle.

[0071] S142: When the hardware resources corresponding to the historical model include GPUs, the GPU score information is obtained by multiplying at least one piece of information about the number of servers, the number of cores, the single-core clock frequency, and the single-cycle floating-point calculation value for each type of GPU.

[0072] In one embodiment, the initial GPU score can be obtained by summing the products of the number of servers, cores, single-core clock speed, and single-cycle floating-point calculation value for each GPU in the historical model. This initial GPU score is then multiplied by a high-performance computing power score conversion factor to obtain the final GPU score. This high-performance computing power score conversion factor, also known as the GPU performance score conversion factor, is used to convert parameters into a score on a unified scale. The high-performance computing power score conversion factor is obtained through standardized calculations based on industry benchmarks.

[0073] In one example, the expression for obtaining the initial GPU score information is shown in formula (5).

[0074] in, Characterizing GPU score, Characterizing the conversion coefficient of general computing power score, Represents the number of GPU servers of type i. The number of cores representing the i-th type of GPU. Characterizing the single-core clock frequency, Characterizes the floating-point calculation value for a single cycle.

[0075] S143: If the hardware resources corresponding to the historical model include memory, obtain memory score information based on the sum of at least one capacity information and speed information corresponding to memory.

[0076] In one embodiment, the capacity and speed information corresponding to the memory in the historical model can be added together to obtain the initial memory score. Then, the initial memory score is multiplied by a memory performance score conversion factor to obtain the final memory score. The memory performance score conversion factor is calculated using industry benchmark standardization and is used to convert parameters into a score on a unified scale.

[0077] In one example, the expression for obtaining memory score information is shown in formula (6).

[0078] in, Characterize memory score, The conversion factor for memory performance scores. Information representing memory capacity Information that represents memory speed.

[0079] S144: If the hardware resources corresponding to the historical model include hard drives, obtain hard drive rating information based on the sum of at least one capacity information and speed information corresponding to the hard drive.

[0080] In one embodiment, the capacity and speed information of the hard drives in the historical model can be added together to obtain initial hard drive rating information. Then, the initial hard drive rating information is multiplied by a hard drive performance score conversion factor to obtain the final hard drive rating. The hard drive performance score conversion factor is calculated using industry benchmark standardization and is used to convert parameters into a rating under a unified scale.

[0081] In one example, the expression for obtaining hard drive rating information is shown in formula (7).

[0082] in, Characterizing hard drive rating The conversion factor for hard drive performance scores. Information representing hard drive capacity Information that represents hard drive speed.

[0083] S145: Obtain the hardware resource score information corresponding to the historical model based on the sum of at least one of the CPU score information, GPU score information, memory score information, and hard disk score information.

[0084] In this embodiment, after obtaining at least one of the CPU rating information, GPU rating information, memory rating information, and hard disk rating information, the sum of the above information can be calculated to obtain the hardware resource rating information corresponding to the historical model.

[0085] This invention provides a hardware resource rating information corresponding to a historical model by summing at least one of CPU rating information, GPU rating information, memory rating information, and hard disk rating information. In other words, it can determine the hardware resource rating information through at least one dimension, thereby improving the accuracy of determining the hardware resource rating information.

[0086] In one embodiment, the preset historical model task data includes at least one of historical task type data, historical dataset data volume, preset historical model inference time, and preset historical model accuracy; a correlation coefficient is obtained by calculating the correlation between the preset historical model task data and the historical model prediction computing power value, such as... Figure 4 As shown, steps S181-S185 may be included.

[0087] S181: If the historical model task data includes historical task type data, perform Pearson correlation calculation between the historical task type data and the historical model prediction computing power value to obtain the first correlation coefficient.

[0088] In this embodiment, if the preset historical model task data includes historical task type data, then the historical task type data and the historical model prediction computing power value can be calculated using Pearson correlation to obtain the first correlation coefficient.

[0089] In one example, the expression for the first correlation coefficient is shown in formula (8).

[0090] in, Characterizing the first correlation coefficient, Data representing historical task types, Characterizes the computing power value predicted by the historical model. The covariance between historical task type data and historical model predicted computing power values. The standard deviation, representing the size of the historical dataset, The standard deviation of the historical model's predicted computing power value.

[0091] S182: Given that the historical model task data includes the amount of historical dataset data, perform Pearson correlation calculation between the amount of historical dataset data and the historical model prediction computing power value to obtain the second correlation coefficient.

[0092] In this embodiment, if the preset historical model task data includes the amount of historical dataset data, then the Pearson correlation between the amount of historical dataset data and the historical model prediction computing power value can be calculated to obtain the second correlation coefficient.

[0093] In one example, the expression for the second correlation coefficient is shown in formula (9).

[0094] in, Characterizing the second correlation coefficient, Characterizing the amount of historical dataset, Characterizes the computing power value predicted by the historical model. The covariance between the amount of historical dataset data and the computational power value predicted by the historical model. The standard deviation, representing the size of the historical dataset, The standard deviation of the historical model's predicted computing power value.

[0095] S183: Given that the preset historical model task data includes the preset historical model inference time, perform Pearson correlation calculation on the preset historical model inference time and the historical model prediction computing power value to obtain the third correlation coefficient.

[0096] In this embodiment, if the preset historical model task data includes the preset historical model inference time, the preset historical model inference time and the historical model prediction computing power value can be calculated using Pearson correlation to obtain the third correlation coefficient.

[0097] In one example, the expression for the third correlation coefficient is shown in formula (10).

[0098] in, Characterizing the third correlation coefficient, The duration of reasoning in the predefined historical model is represented. Characterizes the computing power value predicted by the historical model. The covariance between the inference time of the preset historical model and the computing power value predicted by the historical model represents the variance of the inference time of the preset historical model. The standard deviation, representing the size of the historical dataset, The standard deviation of the historical model's predicted computing power value.

[0099] S184: Given that the preset historical model task data includes the preset historical model accuracy, perform Pearson correlation calculation between the preset historical model accuracy and the historical model prediction computing power value to obtain the fourth correlation coefficient.

[0100] In this embodiment, if the preset historical model task data includes the preset historical model accuracy, the preset historical model accuracy and the historical model prediction computing power value can be calculated using Pearson correlation to obtain the first correlation coefficient.

[0101] In one example, the expression for the first correlation coefficient is shown in formula (11).

[0102] in, Characterizing the first correlation coefficient, Characterizes the accuracy of the preset historical model. Characterizes the computing power value predicted by the historical model. The covariance between the accuracy of the preset historical model and the predicted computing power of the historical model is used to characterize the data. The standard deviation, representing the size of the historical dataset, The standard deviation of the historical model's predicted computing power value.

[0103] S185: The correlation coefficient is obtained by summing at least one of the first, second, third and fourth correlation coefficients.

[0104] In this embodiment, the values ​​of at least one of the first correlation coefficient, the second correlation coefficient, the third correlation coefficient, and the fourth correlation coefficient can be added together to obtain the correlation coefficient.

[0105] In one embodiment, the first correlation coefficient, the second correlation coefficient, the third correlation coefficient, and the fourth correlation coefficient can be added together to obtain the correlation coefficient.

[0106] The embodiments of the present invention obtain the correlation coefficient by summing at least one of the first correlation coefficient, the second correlation coefficient, the third correlation coefficient, and the fourth correlation coefficient, that is, the correlation coefficient can be determined by at least one dimension, which can improve the accuracy of determining the correlation coefficient.

[0107] In one embodiment, after determining the target model configuration parameters based on the predicted computing power value of the target model, such as... Figure 5 As shown, the method for generating model configuration parameters also includes steps S210-S250.

[0108] S210: Construct the target model based on the target model configuration parameters.

[0109] In this embodiment, after obtaining the target model configuration parameters, the target model is constructed based on the hardware resources corresponding to the target model configuration parameters.

[0110] S220: Obtain the actual computing power and operating information of the target model.

[0111] The actual computing power value of the target model represents its computing performance when executing tasks. The runtime information represents the operational status of the target model.

[0112] In this embodiment, the actual computing power and operation information of the target model can be obtained based on the model operation monitoring module.

[0113] In the steps of this invention, obtaining the actual computing power value and operating information of the target model is to determine the quality of the constructed target model from multiple aspects such as the model's computing power prediction accuracy and the model's operating status.

[0114] S230: Determine the accuracy of the computing power prediction corresponding to the actual computing power value of the target model and the predicted computing power value of the target model.

[0115] In this embodiment, after obtaining the actual computing power value of the target model, the computing power prediction accuracy can be determined based on the actual computing power value of the target model and the predicted computing power value of the target model.

[0116] In one example, the expression for determining the accuracy of computing power prediction is shown in formula (12).

[0117] in, Characterizing the accuracy of prediction based on computing power. Characterizes the actual computing power of the target model. Characterize the target model's predictive computing power value.

[0118] S240: Determine the operational quality information corresponding to the computing power prediction accuracy and operational information.

[0119] In one example, runtime information includes network stability information, failure rate information, and model effective runtime information. Model effective runtime information characterizes the effective runtime of the model.

[0120] In one embodiment, operational quality information can be determined based on the accuracy of computing power prediction and operational information, including network operational stability information, failure rate information, and effective model operation information.

[0121] In one example, the expression for determining the running quality information is shown in formula (13).

[0122] in, Characterizing operational quality information, Characterizing the accuracy of prediction based on computing power. Information characterizing network operational stability Information representing the effective operation of the model. Characterizes the failure rate.

[0123] S250: Based on the operational quality information, perform optimization operations on the target model that correspond to the operational quality information.

[0124] In this embodiment, after determining the operational quality information, optimization operations corresponding to the operational quality information can be performed on the target model.

[0125] In one embodiment, after determining the running quality information, it can be determined whether the running quality information is greater than or equal to a preset threshold. If the running quality information is less than the preset threshold, the correlation coefficient in the correspondence between the preset historical model task data and the historical model prediction computing power value can be reset and / or the sample size of historical data can be increased to achieve the optimization operation of the target model.

[0126] This invention obtains the computing power and operation information of the target model, determines the operation quality information of the target model, and performs corresponding optimization operations on the target model based on the operation quality information. This enables continuous iterative updates to the target model, thereby improving its quality.

[0127] In one embodiment, the actual computing power value of the target model is obtained, such as Figure 6 As shown, steps S221-S223 may be included.

[0128] S221: Determine the configuration information and scoring information of the hardware resources corresponding to the target model.

[0129] In this context, the configuration information of the hardware resources corresponding to the target model is equal to the configuration parameters of the target model.

[0130] In one example, the hardware resource rating information corresponding to the target model may include at least one of the following: general computing power rating, high-performance computing power rating, memory performance rating, and disk performance rating.

[0131] In one embodiment, the score information of at least one of the hardware resources, including general computing power score, high-performance computing power score, memory performance score, and disk performance score, can be determined based on the configuration information of the hardware resources corresponding to the target model.

[0132] S222: Obtain the hardware resource utilization information of the target model.

[0133] In one example, the hardware resource utilization information of the target model may include at least one of the following: CPU utilization information, GPU utilization information, memory utilization information, and hard disk utilization information.

[0134] In one embodiment, at least one of the following can be obtained when the target model's hardware resources are executing a task: CPU utilization information, GPU utilization information, memory utilization information, and hard disk utilization information.

[0135] S223: Add the product of the utilization rate information of each hardware resource in the target model and the corresponding hardware resource score information to obtain the actual computing power value of the target model.

[0136] In one embodiment, after obtaining the utilization information of each hardware resource of the target model, the value of the predicted computing power of the target model can be obtained by multiplying it by the corresponding hardware resource score information and then adding them together.

[0137] In one example, the expression for determining the historical model prediction computing power value is shown in formula (14).

[0138] in, Characterize the target model's predictive computing power value. A score representing the hardware resources of the central processing unit of the target model. The utilization rate of the central processing unit of the target model. A score representing the hardware resources of the image processor used to characterize the target model. The utilization rate of the image processor characterizing the target model. A score representing the hardware resources of the target model's memory. Characterizes the memory utilization of the target model. A score representing the hardware resources of the hard drive in the target model. The utilization rate of the hard drive in the target model is characterized.

[0139] In this embodiment of the invention, the utilization information of each hardware resource of the target model is added to the product of the corresponding hardware resource score information because the hardware resource utilization information reflects the dynamic performance information of the target model, while the hardware resource score information reflects the static performance information of the target model. Therefore, by comprehensively evaluating the actual computing power value of the target model from both static and dynamic performance perspectives, the accuracy of determining the predicted computing power value of the target model can be improved.

[0140] In one embodiment, the runtime information of the target model is obtained, such as... Figure 7 As shown, steps S224-S2210 may be included.

[0141] S224: Obtain the communication parameters corresponding to the target model.

[0142] In one example, the communication parameters include at least one of the following: bandwidth, latency, and packet loss rate during the target model's runtime.

[0143] In one embodiment, at least one of the communication parameters, namely bandwidth, latency, and packet loss rate, corresponding to the target model can be obtained.

[0144] S225: Determine the network operation stability information corresponding to the communication parameters.

[0145] In this embodiment, after obtaining the communication parameters, the network operation stability information corresponding to the communication parameters can be determined according to a preset formula.

[0146] In one example, the expression for determining network operational stability information is shown in formula (15).

[0147] in, Information characterizing network operational stability Characterizes the bandwidth at time i during the execution of the target model. Characterizes the time delay at the i-th time step during the execution of the target model. The packet loss rate at time i during the execution of the target model is represented by n1, the bandwidth data volume is represented by n2, the latency data volume is represented by n3, and the packet loss rate data volume is represented by n3.

[0148] S226: Obtain information on the number of failures of the target model within a preset period.

[0149] The preset cycle size is not fixed and can be adjusted according to user needs.

[0150] In this embodiment, the monitoring data processing module can be used to obtain information on the number of faults of the target model within a preset period.

[0151] S227: Determine the number of failures and the failure rate information corresponding to the preset cycle.

[0152] In this embodiment, the number of faults and the fault rate information corresponding to the preset period can be determined according to a preset formula.

[0153] In one example, the expression for determining the failure rate information is shown in formula (16).

[0154] in, Characterizing the failure rate, Characterizes the number of failures within a preset period. Characterizes the preset period.

[0155] S228: Obtain information on the number of faults, fault repair time, and model runtime of the target model within a preset period.

[0156] Among them, the fault repair time information represents the time required to repair the fault.

[0157] In one embodiment, the monitoring data processing module can be used to obtain information on the number of faults, fault repair time, and model runtime of the target model within a preset period.

[0158] S229: Determine the effective running information of the target model corresponding to the fault count, fault repair time, and model running time within a preset period.

[0159] In one embodiment, a preset relational formula can be used to determine the effective running information of the target model corresponding to the fault count information, fault repair time information, and model running time information within a preset period.

[0160] In one example, the expression for determining the effective operating information of the model is shown in formula (17).

[0161] in, Information representing the effective operation of the model. Characterizes the runtime information of the model. Characterizes the number of failures within a preset period. Characterizes the time required for the i-th fault repair.

[0162] S2210: Determine the operating information of the target model based on network operation stability information, failure rate information, and effective model operation information.

[0163] In this embodiment, after obtaining network operation stability information, failure rate information, and model effective operation information, the operation information of the target model can be determined based on the network operation stability information, failure rate information, and model effective operation information.

[0164] By acquiring network stability information, failure rate information, and effective running time of the target model, this embodiment of the invention can determine the operating information of the target model from multiple aspects such as network status, failure status, and running time, thereby improving the accuracy of determining the operating information.

[0165] In one embodiment, such as Figure 8 As shown, this embodiment of the invention includes a system 800 for generating model configuration parameters. This system 800 includes a model information acquisition module 801, a model hardware resource evaluation module 802, a model computing power analysis module 803, a target model task requirement input module 804, a target model computing power prediction module 805, a target model operation monitoring module 806, a monitoring data processing module 807, a target model operation quality index calculation module 808, and a database 809. The model hardware resource evaluation module 802 is connected to the model computing power analysis module 803. The model computing power analysis module 803 and the target model task requirement input module 804 are connected to the target model computing power prediction module 805. The target model computing power prediction module 805, the target model operation monitoring module 806, the monitoring data processing module 807, and the target model operation quality index calculation module 808 are sequentially connected. The target model computing power prediction module 805 is connected to the target model operation quality index calculation module 808. All modules in the system are connected to the database 809.

[0166] The model information acquisition module 801 is used to collect preset task data, including task type, dataset size, preset model inference time, and preset model application accuracy. The model hardware resource evaluation module 802 is used to evaluate the model's hardware resources and obtain hardware resource scoring information. The model computing power analysis module 803 is used to perform correlation calculations between the preset model task data and the calculated hardware resource scoring information. The target model task requirement input module 804 is used by the user to input the task type, dataset size, preset model inference time, and preset model application accuracy required by the target model. The target model computing power prediction module 805 is used to determine the predicted computing power value of the target model and generate corresponding target model configuration parameters. The target model operation monitoring module 806 is used to monitor and record the operation information of the target model. The monitoring data processing module 807 is used to process the model operation data recorded within a preset period. The target model operation quality index calculation module 808 is used to determine the operation quality information of the target model.

[0167] In one embodiment, after calculating the target model's predicted computing power value, if the target model's predicted computing power value is inconsistent with the historical model's predicted computing power value, then the target historical model's predicted computing power value with the smallest difference from the target model's predicted computing power value can be determined. The historical model configuration parameters corresponding to the target historical model's predicted computing power value are obtained, and these historical model configuration parameters are adjusted based on the feature information of the task data corresponding to the target model to obtain the target model's configuration parameters.

[0168] In embodiments of the present invention, when the predicted computing power value of the target model is inconsistent with the predicted computing power value of the historical model, the predicted computing power value of the target historical model with the smallest difference from the predicted computing power value of the target model can be determined, and the configuration parameters corresponding to the predicted computing power value of the target historical model can be adjusted to obtain the configuration parameters of the target model. Since the difference between the predicted computing power value of the target historical model and the predicted computing power value of the target model is the smallest, the required performance of the target historical model is closest to the required performance of the target model. Therefore, the configuration parameters corresponding to the target historical model are modified based on the feature information of the task data corresponding to the target model to obtain the configuration parameters of the target model. Compared with the method of determining the configuration parameters corresponding to the computing power value in related technologies, the present invention makes the configuration parameters corresponding to the target historical model have a high similarity with the configuration parameters of the target model, which is equivalent to fine-tuning the configuration parameters corresponding to the target historical model, thereby improving the accuracy of determining the configuration parameters of the target model.

[0169] like Figure 9 As shown, this embodiment of the invention also provides a device 900 for generating model configuration parameters, the device 900 comprising: The acquisition module 901 is used to acquire preset target model task data, which represents the requirement data corresponding to the target model execution task. The determination module 902 is used to determine the target model prediction computing power value corresponding to the preset target model task data based on the correspondence between the preset historical model task data and the historical model prediction computing power value. The preset historical model task data represents the demand data corresponding to the historical model executing the historical task. The determination module 902 is also used to determine, when the target model prediction computing power value is equal to the historical model prediction computing power value, to use the historical model configuration parameters corresponding to the historical model prediction computing power value as the target model configuration parameters, and the target model configuration parameters are used to construct the target model.

[0170] In one embodiment, the determining module 902 is specifically used for: Determine the hardware resource configuration information and scoring information corresponding to the historical model; Obtain hardware resource utilization information from historical models; The historical model's predicted computing power value is obtained by multiplying the utilization rate information of each hardware resource in the historical model with the corresponding hardware resource's score information. Obtain preset historical model task data; The correlation coefficient is obtained by performing correlation calculations on the pre-set historical model task data and the historical model prediction computing power values. Based on the correlation coefficient, the correspondence between the preset historical model task data and the historical model prediction computing power value is determined.

[0171] In one embodiment, the acquisition module 901 is specifically used for: When the hardware resources corresponding to the historical model include CPUs, the CPU score information is obtained by multiplying at least one piece of information about the number of servers, the number of cores, the single-core clock frequency, and the single-cycle floating-point calculation value for each type of CPU. When the hardware resources corresponding to the historical model include GPUs, the GPU score information is obtained by multiplying at least one piece of information about the number of servers, the number of cores, the single-core clock frequency, and the single-cycle floating-point calculation value for each type of GPU. If the hardware resources corresponding to the historical model include memory, memory score information is obtained based on the sum of at least one capacity information and speed information corresponding to memory. If the hardware resources corresponding to the historical model include hard drives, the hard drive rating information is obtained based on the sum of at least one capacity information and speed information corresponding to the hard drive. The hardware resource rating information corresponding to the historical model is obtained by summing at least one of the CPU rating information, GPU rating information, memory rating information, and hard disk rating information.

[0172] In one embodiment, the acquisition module 901 is specifically used for: Given that the historical model task data includes historical task type data, the Pearson correlation between the historical task type data and the historical model prediction computing power value is calculated to obtain the first correlation coefficient. Given that the historical model task data includes the amount of historical dataset data, the Pearson correlation between the amount of historical dataset data and the historical model prediction computing power value is calculated to obtain the second correlation coefficient. Given that the preset historical model task data includes the preset historical model inference time, the preset historical model inference time and the historical model prediction computing power value are calculated using Pearson correlation to obtain the third correlation coefficient. Given that the preset historical model task data includes the preset historical model accuracy, the preset historical model accuracy and the historical model prediction computing power value are calculated using Pearson correlation to obtain the fourth correlation coefficient. The correlation coefficient is obtained by summing at least one of the first, second, third, and fourth correlation coefficients.

[0173] In one embodiment, the apparatus 900 further includes a construction module for constructing a target model based on target model configuration parameters; Obtain the actual computing power and operational information of the target model; Determine the computing power prediction accuracy corresponding to the actual computing power value of the target model and the predicted computing power value of the target model; Determine the accuracy of computing power prediction and the corresponding operational quality information based on operational information; Based on the operational quality information, perform optimization operations on the target model that correspond to the operational quality information.

[0174] In one embodiment, the determining module 902 is configured to: Determine the configuration information and scoring information of the hardware resources corresponding to the target model; Obtain the hardware resource utilization information of the target model; The actual computing power of the target model is obtained by multiplying the utilization rate information of each hardware resource in the target model with the corresponding hardware resource score information.

[0175] In one embodiment, the acquisition module 901 is used for: Obtain the communication parameters corresponding to the target model; Determine the network operational stability information corresponding to the communication parameters; Obtain information on the number of failures of the target model within a preset period; Determine the number of failures and the failure rate information corresponding to the preset cycle; Obtain information on the number of failures, failure repair time, and model runtime of the target model within a preset period; Determine the effective operational information of the target model corresponding to the fault count, fault repair time, and model runtime information within a preset period; Based on network stability information, failure rate information, and effective model operation information, the operation information of the target model is determined.

[0176] Figure 9 Each module / unit in the illustrated device has the ability to implement Figures 1 to 7 The functions of each step performed by the application system and the corresponding technical effects it achieves are described briefly and will not be elaborated here.

[0177] Figure 10 A schematic diagram of the hardware structure for generating model configuration parameters provided in an embodiment of the present invention is shown.

[0178] The device that generates the model configuration parameters may include a processor 1001 and a memory 1002 storing computer program instructions.

[0179] Specifically, the processor 1001 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0180] Memory 1002 may include mass storage for data or instructions. For example, and not limitingly, memory 1002 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 1002 may include removable or non-removable (or fixed) media. Where appropriate, memory 1002 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 1002 is non-volatile solid-state memory.

[0181] Memory 1002 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of this disclosure.

[0182] The processor 1001 reads and executes computer program instructions stored in the memory 1002 to implement any of the model configuration parameter generation methods in the above embodiments.

[0183] In one example, the device for generating model configuration parameters may also include a communication interface 1003 and a bus 1004. For example, Figure 10 As shown, the processor 1001, memory 1002, and communication interface 1003 are connected through bus 1004 and complete communication with each other.

[0184] The communication interface 1003 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of the present invention.

[0185] Bus 1004 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 1004 may include one or more buses. Although specific buses are described and illustrated in embodiments of the invention, the invention contemplates any suitable bus or interconnect. Additionally, in conjunction with the method for generating model configuration parameters in the above embodiments, embodiments of the invention also provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by the processor, they implement any of the methods for generating model configuration parameters in the above embodiments.

[0186] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements a method for generating any of the model configuration parameters described in the above embodiments.

[0187] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0188] The functional blocks shown in the above structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (ROM), flash memory, erasable read-only memory (EROM), floppy disks, compact disc read-only memory (CD-ROM), optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0189] It should also be noted that the exemplary embodiments mentioned in this invention describe methods or systems based on a series of steps or apparatus. However, this invention is not limited to the order of the steps described above; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0190] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0191] The above are merely specific embodiments of the present invention. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A method for generating model configuration parameters, characterized in that, The method includes: Acquire preset target model task data, wherein the preset target model task data represents the requirement data corresponding to the target model execution task; Based on the correspondence between the preset historical model task data and the historical model prediction computing power value, the target model prediction computing power value corresponding to the preset target model task data is determined. The preset historical model task data represents the demand data corresponding to the historical model executing historical tasks. If the target model's predicted computing power value is equal to the historical model's predicted computing power value, then the historical model configuration parameters corresponding to the historical model's predicted computing power value are determined as the target model configuration parameters, which are used to construct the target model.

2. The method according to claim 1, characterized in that, Before determining the target model prediction computing power value corresponding to the preset target model task data based on the correspondence between preset historical model task data and historical model prediction computing power values, the method further includes: Determine the hardware resource configuration information and corresponding hardware resource scoring information for the historical model. Obtain the hardware resource utilization information of the historical model; The utilization rate information of each hardware resource in the historical model is multiplied by the corresponding score information of the hardware resource to obtain the predicted computing power value of the historical model. Obtain preset historical model task data; The correlation coefficient is obtained by performing correlation calculation on the preset historical model task data and the historical model prediction computing power value. Based on the correlation coefficient, the correspondence between the preset historical model task data and the historical model prediction computing power value is determined.

3. The method according to claim 2, characterized in that, The hardware resources corresponding to the historical model include at least one of a central processing unit (CPU), a graphics processing unit (GPU), memory, and hard disk; the CPU configuration information includes at least one of the following: number of servers, number of cores, single-core clock speed, and single-cycle floating-point calculation value; the GPU configuration information includes at least one of the following: number of servers, number of cores, single-core clock speed, and single-cycle floating-point calculation value; the memory configuration information includes at least one of the following: capacity and speed; the hard disk configuration information includes at least one of the following: capacity and speed; the scoring information for determining the hardware resources corresponding to the configuration information of the hardware resources corresponding to the historical model includes: When the hardware resources corresponding to the historical model include the CPU, the CPU score information is obtained by multiplying at least one piece of information about the number of servers, the number of cores, the single-core clock frequency, and the single-cycle floating-point calculation value for each type of CPU. When the hardware resources corresponding to the historical model include the GPU, the GPU score information is obtained by multiplying at least one piece of information about the number of servers, the number of cores, the single-core clock frequency, and the single-cycle floating-point calculation value for each type of GPU. If the hardware resources corresponding to the historical model include the memory, memory scoring information is obtained based on the sum of at least one capacity information and speed information corresponding to the memory. When the hardware resources corresponding to the historical model include the hard disk, the hard disk rating information is obtained based on the sum of at least one capacity information and speed information corresponding to the hard disk; The hardware resource rating information corresponding to the historical model is obtained by summing at least one of the CPU rating information, the GPU rating information, the memory rating information, and the hard disk rating information.

4. The method according to claim 2, characterized in that, The preset historical model task data includes at least one of the following: historical task type data, historical dataset data volume, preset historical model inference time, and preset historical model accuracy. The step of calculating the correlation coefficient between the preset historical model task data and the historical model prediction computing power value includes: When the preset historical model task data includes the historical task type data, the historical task type data and the historical model prediction computing power value are subjected to Pearson correlation calculation to obtain the first correlation coefficient. When the preset historical model task data includes the amount of historical dataset data, the amount of historical dataset data and the historical model prediction computing power value are calculated using Pearson correlation to obtain a second correlation coefficient. When the preset historical model task data includes the preset historical model inference time, the preset historical model inference time and the historical model prediction computing power value are calculated using Pearson correlation to obtain a third correlation coefficient. When the preset historical model task data includes the preset historical model accuracy, the preset historical model accuracy and the historical model prediction computing power value are calculated using Pearson correlation to obtain the fourth correlation coefficient. The correlation coefficient is obtained by summing at least one of the first correlation coefficient, the second correlation coefficient, the third correlation coefficient, and the fourth correlation coefficient.

5. The method according to claim 1, characterized in that, After determining the target model configuration parameters based on the predicted computing power value of the target model, the method further includes: Construct the target model based on the target model configuration parameters; Obtain the actual computing power and operational information of the target model; Determine the computing power prediction accuracy corresponding to the actual computing power value of the target model and the predicted computing power value of the target model; Determine the accuracy of the computing power prediction and the corresponding operational quality information of the operational information; Based on the operational quality information, optimization operations corresponding to the operational quality information are performed on the target model.

6. The method according to claim 5, characterized in that, Obtaining the actual computing power value of the target model includes: Determine the hardware resource configuration information and corresponding hardware resource scoring information for the target model; Obtain the hardware resource utilization information of the target model; The actual computing power value of the target model is obtained by multiplying the utilization rate information of each hardware resource of the target model with the corresponding hardware resource score information.

7. The method according to claim 5, characterized in that, The step of obtaining the runtime information of the target model includes: Obtain the communication parameters corresponding to the target model; Determine the network operation stability information corresponding to the communication parameters; Obtain the number of failures of the target model within a preset period; Determine the number of failures and the failure rate information corresponding to the preset period; Obtain information on the number of faults, fault repair time, and model runtime of the target model within a preset period; Determine the effective running information of the target model corresponding to the fault count information, fault repair time information, and model runtime information within a preset period; The operating information of the target model is determined based on the network operation stability information, the failure rate information, and the effective operation information of the model.

8. An apparatus for generating model configuration parameters, characterized in that, The device includes: The acquisition module is used to acquire preset target model task data, which represents the requirement data corresponding to the target model execution task. The determining module is used to determine the target model prediction computing power value corresponding to the preset target model task data based on the correspondence between preset historical model task data and historical model prediction computing power value. The preset historical model task data represents the demand data corresponding to the historical model executing historical tasks. The determining module is further configured to, when the target model prediction computing power value is equal to the historical model prediction computing power value, determine the historical model configuration parameters corresponding to the historical model prediction computing power value as the target model configuration parameters, and the target model configuration parameters are used to construct the target model.

9. A device for generating model configuration parameters, characterized in that, The device includes: a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the method for generating model configuration parameters as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer storage medium stores computer program instructions, which, when executed by a processor, implement the method for generating model configuration parameters as described in any one of claims 1-7.

11. A computer program product, characterized in that, When the instructions in the computer program product are executed by a processor in an electronic device, the electronic device causes the electronic device to perform the method for generating model configuration parameters as described in any one of claims 1-7.

Images