Communication mode determination method and device, computer equipment, storage medium and computer program product

Abstract

The embodiment of the invention provides a communication mode determination method and device, computer equipment, a storage medium and a computer program product, and relates to the technical field of computers. The method comprises the steps of obtaining task description information of a target calculation task; obtaining an index weight of each influence index according to the task description information and a plurality of influence indexes of the target calculation task; obtaining an index score of each communication mode in a plurality of communication modes of the target calculation task under each influence index based on the task description information by using a pre-constructed communication mode evaluation model; obtaining a comprehensive score of each communication mode according to the index score and the index weight of each influence index; and according to the comprehensive score, determining a target communication mode for the target computing service in the plurality of communication modes. According to the method, the accuracy of determining the communication mode is improved, so that the efficiency and the accuracy of data transmission between the GPUs are improved, and the efficiency and the accuracy of executing the target calculation task are improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a communication mode determination method, apparatus, computer equipment, storage medium and computer program product. Background Art

[0002] In the training process of large models, it is usually necessary to realize data transmission between different GPUs in large-scale graphics processing units (GPUs), for example, to transmit a large amount of parameters and gradient information.

[0003] In the current communication technology between GPUs of servers, most of them adopt a network architecture CLOS architecture with a multi-layer switch design. The transmission method is relatively single and cannot meet different business scenarios. The universality is low. For some business scenarios, this transmission method is prone to link congestion, which will lead to low efficiency and accuracy of data transmission between GPUs. Summary of the invention

[0004] Based on this, it is necessary to provide a communication mode determination method, apparatus, computer device, storage medium and computer program product to address the above technical issues.

[0005] In a first aspect, the present application provides a method for determining a communication mode. The method comprises:

[0006] Get the task description information of the target computing task;

[0007] According to the task description information and a plurality of influencing indicators of the target computing task, obtaining an indicator weight of each of the influencing indicators;

[0008] Using a pre-built communication mode evaluation model, based on the task description information, obtaining an indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task;

[0009] Obtaining a comprehensive score for each of the communication methods according to the indicator score and the indicator weight of each of the influencing indicators;

[0010] According to the comprehensive score, a target communication mode for the target computing task is determined among the multiple communication modes.

[0011] In one of the embodiments, the pre-built communication mode evaluation model is used to obtain, based on the task description information, an index score of each of the multiple communication modes of the target computing task under each of the influencing indicators, including: obtaining transmission description information of each of the communication modes; inputting the task description information and the transmission description information into the pre-built communication mode evaluation model to obtain the index score of each of the communication modes under each of the influencing indicators; wherein the communication mode evaluation model calculates the index score of each of the communication modes under each of the influencing indicators based on the task description information and the transmission description information.

[0012] In one of the embodiments, the task description information and the transmission description information are input into the pre-built communication mode evaluation model to obtain the indicator score of each communication mode under each influencing indicator, including: inputting the task description information and the transmission description information into the pre-built communication mode evaluation model; having the feature module in the communication mode evaluation model perform feature extraction on the task description information and the transmission description information respectively, determine and output the task feature information of the target computing task, and the transmission feature information of each communication mode; and obtaining the indicator score of each communication mode under each influencing indicator based on the task feature information and the transmission feature information.

[0013] In one of the embodiments, obtaining a comprehensive score for each of the communication methods based on the indicator score and the indicator weight of each of the influencing indicators includes: determining the weighted indicator score for each of the communication methods under each of the influencing indicators by multiplying the indicator score and the indicator weight; and obtaining the comprehensive score for each of the communication methods based on the sum of the weighted indicator scores for each of the communication methods under multiple influencing indicators.

[0014] In one of the embodiments, determining a target communication method for the target computing task among the multiple communication methods based on the comprehensive score includes: determining the communication method with the highest comprehensive score among the multiple communication methods as the target communication method.

[0015] In one of the embodiments, the indicator weight of each influencing indicator is obtained based on the task description information and multiple influencing indicators of the target computing task, including: determining the correlation between each influencing indicator and the target computing task based on the task description information; and obtaining the indicator weight of each influencing indicator based on the correlation corresponding to each influencing indicator.

[0016] In a second aspect, the present application provides a communication mode determination device. The device includes:

[0017] A data acquisition module is used to obtain task description information of a target computing task;

[0018] A weight calculation module, used to obtain the indicator weight of each influencing indicator according to the task description information and multiple influencing indicators of the target computing task;

[0019] A first scoring calculation module is used to obtain an indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators by using a pre-built communication mode evaluation model based on the task description information; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task;

[0020] A second score calculation module, used for obtaining a comprehensive score of each of the communication modes according to the indicator score and the indicator weight of each of the influencing indicators;

[0021] A screening module is used to determine a target communication mode for the target computing task from among the multiple communication modes according to the comprehensive score.

[0022] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0023] Get the task description information of the target computing task;

[0024] According to the task description information and a plurality of influencing indicators of the target computing task, obtaining an indicator weight of each of the influencing indicators;

[0025] Using a pre-built communication mode evaluation model, based on the task description information, obtaining an indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task;

[0026] Obtaining a comprehensive score for each of the communication methods according to the indicator score and the indicator weight of each of the influencing indicators;

[0027] According to the comprehensive score, a target communication mode for the target computing task is determined among the multiple communication modes.

[0028] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0029] Get the task description information of the target computing task;

[0030] According to the task description information and a plurality of influencing indicators of the target computing task, obtaining an indicator weight of each of the influencing indicators;

[0031] Using a pre-built communication mode evaluation model, based on the task description information, obtaining an indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task;

[0032] Obtaining a comprehensive score for each of the communication methods according to the indicator score and the indicator weight of each of the influencing indicators;

[0033] According to the comprehensive score, a target communication mode for the target computing task is determined among the multiple communication modes.

[0034] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0035] Get the task description information of the target computing task;

[0036] According to the task description information and a plurality of influencing indicators of the target computing task, obtaining an indicator weight of each of the influencing indicators;

[0037] Using a pre-built communication mode evaluation model, based on the task description information, obtaining an indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task;

[0038] Obtaining a comprehensive score for each of the communication methods according to the indicator score and the indicator weight of each of the influencing indicators;

[0039] According to the comprehensive score, a target communication mode for the target computing task is determined among the multiple communication modes.

[0040] In the above communication mode determination method, device, computer equipment, storage medium and computer program product, first, the task description information of the target computing task can be obtained; next, the indicator weight of each influencing indicator can be obtained according to the task description information and multiple influencing indicators of the target computing task; further, a pre-built communication mode evaluation model can be used to obtain the indicator score of each communication mode under each influencing indicator in the multiple communication modes of the target computing task based on the task description information; multiple communication modes are used to realize communication between multiple graphics processors; multiple graphics processors are used to execute the target computing task; further, the comprehensive score of each communication mode can be obtained according to the indicator score and the indicator weight of each influencing indicator; finally, the target communication mode for the target computing task can be determined in multiple communication modes according to the comprehensive score. In the method provided in the embodiment of the present application, a communication mode evaluation model can be pre-trained, and the communication mode evaluation model can realize the evaluation of different communication modes in combination with multiple influencing indicators and multiple influencing factors, and further, the optimal communication mode of the target computing task can be determined in multiple communication modes, which can improve the accuracy of communication mode determination, and further, can improve the efficiency and accuracy of data transmission between GPUs, and can improve the efficiency and accuracy of executing the target computing task. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A flow chart of a method for determining a communication mode provided in an embodiment of the present application;

[0043] Figure 2 A structural block diagram of a CLOS architecture provided in an embodiment of the present application;

[0044] Figure 3 A schematic diagram of a process for obtaining a comprehensive score of each communication method provided in an embodiment of the present application;

[0045] Figure 4 A schematic diagram of a process for obtaining the indicator weights of various influencing indicators provided in an embodiment of the present application;

[0046] Figure 5 A structural block diagram of a communication mode determination device provided in an embodiment of the present application;

[0047] Figure 6An internal structure diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] In an exemplary embodiment, Figure 1 As shown, a method for determining a communication mode is provided. This embodiment uses the method applied to a server as an example for illustration. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0050] Step 102: Obtain task description information of the target computing task.

[0051] The target computing task may be a computing task involving multiple graphics processing units (GPUs) computing, and the task description information may be information used to describe the characteristics of the target computing task, such as attribute information, bandwidth condition information, network delay condition information, task volume and transmission flow information of the target computing task. In the process of implementing the target computing task, multiple GPUs are usually required to process in parallel, such as computing tasks in the fields of deep learning, large-scale scientific computing, graphics rendering and distributed computing. The graphics processing unit (GPU) is an electronic processor specifically used to process graphics rendering and parallel computing tasks. It contains hundreds or thousands of smaller, more efficient cores that are specifically used to process large amounts of data and perform complex mathematical and geometric calculations. Compared with traditional CPUs (central processing units), GPUs excel in processing visually intensive tasks such as video games, 3D animation and graphic design, and their parallel computing capabilities make them increasingly important in the fields of scientific simulation, data analysis, deep learning and machine learning. In a multi-GPU system, different GPUs can interact with each other for data transmission, such as exchanging data, synchronizing computing results or coordinating computing tasks. There are many communication methods for data transmission between GPUs of servers, such as cross-Leaf / spine communication and GPU direct connection in CLOS architecture. Figure 2As shown, the first communication mode can be a CLOS architecture. The CLOS architecture is a network architecture designed based on a multi-layer switch. It is widely used in data centers and large-scale network environments to provide efficient bandwidth and low latency. The core idea of ​​the CLOS architecture is to form a network structure through multi-layer switches, which can usually include a Leaf layer and a Spine layer. The Leaf layer can be an access layer that directly connects terminal devices (such as servers, storage devices, etc.). The Leaf switch is responsible for processing traffic entering or leaving the network; the Spine layer can be a core layer that connects all Leaf switches. The Spine switches are not directly connected to terminal devices, but they are core components for communication across Leaf switches. The second communication mode can be a GPU direct connection, which can be understood as direct data transmission between GPUs and between GPUs and other hardware devices (such as network interface cards, storage devices, etc.) without the intermediary of the CPU or host memory. The main purpose of the GPU direct connection technology is to optimize the performance of computing-intensive applications (such as deep learning, scientific computing, big data analysis, etc.) by reducing latency in data transmission, reducing CPU burden, and increasing data throughput.

[0052] Step 104 , obtaining the indicator weight of each influencing indicator according to the task description information and multiple influencing indicators of the target computing task.

[0053] Among them, the multiple influencing indicators of the target computing task can be factors that affect the task realization status of the target computing task and the data transmission quality during the task execution. For different types of computing tasks, the corresponding influencing factors may be different. If it is necessary to determine the communication mode for the data transmission of the target computing task, multiple influencing indicators of the target computing task can be pre-set, such as network bandwidth, network delay and communication cost. The multiple influencing indicators can be indicators that affect the data transmission between different GPUs during the execution of the target computing task. In a possible implementation, multiple initial influencing indicators can be determined, and the correlation between each initial influencing indicator and the target computing task is obtained according to the task description information of the target computing task, and then the initial influencing indicator with the correlation greater than the preset correlation threshold can be determined as the influencing indicator of the target computing task. Further, the indicator weight of each influencing indicator can also be obtained according to the correlation between each influencing indicator in the multiple influencing indicators and the target computing task. The greater the correlation, the greater the indicator weight of the corresponding influencing indicator; the smaller the correlation, the smaller the indicator weight of the corresponding influencing indicator.

[0054] Step 106, using a pre-built communication mode evaluation model, based on the task description information, obtain the indicator score of each communication mode of the target computing task under each influencing indicator; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task.

[0055] Among them, a communication mode evaluation model can be pre-constructed, and the communication mode evaluation model can obtain the comprehensive score of each of the multiple communication modes between GPUs under multiple dimensions and multiple influencing indicators according to the task description information of the target computing task for the target computing task, that is, the comprehensive evaluation result. Then, according to the comprehensive score of each of the multiple communication modes, the target communication mode of the target computing task can be screened out from the multiple communication modes, and the target communication mode can be used to realize the data transmission between different GPUs during the execution of the target computing task. In this step, the index score of each of the multiple communication modes of the target computing task under each influencing indicator can be obtained. In the process of evaluating each communication mode, evaluations can be performed from multiple angles and dimensions, such as transmission performance, bandwidth, delay, scalability and compatibility. In the process of selecting the target communication mode of the target computing task from the multiple communication modes, multiple influencing indicators of the target computing task can be used as evaluation indicators for each communication mode, such as network bandwidth, network delay and communication cost. Then, the indicator evaluation results of each communication mode under each of the multiple influencing indicators can be determined according to the characteristic information of each communication mode and the task description information of the target computing task, that is, the indicator score. For example, for the first communication mode that uses the CLOS architecture for transmission, its indicator score in terms of the first influencing indicator (e.g., network bandwidth) is , the score of the second influencing indicator (e.g., network delay) is , the indicator score for the third impact indicator (e.g., communication cost) is ,…, the indicator score for the nth impact indicator is , the multi-dimensional scoring data corresponding to the first communication mode can be expressed as Similarly, for the second communication method of GPU direct connection, its index score in the first influencing index (e.g., network bandwidth) is , the score of the second influencing indicator (e.g., network delay) is , the indicator score for the third impact indicator (e.g., communication cost) is ,…, the indicator score for the nth impact indicator is , the multi-dimensional scoring data corresponding to the second communication mode can be expressed as Moreover, the higher the index score, the better the performance of the corresponding communication method under the corresponding impact index, and the more conducive to the data transmission between different GPUs during the execution of the target computing task. For example, the more sufficient the bandwidth, the lower the latency, and the lower the communication cost.

[0056] Step 108: Obtain a comprehensive score of each communication mode according to the indicator score and the indicator weight of each influencing indicator.

[0057] Among them, in this step, the indicator scores of each of the multiple communication modes under the multiple influencing indicators can be weighted and summed to obtain the corresponding comprehensive scores of each communication mode, and the comprehensive scores are used to evaluate the overall transmission performance of the corresponding communication mode.

[0058] Step 110 , determining a target communication mode for the target computing task from among multiple communication modes according to the comprehensive score.

[0059] Among them, in this step, the communication method with the highest comprehensive score can be determined as the target communication method of the target computing task, and the target communication method can be used to realize data transmission between different GPUs during the execution of the target computing task.

[0060] In the method of this embodiment, first, the task description information of the target computing task can be obtained; next, the indicator weight of each of the influencing indicators can be obtained according to the task description information and the multiple influencing indicators of the target computing task; further, the pre-constructed communication mode evaluation model can be used to obtain the indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators based on the task description information; then, the comprehensive score of each of the communication modes can be obtained according to the indicator score and the indicator weight of each of the influencing indicators; finally, the target communication mode for the target computing task can be determined according to the comprehensive score. In the method provided by the embodiment of the present application, a communication mode evaluation model can be pre-trained, and the communication mode evaluation model can realize the evaluation of different communication modes in combination with multiple influencing indicators and multiple influencing factors, and then, the optimal communication mode of the target computing task can be determined in multiple communication modes, which can improve the accuracy of communication mode determination, and then, the efficiency and accuracy of data transmission between GPUs can be improved, and the efficiency and accuracy of executing the target computing task can be improved.

[0061] In an exemplary embodiment, step 106 may include:

[0062] Acquire transmission description information of each communication mode; input the task description information and the transmission description information into a pre-built communication mode evaluation model to obtain the indicator score of each communication mode under each influencing indicator; wherein the communication mode evaluation model calculates the indicator score of each communication mode under each influencing indicator based on the task description information and the transmission description information.

[0063] Among them, multiple impact indicators of the target computing task can be used as evaluation indicators of each of the multiple communication methods, and each of the multiple communication methods can be evaluated in multiple dimensions based on the multiple evaluation indicators. In an embodiment of the present application, a pre-built communication method evaluation model can be used to evaluate each communication method, and the basis data for the evaluation may include the transmission description information of the corresponding communication method, the task description information of the target computing task, and the current network status information. In a possible implementation, the communication method evaluation model can be a neural network model, and the communication method evaluation model can include multiple modules, including but not limited to a feature extraction module and a classification module.

[0064] In the method of this embodiment, a communication mode evaluation model can be pre-constructed, and each communication mode can be evaluated using the model, thereby improving the accuracy of the communication mode evaluation.

[0065] In an exemplary embodiment, the step of inputting the task description information and the transmission description information into a pre-built communication mode evaluation model to obtain the indicator score of each communication mode under each influencing indicator may include:

[0066] The task description information and the transmission description information are input into a pre-built communication mode evaluation model; the feature module in the communication mode evaluation model extracts features from the task description information and the transmission description information respectively, determines and outputs the task feature information of the target computing task, and the transmission feature information of each communication mode; based on the task feature information and the transmission feature information, obtains the indicator score of each communication mode under each influencing indicator.

[0067] Among them, the feature extraction module can obtain the task feature information of the target computing task according to the task description information, and the task feature information can include the indicator feature corresponding to each of the multiple influencing indicators, and the indicator feature can represent the required feature of the target computing task in the corresponding influencing indicator. For example, when the influencing indicator is network delay, the indicator feature can represent the required feature of the target computing task in terms of network delay, such as the size of the delay and the preset delay interval. The feature extraction module can also obtain the transmission feature information of the corresponding communication mode according to the transmission description information of each communication mode, and the transmission feature information can include the feature information of each communication mode under each of the multiple influencing indicators, such as network delay feature and bandwidth feature. In a possible implementation, when obtaining the indicator feature of each communication mode under each of the multiple influencing indicators, the task feature information of the target computing task can be used as reference information, and the communication mode can be evaluated according to the matching degree of the feature information of each communication mode under each of the multiple influencing indicators and the indicator feature of the target computing task in the same influencing indicator, and the indicator score of the communication mode under the corresponding influencing indicator can be obtained.

[0068] In the method of this embodiment, a pre-trained communication mode evaluation model can be combined to evaluate different communication modes, and then, the optimal communication mode for the target computing task can be determined among multiple communication modes, which can improve the accuracy of communication mode determination.

[0069] In an exemplary embodiment, Figure 3 As shown, step 108 may include steps 302 to 304. Among them:

[0070] Step 302: The product of the indicator score and the indicator weight is determined as the weighted indicator score of each communication mode under each influencing indicator.

[0071] Step 304: Obtain a comprehensive score of each communication method according to the sum of the weighted index scores of each communication method under multiple influencing indicators.

[0072] The indicator scores of each of the multiple communication modes under the multiple influencing indicators can be weighted and summed to obtain the corresponding comprehensive scores of each communication mode. The comprehensive scores are used to evaluate the overall transmission performance of the corresponding communication mode. See formula (1).

[0073]

[0074] in, It can be the corresponding communication method The indicator score under the nth impact indicator; It can be the indicator weight of the nth influencing indicator; It can be the corresponding communication method The comprehensive rating of .

[0075] In the method of this embodiment, different communication modes can be evaluated in multiple dimensions, and multiple influencing factors can be considered. A more comprehensive evaluation of each communication mode can be performed for the target computing task, which can improve the accuracy of the evaluation and the accuracy of determining the communication mode. In turn, the efficiency and accuracy of data transmission between GPUs can be improved, and the efficiency and accuracy of executing the target computing task can be improved.

[0076] In an exemplary embodiment, step 110 may include:

[0077] A communication method with the highest comprehensive score among the multiple communication methods is determined as a target communication method.

[0078] In an exemplary embodiment, Figure 4 As shown, step 104 may include steps 402 to 404. Among them:

[0079] Step 402: Determine the relevance of each influencing indicator to the target computing task according to the task description information.

[0080] Step 404: Obtain the indicator weight of each influencing indicator according to the correlation corresponding to each influencing indicator.

[0081] The correlation can be used to indicate the degree of association between each influencing indicator and the target computing task, the degree of influence on the data transmission between GPUs of the target computing task, and the degree of contribution to the selection of the target communication method of the target computing task. Thus, the indicator weight of each influencing indicator can be obtained according to the correlation between each influencing indicator among the multiple influencing indicators and the target computing task. The greater the correlation, the greater the indicator weight of the corresponding influencing indicator; the smaller the correlation, the smaller the indicator weight of the corresponding influencing indicator.

[0082] In the method of this embodiment, the indicator weight of each influencing indicator can be obtained through correlation calculation, which can improve the accuracy of subsequent communication mode evaluation.

[0083] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0084] Based on the same inventive concept, the embodiment of the present application also provides a communication mode determination device for implementing the communication mode determination method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more communication mode determination device embodiments provided below can refer to the limitations of the communication mode determination method above, and will not be repeated here.

[0085] In one embodiment, Figure 5 As shown, a communication mode determination device is provided, comprising: a data acquisition module 502, a weight calculation module 504, a first score calculation module 506, a second score calculation module 508 and a screening module 510, wherein:

[0086] A data acquisition module 502 is used to acquire task description information of a target computing task;

[0087] A weight calculation module 504, configured to obtain an indicator weight of each influencing indicator according to the task description information and a plurality of influencing indicators of the target computing task;

[0088] A first score calculation module 506 is used to obtain the index score of each of the multiple communication modes of the target computing task under each of the influencing indicators by using a pre-built communication mode evaluation model based on the task description information; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task;

[0089] A second score calculation module 508, configured to obtain a comprehensive score of each of the communication modes according to the indicator score and the indicator weight of each of the influencing indicators;

[0090] The screening module 510 is configured to determine, according to the comprehensive score, a target communication method for the target computing service from among the multiple communication methods.

[0091] In one embodiment, the first scoring calculation module 506 is further used to: obtain transmission description information of each of the communication methods; input the task description information and the transmission description information into a pre-built communication method evaluation model to obtain the indicator score of each of the communication methods under each of the impact indicators; wherein the communication method evaluation model calculates the indicator score of each of the communication methods under each of the impact indicators based on the task description information and the transmission description information.

[0092] In one embodiment, the first scoring calculation module 506 is further used to: input the task description information and the transmission description information into the pre-built communication mode evaluation model; the feature module in the communication mode evaluation model respectively extracts features from the task description information and the transmission description information, determines and outputs the task feature information of the target computing task, and the transmission feature information of each of the communication modes; and obtains the indicator score of each of the communication modes under each of the influencing indicators based on the task feature information and the transmission feature information.

[0093] In one embodiment, the second score calculation module 508 is also used to: determine the product of the indicator score and the indicator weight as the weighted indicator score of each communication method under each influencing indicator; and obtain the comprehensive score of each communication method according to the sum of the weighted indicator scores of each communication method under multiple influencing indicators.

[0094] In one embodiment, the screening module 510 is further configured to: determine the communication method with the highest comprehensive score among the multiple communication methods as the target communication method.

[0095] In one embodiment, the weight calculation module 504 is further used to: determine the relevance between each of the influencing indicators and the target computing task according to the task description information; and obtain the indicator weight of each of the influencing indicators according to the relevance corresponding to each of the influencing indicators.

[0096] Each module in the above communication mode determination device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0097] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 6As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store communication mode determination related data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a communication mode determination method is implemented.

[0098] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0099] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0100] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0101] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0102] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0103] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0104] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for determining a communication mode, characterized in that: The method comprises: Get the task description information of the target computing task; According to the task description information and a plurality of influencing indicators of the target computing task, obtaining an indicator weight of each of the influencing indicators; Using a pre-built communication mode evaluation model, based on the task description information, obtaining an indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task; Obtaining a comprehensive score for each of the communication methods according to the indicator score and the indicator weight of each of the influencing indicators; According to the comprehensive score, a target communication mode for the target computing task is determined among the multiple communication modes.

2. The method according to claim 1, characterized in that The using of the pre-built communication mode evaluation model to obtain the indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators based on the task description information includes: Obtaining transmission description information of each of the communication modes; Inputting the task description information and the transmission description information into the pre-built communication mode evaluation model to obtain the indicator score of each communication mode under each influencing indicator; The communication mode evaluation model calculates the indicator score of each communication mode under each influencing indicator according to the task description information and the transmission description information.

3. The method according to claim 2, characterized in that The step of inputting the task description information and the transmission description information into the pre-built communication mode evaluation model to obtain the indicator score of each communication mode under each influencing indicator includes: Inputting the task description information and the transmission description information into the pre-built communication mode evaluation model; The feature module in the communication mode evaluation model respectively extracts features from the task description information and the transmission description information, determines and outputs the task feature information of the target computing task, and the transmission feature information of each of the communication modes; The indicator score of each of the communication modes under each of the influencing indicators is obtained according to the task characteristic information and the transmission characteristic information.

4. The method according to claim 1, characterized in that The obtaining of a comprehensive score of each of the communication modes according to the indicator score and the indicator weight of each of the influencing indicators includes: The product of the indicator score and the indicator weight is determined as the weighted indicator score of each communication mode under each influencing indicator; The comprehensive score of each communication method is obtained according to the sum of the weighted index scores of each communication method under the multiple influencing indicators.

5. The method according to claim 1, characterized in that Determining a target communication mode for the target computing task from among the multiple communication modes according to the comprehensive score includes: The communication method with the highest comprehensive score among the plurality of communication methods is determined as the target communication method.

6. The method according to any one of claims 1 to 5, characterized in that: The step of obtaining the indicator weight of each influencing indicator according to the task description information and the multiple influencing indicators of the target computing task includes: Determining the relevance of each of the influencing indicators to the target computing task according to the task description information; The indicator weight of each of the influencing indicators is obtained according to the correlation corresponding to each of the influencing indicators.

7. A communication mode determination device, characterized in that: The device comprises: A data acquisition module is used to obtain task description information of a target computing task; A weight calculation module, used to obtain the indicator weight of each influencing indicator according to the task description information and multiple influencing indicators of the target computing task; A first scoring calculation module is used to obtain an indicator score of each of the multiple communication modes of the target computing task under each of the influencing indicators by using a pre-built communication mode evaluation model based on the task description information; the multiple communication modes are used to realize communication between multiple graphics processors; and the multiple graphics processors are used to execute the target computing task; A second score calculation module, used for obtaining a comprehensive score of each of the communication modes according to the indicator score and the indicator weight of each of the influencing indicators; A screening module is used to determine a target communication mode for the target computing task from among the multiple communication modes according to the comprehensive score.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images