Data processing method and system, computing device and storage medium

Abstract

The embodiment of the invention relates to the technical field of computers, in particular to a data processing method and system, computing equipment and a storage medium, and the data processing method comprises the following steps: determining a data distribution network contained in a task processing model and a plurality of task processing networks corresponding to the data distribution network; obtaining distribution behavior data of task data distribution to a plurality of task processing networks by the data distribution network; the data volume of the distribution behavior data is analyzed, an analysis result is obtained, the task processing model is optimized according to the analysis result, and the analysis result is used for judging the balance of the data distribution behaviors of the data distribution network. The data processing conditions of the task processing networks in the task processing model can be known in time, and the distribution conditions of the data distribution networks in the task processing model can also be known in time.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and particularly to data processing methods and systems, computing devices, and storage media. Background Art

[0002] A mixture-of-experts model is a deep learning architecture that can improve the efficiency and performance of a model by combining multiple expert models (i.e., sub-models specialized in handling specific tasks or data types) with a gating network (responsible for allocating input data to appropriate expert models).

[0003] However, since the mixture-of-experts model includes multiple expert models, in the case where some expert models are overused while others are rarely used, it will lead to waste of model computing resources and performance degradation. Moreover, during the model training process, if multiple expert models learn similar knowledge, it will also affect the efficiency and performance of the model, resulting in a load balancing problem for the mixture-of-experts model. For current mixture-of-experts models, due to the difficulty in timely understanding the data processing situations of multiple expert models, the above load balancing problem is difficult to solve in a timely manner, further affecting the task processing performance and efficiency of the mixture-of-experts model. Therefore, an effective technical solution is urgently needed to solve the above problems. Summary of the Invention

[0004] In view of this, the embodiments of this specification provide two data processing methods. One or more embodiments of this specification simultaneously relate to a data processing device, a data processing system, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0005] According to the first aspect of the embodiments of this specification, a data processing method is provided, including: Determine a data distribution network included in a task processing model and multiple task processing networks corresponding to the data distribution network, where the data distribution network is used to distribute task data of a target task to the multiple task processing networks when the task processing model executes the target task, and the multiple task processing networks are used to process the task data distributed by the data distribution network; Obtain distribution behavior data of the data distribution network distributing task data to the multiple task processing networks; Analyze the data volume of the distribution behavior data to obtain an analysis result, and optimize the task processing model according to the analysis result, where the analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0006] According to the second aspect of the embodiments of this specification, a data processing device is provided, including: A determination module, configured to determine a data distribution network included in a task processing model and a plurality of task processing networks corresponding to the data distribution network, where the data distribution network is used to distribute task data of a target task to the plurality of task processing networks when the task processing model executes the target task, and the plurality of task processing networks are used to process the task data distributed by the data distribution network; An acquisition module, configured to acquire distribution behavior data of the data distribution network for distributing task data to the plurality of task processing networks; An analysis module, configured to analyze the data volume of the distribution behavior data to obtain an analysis result, and optimize the task processing model according to the analysis result, where the analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0007] According to the third aspect of the embodiments of this specification, a data processing method is provided, including: A client determines a data distribution network included in a task processing model and a plurality of task processing networks corresponding to the data distribution network, acquires distribution behavior data of the data distribution network for distributing task data to the plurality of task processing networks, and sends the distribution behavior data to a server, where the data distribution network is used to distribute task data of a target task to the plurality of task processing networks when the task processing model executes the target task, and the plurality of task processing networks are used to process the task data distributed by the data distribution network; The server analyzes the data volume of the distribution behavior data to obtain an analysis result, and optimizes the task processing model according to the analysis result, where the analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0008] According to the fourth aspect of the embodiments of this specification, a data processing system is provided, including a client and a server, where The client is configured to determine a data distribution network included in a task processing model and a plurality of task processing networks corresponding to the data distribution network, acquire distribution behavior data of the data distribution network for distributing task data to the plurality of task processing networks, and send the distribution behavior data to the server, where the data distribution network is used to distribute task data of a target task to the plurality of task processing networks when the task processing model executes the target task, and the plurality of task processing networks are used to process the task data distributed by the data distribution network; The server is configured to analyze the data volume of the distribution behavior data, obtain an analysis result, and optimize the task processing model according to the analysis result, where the analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0009] According to a fifth aspect of the embodiments of the present specification, a computing device is provided, including: A memory and a processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the above method are implemented.

[0010] According to a sixth aspect of the embodiments of the present specification, a computer-readable storage medium is provided, which stores computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the above method are implemented.

[0011] According to a seventh aspect of the embodiments of the present specification, a computer program product is provided, including computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the above method are implemented.

[0012] An embodiment of the present specification provides a data processing method, including: determining a data distribution network included in a task processing model and a plurality of task processing networks corresponding to the data distribution network, where the data distribution network is used to distribute task data of a target task to the plurality of task processing networks when the task processing model executes the target task, and the plurality of task processing networks are used to process the task data distributed by the data distribution network; obtaining distribution behavior data of the data distribution network for distributing task data to the plurality of task processing networks; analyzing the data volume of the distribution behavior data to obtain an analysis result, and optimizing the task processing model according to the analysis result, where the analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0013] The above method collects distribution behavior data of a data distribution network for distributing task data to a plurality of task processing networks, analyzes the data volume of the distribution behavior data to obtain an analysis result, realizes the statistical analysis of the distribution behavior data of the data distribution network in the task processing model for distributing task data to the plurality of task processing networks, realizes the judgment of the balance of the data distribution behavior of the data distribution network, realizes timely understanding of the data processing conditions of the plurality of task processing networks in the task processing model, and can also timely understand the distribution conditions of the data distribution network in the task processing model, so as to facilitate optimizing the task processing model according to the analysis result and further ensuring the task processing performance and efficiency of the task processing model. Brief Description of the Drawings

[0014] Figure 1 is a schematic diagram of an application scenario of a data processing method provided by an embodiment of this specification; Figure 2 is a flowchart of a data processing method provided by an embodiment of this specification; Figure 3 is a schematic diagram of a visualization chart in a data processing method provided by an embodiment of this specification; Figure 4 is a schematic diagram of another visualization chart in a data processing method provided by an embodiment of this specification; Figure 5 is a schematic diagram of the structure of a task processing model in a data processing method provided by an embodiment of this specification; Figure 6 is a flowchart of the processing process of a data processing method provided by an embodiment of this specification; Figure 7 is a schematic diagram of the structure of a data processing device provided by an embodiment of this specification; Figure 8 is a flowchart of another data processing method provided by an embodiment of this specification; Figure 9 is a schematic diagram of the structure of a data processing system provided by an embodiment of this specification; Figure 10 is a structural block diagram of a computing device provided by an embodiment of this specification. Detailed Description of the Embodiments

[0015] Many specific details are set forth in the following description in order to provide a thorough understanding of this specification. However, this specification can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the connotation of this specification. Therefore, this specification is not limited by the specific embodiments disclosed below.

[0016] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a", "the", and "said" used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more of the associated listed items.

[0017] It should be understood that although the terms first, second, etc. may be used in one or more embodiments of this specification to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to determining".

[0018] In addition, 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 for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data that have been authorized by the user or fully authorized by all parties. And the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0019] In one or more embodiments of this specification, a large model refers to a deep learning model with a large number of model parameters, usually including hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than one hundred trillion model parameters. A large model can also be referred to as a Foundation Model. Through pre-training of the large model with a large amount of unlabeled corpus, a pre-trained model with more than one billion parameters is produced. This model can adapt to a wide range of downstream tasks and has good generalization ability, such as large language models (LLMs), multi-modal pre-training models, etc.

[0020] When a large model is actually applied, only a small number of samples are needed to fine-tune the pre-trained model for application to different tasks. Large models can be widely applied in the fields of natural language processing (NLP), computer vision, etc. Specifically, they can be applied to tasks in the field of computer vision such as visual question answering (VQA), image captioning (IC), image generation, etc., and tasks in the field of natural language processing such as text-based sentiment classification, text summary generation, machine translation, etc. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0021] First, the noun terms involved in one or more embodiments of this specification are explained.

[0022] In practical applications, training a mixture of experts model is difficult. In particular, there is a problem that the indicators in the pre-training stage are good, but after supervised fine-tuning, the performance in downstream scenarios is not good. Therefore, an effective technical solution is urgently needed to solve the above problems.

[0023] In this specification, two data processing methods are provided. This specification also relates to a data processing device, a data processing system, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail one by one in the following embodiments.

[0024] See Figure 1 , Figure 1 which shows a schematic diagram of the application scenario of a data processing method provided according to an embodiment of this specification.

[0025] Figure 1 It includes an edge device 102 and a cloud device 104.

[0026] Specifically, a task processing model can be deployed in the edge device 102. The task processing model includes a data distribution network and multiple task processing networks corresponding to the data distribution network. The data distribution network is used to distribute the task data of the target task to the multiple task processing networks when the task processing model executes the target task. The multiple task processing networks are used to process the task data distributed by the data distribution network. The edge device 102 can be used as a distribution behavior analysis collector, and can obtain the distribution behavior data of the task data distributed by the data distribution network to the multiple task processing networks through a data embedding program, and send the distribution behavior data to the cloud device 104. The cloud device 104, as a distribution behavior analysis server, can store the distribution behavior data in a database, display the distribution behavior data in multiple dimensions (that is, analyze the distribution behavior data to obtain an analysis result), generate a visualization interface, and optimize the task processing model according to the analysis result to achieve the training white-boxing of the task processing model.

[0027] The edge device 102 may include a browser, an APP (Application), or a web application such as an H5 (Hyper Text Markup Language 5) application, or a light application (also known as a mini-program, a lightweight application), or a cloud application, etc. The edge device may be developed based on the software development kit (SDK) of the corresponding service provided by the server side, such as developed based on the real-time communication (RTC) SDK. The edge device may be deployed in an electronic device and needs to rely on the device or certain APPs in the device to run. The electronic device may have a display screen and support information browsing, etc., such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications can usually be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0028] The cloud device 104 can be understood as a server that provides various services, including physical servers and cloud servers. For example, a server that provides communication services for multiple clients, or a server for background training that supports the models used on the client, or a server that processes the data sent by the client, etc. It should be noted that the cloud device 104 can be implemented as a distributed server cluster composed of multiple servers, or can be implemented as a single server. The cloud device 104 can also be a server of a distributed system, or a server combined with a blockchain. The cloud device 104 can also be a cloud server of basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery network (CDN), and big data and artificial intelligence platforms, or an intelligent cloud computing server or an intelligent cloud host with artificial intelligence technology.

[0029] It is worth noting that the data processing method provided in the embodiments of this specification can be executed by the cloud device 104. In other embodiments of this specification, the edge device 102 may also have similar functions to the cloud device 104, so as to execute the data processing method provided in the embodiments of this specification; in other embodiments, the data processing method provided in the embodiments of this specification may also be jointly executed by the edge device 102 and the cloud device 104.

[0030] See Figure 2 , Figure 2The flowchart of a data processing method provided according to an embodiment of this specification is shown, which specifically includes the following steps.

[0031] Step 202: Determine the data distribution network included in the task processing model and the multiple task processing networks corresponding to the data distribution network. Among them, the data distribution network is used to distribute the task data of the target task to the multiple task processing networks when the task processing model executes the target task, and the multiple task processing networks are used to process the task data distributed by the data distribution network.

[0032] Specifically, the data processing method provided in the embodiments of this specification can be applied to the training stage of the task processing model or the application stage of the task processing model.

[0033] Among them, the task processing model can be understood as a mixture-of-experts model, the data distribution network can be understood as the gating network in the mixture-of-experts model, and the task processing network can be understood as the expert models included in the mixture-of-experts model. In the training stage of the task processing model, the target task can be the model training task. Then, the task data of the target task can be the training data. In the application stage of the task processing model, the target task can be the data processing task in model application. Then, the task data of the target task is the data to be processed. For example, if the target task is an image-to-text task, then the task data of the target task can be image data or text data.

[0034] Based on this, the gating network included in the mixture-of-experts model and the multiple expert models corresponding to the gating network can be determined. The gating network can, according to a preset distribution rule, distribute the task data of the target task to the multiple expert models when the mixture-of-experts model executes the target task. Among them, the preset distribution rule can be to distribute according to the data type of the task data. For example, send the task data of the text type to expert model 1, send the task data of the character type to expert model 2, send the task data of the digital type to expert model 3, etc. The embodiments of this specification do not limit this.

[0035] For example, the data distribution network A included in the task processing model and the task processing networks B1, B2, and B3 corresponding to the data distribution network A can be determined. The data distribution network A can distribute the task data of the target task to the task processing networks B1, B2, and B3, and the task processing networks B1, B2, and B3 process the distributed task data.

[0036] In one embodiment of this specification, the task data of the target task includes at least any one of text data, visual data, and audio data.

[0037] Among them, the visual data may include image data, video data, etc.

[0038] Specifically, the task processing model can be applicable to a variety of target tasks, which may include but are not limited to text processing tasks, image processing tasks, video processing tasks, audio processing tasks, etc. Then, the task data of the target tasks may include but are not limited to text data, image data, video data, and audio data, etc. The embodiments of this specification do not make limitations in this regard.

[0039] Step 204: Obtain the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks.

[0040] Among them, the distribution behavior data may include the data volume and data type of the task data distributed to each task processing network when the data distribution network distributes task data to the multiple task processing networks. For example, if the data distribution network distributes 500 text-type data to task processing network 1, then the distribution behavior data may include a data volume of 500 and a data type of text type.

[0041] Specifically, each time the task processing model executes a target task, the data volume and data type of the task data distributed to each task processing network can be obtained when the data distribution network distributes task data to the multiple task processing networks.

[0042] In practical applications, the data distribution network may be a router. Then, the distribution behavior data of the data distribution network is the routing behavior data, and the task data distributed by the data distribution network to the task processing network may be Tokens.

[0043] Continuing with the above example, it can be obtained that data distribution network A distributes 100 task data to task processing network B1, 150 task data to task processing network B2, and 90 task data to task processing network B3.

[0044] In specific implementation, the obtaining of the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks includes: Using a data logging program to obtain the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks.

[0045] Among them, the data logging program can be used to collect the distribution behavior data of the data distribution network.

[0046] Specifically, the task processing model can be deployed in the client. Then, the data logging program of the client can be used to collect the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks.

[0047] In practical applications, in the training program deployed on the client side, a data collection program can be used to collect distribution behavior data.

[0048] In summary, by using the data collection program, the collection of distribution behavior data of the data distribution network is realized, which is convenient for subsequent analysis of the distribution behavior data.

[0049] Step 206: Analyze the data volume of the distribution behavior data to obtain an analysis result, and optimize the task processing model according to the analysis result, where the analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0050] Furthermore, after obtaining the analysis result, a visualization interface of the analysis result can also be generated.

[0051] Specifically, after obtaining the distribution behavior data, the distribution behavior data can be analyzed to obtain an analysis result, and a visualization interface of the analysis result can be generated.

[0052] In an embodiment of this specification, when the distribution behavior data is the data volume of the task data distributed by the data distribution network to multiple task processing networks, the data volume can be analyzed to determine whether the task data distributed by the data distribution network to each task processing network is balanced, so as to judge whether the data distribution behavior of the data distribution network is balanced.

[0053] In another embodiment of this specification, when the distribution behavior data is the data type of the task data distributed by the data distribution network to multiple task processing networks, it can be determined whether the data distribution network distributes the task data according to a preset distribution rule. For example, the preset distribution rule is to distribute task data of text type to task processing network 1 and task data of digital type to task processing network 2. Then, when the data distribution network distributes task data to multiple task processing networks, the data type of the task data distributed to each task processing network can be obtained, and it can be determined whether the data distribution network distributes the task data according to the preset distribution rule. If not, the data distribution network can be adjusted. Based on this, both the data volume of the task data distributed by the data distribution network to each task processing network and the data type of the task data distributed to which task processing network can be analyzed, realizing the display of a more rich visualization interface, which is convenient for the subsequent determined optimization strategy to be more comprehensive and accurate.

[0054] Optionally, the distribution behavior data includes the data volume of the task data distributed by the data distribution network to the multiple task processing networks; Analyzing the data volume of the distribution behavior data to obtain an analysis result, including: Statistical analysis is performed on the data volume of the task data distributed by the data distribution network to the multiple task processing networks to obtain a statistical result; According to the statistical result, a visualization chart is generated, and the visualization chart is determined as the analysis result. Wherein, the visualization chart is used to represent the comparison result of the data volume of the task data distributed by the data distribution network to each task processing network. The visualization chart contains multiple network markers, and the multiple network markers correspond to the multiple task processing networks one by one.

[0055] Among them, the network marker can be understood as the network marker corresponding to the task processing network. The network marker can be, for example, a color marker. Each color corresponds to a task processing network. For example, the color corresponding to task processing network 1 is purple, and the color corresponding to task processing network 2 is red. In the visualization chart, if the data distribution network distributes 1 task data to task processing network 1, then it is represented by purple in the visualization chart.

[0056] For practical applications, please refer to Figure 3 , Figure 3 shows a schematic diagram of a visualization chart in a data processing method provided according to an embodiment of the present specification. As Figure 3 shown, there can be multiple training rounds for the task processing model. In each round of training of the task processing model, according to the foregoing steps, the distribution behavior data of the task data distributed by the data distribution network to the multiple task processing networks is obtained, and the distribution behavior data is analyzed to obtain an analysis result, and a visualization interface of the analysis result is generated. Figure 3 In (a) of Figure 3 shows a schematic diagram of task data being distributed to different task processing networks during the first round of iterative training of the task processing model (i.e., the initial state). Among them, different colors represent different task processing networks.

[0057] Specifically, in an embodiment of the present specification, statistical analysis can be performed on the data volume of the task data distributed by the data distribution network to the multiple task processing networks to obtain a statistical result. According to the statistical result, a visualization chart is generated, and the visualization chart is determined as the analysis result. Subsequently, an optimization strategy can be generated based on the analysis result.

[0058] In another embodiment of this specification, it is also possible to compare the visualization chart of the foregoing initial state with the visualization chart of the final state to obtain a comparison result, and subsequently, an optimization strategy can be generated based on this comparison result.

[0059] In summary, by generating a visualization chart, the visualization analysis of the data distribution network is realized, and the understanding of the data distribution behavior of the data distribution network of the task processing model is improved.

[0060] During specific implementation, after obtaining the analysis result, it further includes: Generating a visualization interface including the visualization chart according to the analysis result; Determining the data distribution state of the data distribution network according to the distribution behavior data of the task data distributed by the data distribution network to the multiple task processing networks shown in the visualization interface; Generating an optimization strategy for the data distribution network according to the data distribution state; The optimization of the task processing model according to the analysis result includes: Optimizing the data distribution network in the task processing model according to the optimization strategy to obtain an optimized data distribution network.

[0061] Among them, the optimization strategy can be understood as an optimization strategy for the data distribution network. The optimization strategy can be to adjust the data distribution rules of the data distribution network or to adjust the amount of task data distributed by the data distribution network to the task processing model, etc. The data distribution state can be understood as the state of whether the task data distributed by the data distribution network to multiple task processing networks is balanced. For example, if the data distribution network distributes 100 task data to each of the multiple task processing networks, then the data distribution state can be a balanced state. Or, if the data distribution network sends 200 task data to task processing network 1, 1000 task data to task processing network 2, and 5000 task data to task processing network 3, that is to say, the amount of task data distributed by the data distribution network to multiple task processing networks varies greatly, then the data distribution state can be an unbalanced state. Then, when the data distribution state is a balanced state, the data distribution network does not need to be optimized. When the data distribution state is an unbalanced state, the optimization strategy can be to adjust the distribution behavior of the data distribution network.

[0062] Specifically, according to the analysis result, a visualization interface including the visualization chart can be generated. Subsequently, the visualization interface can be sent to the client and presented to the algorithm researchers through the display interface of the client, so as to white-box the training process of the task processing model. According to the distribution behavior data of the task data distributed by the data distribution network to each task processing network shown in the visualization interface, it can be determined whether the data distribution state of the data distribution network is balanced. And according to this data distribution state, an optimization strategy for the data distribution network can be generated. According to this optimization strategy, the data distribution network in the task processing model is optimized to obtain an optimized data distribution network, thereby achieving further optimization of the task processing model.

[0063] See Figure 4 , Figure 4 FIG. shows a schematic diagram of another visualization chart in a data processing method provided according to an embodiment of the present specification. As Figure 4 shown, the abscissa represents the number of iterations in the training process of the task processing model, and the ordinate represents the amount of task data assigned to each task processing network in each iteration during the training process of the task processing model. Different color markers represent different task processing networks. In the case where the amount of task data assigned to some task processing networks is particularly large, this indicates that there may be a problem of data skew, that is, some types of input data tend to be assigned to specific task processing networks, while most of the curves are relatively flat, which means that in most cases, the amount of task data processed by each task processing model is relatively stable, that is, the data distribution network can distribute tasks to each task processing network more evenly. However, some curves have sharp fluctuations when the abscissa is from 400 to 450, which indicates that there are sharp fluctuations in the amount of task data assigned to some task processing networks during the 400th and 500th iterations in the training process of the task processing model.

[0064] In summary, by generating an optimization strategy for the data distribution network, the optimization of the task processing model can be achieved.

[0065] Further, after obtaining the optimized data distribution network, it further includes: Obtaining the distribution behavior data of the optimized data distribution network for distributing task data to the multiple task processing networks; Continuing to execute the step of analyzing the distribution behavior data to obtain an analysis result, generating the next visualization interface of the analysis result, and generating the next optimization strategy for the task processing model according to the next visualization interface until a task processing model that meets the optimization stop condition is obtained.

[0066] Among them, the optimization stop condition can be that the number of optimization rounds reaches a preset round threshold or the distribution behavior of the data distribution network is stable and uniform.

[0067] Specifically, after obtaining the optimized data distribution network, the above steps can be continued to obtain the distribution behavior data of the optimized data distribution network for task data distribution to the multiple task processing networks, analyze the distribution behavior data to obtain an analysis result, generate the next visualization interface of the analysis result, and continue to generate the next optimization strategy for the task processing model according to the next visualization interface until a task processing model that meets the optimization stop condition is obtained.

[0068] In summary, through the above iterative process, the analysis and optimization of the task processing model can be realized.

[0069] Specifically, generating the next optimization strategy for the task processing model according to the next visualization interface includes: Comparing the distribution behavior data of the optimized data distribution network for task data distribution to the multiple task processing networks shown in the next visualization interface with the distribution behavior data of the data distribution network for task data distribution to the multiple task processing networks shown in the visualization interface to obtain a comparison result; Generating the next optimization strategy for the optimized data distribution network according to the comparison result.

[0070] Among them, the next visualization interface can be understood as the visualization interface generated by the analysis result of the next iteration process of the task processing model.

[0071] Specifically, the next optimization strategy for the optimized data distribution network can be generated according to the comparison result between the distribution behavior data in two iterative processes.

[0072] In summary, through multiple iterative optimizations, the further optimization of the task processing model is realized, and the performance and efficiency of the task processing model are further guaranteed.

[0073] Further, the task processing model includes multiple data distribution networks, and multiple task processing networks corresponding to each data distribution network in the multiple data distribution networks; The obtaining the distribution behavior data of the data distribution network for task data distribution to the multiple task processing networks includes: Obtaining the distribution behavior data of each data distribution network for task data distribution to the corresponding multiple task processing networks; The analyzing the data volume of the distribution behavior data to obtain an analysis result includes: Analyze the data volume of the distribution behavior data corresponding to each data distribution network to obtain the analysis results of each data distribution network; After obtaining the analysis results, it further includes: Generate a visualization interface corresponding to each data distribution network.

[0074] Specifically, the task processing model may include multiple layers of networks, and each layer of network includes a data distribution network and multiple task processing networks corresponding to the data distribution network. Combining the above Figure 4 , Figure 4 (a) in shows the visualization interface corresponding to the first layer of network in the task processing model, Figure 4 (b) in shows the visualization interface corresponding to the second layer of network in the task processing model.

[0075] Based on this, the distribution behavior data of each layer of network in the task processing model can be analyzed and the visualization interface can be generated.

[0076] In practical applications, referring to Figure 5 , Figure 5 shows the structural schematic diagram of the task processing model in a data processing method provided in an embodiment of this specification. As Figure 5 shown, the task processing model may include multiple layers of networks. Taking one layer of network as an example, this layer of network includes a data distribution network and multiple task processing networks. The task data is input into the data distribution network, and the data distribution network distributes the input task data to multiple task processing networks, and the multiple task processing networks are used to process the task data to obtain an output result, and this output result can continue to be input into the next layer of network.

[0077] In summary, by analyzing and visualizing each layer of network in the task processing model, the comprehensive optimization of the task processing model is realized.

[0078] In addition, after generating the visualization interface corresponding to each data distribution network, it further includes: Adjust the second data distribution network according to the visualization interface corresponding to the first data distribution network; Wherein, the first data distribution network is any one of the multiple data distribution networks, and the second data distribution network is the next data distribution network of the first data distribution network among the multiple data distribution networks.

[0079] Specifically, the first data distribution network can be understood as the data distribution network in a certain layer of the task processing model, and the second data distribution network can be understood as the data distribution network in the next layer of the network of the first data distribution network. Then, according to the visualization interface corresponding to the data distribution network in a certain layer of the network, the data distribution network in the next layer of this layer can be adjusted to optimize the task processing model.

[0080] In summary, the above method collects the distribution behavior data of the data distribution network for task data distribution to multiple task processing networks, analyzes the data volume of the distribution behavior data, obtains the analysis result, realizes the statistical analysis of the distribution behavior data of the data distribution network in the task processing model for task data distribution to multiple task processing networks, realizes the judgment of the balance of the data distribution behavior of the data distribution network, realizes timely understanding of the data processing situation of multiple task processing networks in the task processing model, and can also timely understand the distribution situation of the data distribution network in the task processing model, so as to facilitate optimizing the task processing model according to the analysis result and further ensuring the task processing performance and efficiency of the task processing model.

[0081] The following combines the attached Figure 6 , taking the application of the data processing method provided in this specification in the mixture of experts model as an example, to further illustrate the data processing method. Among them, Figure 6 shows the processing procedure flowchart of a data processing method provided by an embodiment of this specification, which specifically includes the following steps.

[0082] Step 602: Determine the data distribution network included in the task processing model and the multiple task processing networks corresponding to the data distribution network.

[0083] Step 604: Obtain the distribution behavior data of the data distribution network for task data distribution to the multiple task processing networks during each iteration training process.

[0084] Step 606: Statistically analyze the data volume of the task data distributed by the data distribution network to the multiple task processing networks to obtain the statistical result.

[0085] Step 608: Generate a visualization chart according to the statistical result, and determine the visualization chart as the analysis result.

[0086] Step 610: Generate a visualization interface including the visualization chart according to the analysis result.

[0087] Step 612: Determine the optimization strategy for the data distribution network according to the visualization interface.

[0088] Step 614: Optimize the data distribution network according to the optimization strategy to obtain an optimized data distribution network.

[0089] In summary, the above method collects the distribution behavior data of the data distribution network for distributing task data to multiple task processing networks, analyzes the distribution behavior data to obtain an analysis result, realizes the statistical analysis of the distribution of task data from the data distribution network to multiple task processing networks in the task processing model, and generates a visualization interface of the analysis result, enabling the analysis result to be displayed and analyzed, so as to timely understand the data processing situation of multiple task processing networks in the task processing model and also timely understand the distribution situation of the data distribution network in the task processing model, thereby facilitating the determination of the optimization strategy based on this visualization interface and facilitating the subsequent optimization of the task processing model according to the optimization strategy, further ensuring the task processing performance and efficiency of the task processing model.

[0090] Corresponding to the above method embodiment, this specification also provides a data processing device embodiment. Figure 7 It shows a schematic structural diagram of a data processing device provided by an embodiment of this specification. As Figure 7 shown, the device includes: A determination module 702, configured to determine the data distribution network included in the task processing model and multiple task processing networks corresponding to the data distribution network, where the data distribution network is used to distribute the task data of the target task to the multiple task processing networks when the task processing model executes the target task, and the multiple task processing networks are used to process the task data distributed by the data distribution network; An acquisition module 704, configured to acquire the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks; An analysis module 706, configured to analyze the data volume of the distribution behavior data to obtain an analysis result, and optimize the task processing model according to the analysis result, where the analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0091] In an optional embodiment, the distribution behavior data includes the data volume of the task data distributed by the data distribution network to the multiple task processing networks; The analysis module 706 is further configured to: Statistically analyze the data volume of the task data distributed by the data distribution network to the multiple task processing networks to obtain a statistical result; Generate a visualization chart based on the statistical results, and determine the visualization chart as the analysis result, where the visualization chart is used to represent the comparison result of the amounts of task data distributed by the data distribution network to each task processing network, and the visualization chart includes a plurality of network markers, and the plurality of network markers correspond to the plurality of task processing networks one by one.

[0092] In an optional embodiment, the analysis module 706 is further configured to: Generate a visualization interface including the visualization chart according to the analysis result; Determine the data distribution status of the data distribution network according to the distribution behavior data of the task data distributed by the data distribution network to the plurality of task processing networks shown in the visualization interface; Generate an optimization strategy for the data distribution network according to the data distribution status; Optimize the data distribution network in the task processing model according to the optimization strategy to obtain an optimized data distribution network.

[0093] In an optional embodiment, the analysis module 706 is further configured to: Obtain the distribution behavior data of the optimized data distribution network for distributing task data to the plurality of task processing networks; Continue to execute the step of analyzing the distribution behavior data to obtain an analysis result, generating the next visualization interface of the analysis result, and generating the next optimization strategy for the task processing model according to the next visualization interface until a task processing model that meets the optimization stop condition is obtained.

[0094] In an optional embodiment, the analysis module 706 is further configured to: Compare the distribution behavior data of the optimized data distribution network for distributing task data to the plurality of task processing networks shown in the next visualization interface with the distribution behavior data of the data distribution network for distributing task data to the plurality of task processing networks shown in the visualization interface to obtain a comparison result; Generate the next optimization strategy for the optimized data distribution network according to the comparison result.

[0095] In an optional embodiment, the task processing model includes a plurality of data distribution networks, and a plurality of task processing networks corresponding to each data distribution network in the plurality of data distribution networks; The obtaining module 704 is further configured to: Obtain the distribution behavior data of each of the data distribution networks for distributing task data to the corresponding multiple task processing networks; The analysis module 706 is further configured to: Analyze the data volume of the distribution behavior data corresponding to each of the data distribution networks, obtain the analysis results of each of the data distribution networks, and generate a visualization interface corresponding to each of the data distribution networks.

[0096] In an alternative embodiment, the analysis module 706 is further configured to: Adjust the second data distribution network according to the visualization interface corresponding to the first data distribution network; Wherein, the first data distribution network is any one of the multiple data distribution networks, and the second data distribution network is the next data distribution network of the first data distribution network among the multiple data distribution networks.

[0097] In an alternative embodiment, the task data of the target task includes at least any one of text data, visual data, and audio data.

[0098] In an alternative embodiment, the acquisition module 704 is further configured to: Use a buried point program to obtain the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks.

[0099] In summary, the above device collects the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks, analyzes the data volume of the distribution behavior data, obtains the analysis results, realizes the statistical analysis of the distribution behavior data of the data distribution network in the task processing model for distributing task data to the multiple task processing networks, realizes the judgment of the balance of the data distribution behavior of the data distribution network, realizes timely understanding of the data processing conditions of the multiple task processing networks in the task processing model, and can also timely understand the distribution conditions of the data distribution network in the task processing model, so as to facilitate optimizing the task processing model according to the analysis results and further ensure the task processing performance and efficiency of the task processing model.

[0100] The above is a schematic solution of a data processing device in this embodiment. It should be noted that the technical solution of this data processing device and the technical solution of the above data processing method belong to the same concept. For the details not described in the technical solution of the data processing device, reference can be made to the description of the technical solution of the above data processing method.

[0101] Corresponding to the above method embodiment, refer to Figure 8 , Figure 8The flowchart of another data processing method provided according to an embodiment of this specification is shown, which specifically includes the following steps.

[0102] Step 802: The client determines the data distribution network included in the task processing model and the multiple task processing networks corresponding to the data distribution network, obtains the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks, and sends the distribution behavior data to the server. Among them, the data distribution network is used to distribute the task data of the target task to the multiple task processing networks when the task processing model executes the target task, and the multiple task processing networks are used to process the task data distributed by the data distribution network. Step 804: The server analyzes the data volume of the distribution behavior data to obtain an analysis result, and optimizes the task processing model according to the analysis result. The analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0103] Specifically, the above method collects the distribution behavior data of the data distribution network for distributing task data to multiple task processing networks, analyzes the data volume of the distribution behavior data to obtain an analysis result, realizes the statistical analysis of the distribution behavior data of the data distribution network in the task processing model for distributing task data to multiple task processing networks, realizes the judgment of the balance of the data distribution behavior of the data distribution network, realizes timely understanding of the data processing situation of multiple task processing networks in the task processing model, and can also timely understand the distribution situation of the data distribution network in the task processing model, so as to facilitate optimizing the task processing model according to the analysis result and further ensuring the task processing performance and efficiency of the task processing model.

[0104] The above is a schematic solution of a data processing method in this embodiment. It should be noted that the technical solution of this data processing method belongs to the same concept as the technical solution of the above data processing method. For the details not described in the technical solution of the data processing method, reference can be made to the description of the technical solution of the above data processing method.

[0105] Corresponding to the above method embodiment, this specification also provides a data processing system embodiment. Figure 9 The structural schematic diagram of a data processing system provided according to an embodiment of this specification is shown. As Figure 9 shown, the system includes a client 902 and a server 904, where The client 902 is configured to determine the data distribution network included in the task processing model and the multiple task processing networks corresponding to the data distribution network, obtain the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks, and send the distribution behavior data to the server 904. Among them, the data distribution network is used to distribute the task data of the target task to the multiple task processing networks when the task processing model executes the target task, and the multiple task processing networks are used to process the task data distributed by the data distribution network; The server 904 is configured to analyze the data volume of the distribution behavior data to obtain an analysis result, and optimize the task processing model according to the analysis result. Among them, the analysis result is used to judge the balance of the data distribution behavior of the data distribution network.

[0106] Specifically, the above system collects the distribution behavior data of the data distribution network for distributing task data to the multiple task processing networks, analyzes the data volume of the distribution behavior data to obtain an analysis result, realizes the statistical analysis of the distribution behavior data of the data distribution network in the task processing model for distributing task data to the multiple task processing networks, realizes the judgment of the balance of the data distribution behavior of the data distribution network, realizes the timely understanding of the data processing situation of the multiple task processing networks in the task processing model, and can also timely understand the distribution situation of the data distribution network in the task processing model, so as to facilitate optimizing the task processing model according to the analysis result and further ensuring the task processing performance and efficiency of the task processing model.

[0107] The above is a schematic solution of a data processing system according to an embodiment of the present specification. It should be noted that the technical solution of this data processing system and the technical solution of the above data processing method belong to the same concept. For the details not described in the technical solution of the data processing system, reference can be made to the description of the technical solution of the above data processing method.

[0108] Figure 10 The structural block diagram of a computing device 1000 provided according to an embodiment of this specification is shown. The components of the computing device 1000 include but are not limited to a memory 1010 and a processor 1020. The processor 1020 is connected to the memory 1010 through a bus 1030, and a database 1050 is used to store data.

[0109] The computing device 1000 further includes an access device 1040, which enables the computing device 1000 to communicate via one or more networks 1060. Examples of such networks include the Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 1040 may include one or more of any type of wired or wireless network interfaces (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0110] In one embodiment of the present application, the above components of the computing device 1000 and Figure 10 other components not shown may also be connected to each other, for example, via a bus. It should be understood that Figure 10 the block diagram of the computing device shown is for illustrative purposes only and is not a limitation on the scope of the present application. Those skilled in the art may add or replace other components as needed.

[0111] The computing device 1000 can be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.) or other types of mobile devices, or a stationary computing device such as a desktop computer or a Personal Computer (PC). The computing device 1000 can also be a mobile or stationary server.

[0112] Wherein, the processor 1020 is used to execute the following computer program / instructions, and when the computer program / instructions are executed by the processor, the steps of the above data processing method are implemented.

[0113] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the embodiment of the computing device, since it is basically similar to the embodiment of the data processing method, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the embodiment of the data processing method.

[0114] An embodiment of this specification also provides a computer-readable storage medium, which stores computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the above data processing method are implemented.

[0115] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the embodiment of the computer-readable storage medium, since it is basically similar to the embodiment of the data processing method, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the embodiment of the data processing method.

[0116] An embodiment of this specification also provides a computer program product, including computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the above data processing method are implemented.

[0117] The above is a schematic solution of a computer program product of this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the above data processing method belong to the same concept. For the details not described in detail in the technical solution of the computer program product, reference can be made to the description of the technical solution of the above data processing method.

[0118] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be executed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0119] The computer instructions include computer program code, which may be in the form of source code, object code, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, removable hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0120] It should be noted that for the foregoing method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the embodiments of this specification are not limited by the described action sequence, because according to the embodiments of this specification, some steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential for the embodiments of this specification.

[0121] In the above embodiments, the descriptions of the respective embodiments have their own focuses. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0122] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can well understand and utilize this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A data processing method, characterized in that: include: Determine a data distribution network included in a task processing model and a plurality of task processing networks corresponding to the data distribution network, wherein the data distribution network is used to distribute task data of the target task to the plurality of task processing networks when the task processing model executes the target task, and the plurality of task processing networks are used to process the task data distributed by the data distribution network; Acquire distribution behavior data of the data distribution network distributing task data to the plurality of task processing networks; The data volume of the distribution behavior data is analyzed to obtain analysis results, and the task processing model is optimized according to the analysis results, wherein the analysis results are used to determine the balance of the data distribution behavior of the data distribution network.

2. The method according to claim 1, characterized in that The distribution behavior data includes the data volume of the task data distributed by the data distribution network to the plurality of task processing networks; The step of analyzing the amount of the distribution behavior data to obtain analysis results includes: Counting the data volume of the task data distributed by the data distribution network to the plurality of task processing networks to obtain a statistical result; A visualization chart is generated based on the statistical results, and the visualization chart is determined as the analysis result, wherein the visualization chart is used to represent the comparison results between the data volumes of the task data distributed by the data distribution network to each task processing network, and the visualization chart includes multiple network tags, and the multiple network tags correspond one-to-one to the multiple task processing networks.

3. The method according to claim 2, characterized in that After obtaining the analysis results, the method further includes: According to the analysis result, generating a visualization interface including the visualization chart; Determine a data distribution state of the data distribution network according to distribution behavior data displayed in the visualization interface, in which the data distribution network distributes task data to the plurality of task processing networks; Generating an optimization strategy for the data distribution network according to the data distribution state; Optimizing the task processing model according to the analysis result includes: According to the optimization strategy, the data distribution network in the task processing model is optimized to obtain an optimized data distribution network.

4. The method according to claim 3, characterized in that After obtaining the optimized data distribution network, the method further includes: Acquire distribution behavior data of the optimized data distribution network distributing task data to the plurality of task processing networks; Continue to analyze the distribution behavior data, obtain analysis results, and generate a next visualization interface of the analysis results, and based on the next visualization interface, generate the next optimization strategy for the task processing model until a task processing model that meets the optimization stop condition is obtained.

5. The method according to claim 4, characterized in that Generating a next optimization strategy for the task processing model according to the next visualization interface includes: Comparing the distribution behavior data displayed in the next visualization interface, in which the optimized data distribution network distributes task data to the multiple task processing networks, with the distribution behavior data displayed in the visualization interface, in which the data distribution network distributes task data to the multiple task processing networks, to obtain a comparison result; According to the comparison result, a next optimization strategy for the optimized data distribution network is generated.

6. The method according to claim 1, characterized in that The task processing model includes a plurality of data distribution networks and a plurality of task processing networks corresponding to each of the plurality of data distribution networks; The obtaining of distribution behavior data of the data distribution network distributing task data to the plurality of task processing networks includes: Obtaining distribution behavior data of each data distribution network to distribute task data to the corresponding plurality of task processing networks; The step of analyzing the amount of the distribution behavior data to obtain analysis results includes: Analyzing the data volume of the distribution behavior data corresponding to each data distribution network to obtain analysis results of each data distribution network; After obtaining the analysis results, the method further includes: Generate a visualization interface corresponding to each data distribution network.

7. The method according to claim 6, characterized in that After generating the visualization interface corresponding to each data distribution network, the method further includes: Adjusting the second data distribution network according to the visualization interface corresponding to the first data distribution network; The first data distribution network is any one of the multiple data distribution networks, and the second data distribution network is the next data distribution network of the multiple data distribution networks after the first data distribution network.

8. A data processing method, characterized in that: include: The client determines a data distribution network included in a task processing model and a plurality of task processing networks corresponding to the data distribution network, obtains distribution behavior data of the data distribution network for distributing task data to the plurality of task processing networks, and sends the distribution behavior data to the server, wherein the data distribution network is used to distribute the task data of the target task to the plurality of task processing networks when the task processing model executes the target task, and the plurality of task processing networks are used to process the task data distributed by the data distribution network; The server analyzes the data volume of the distribution behavior data to obtain analysis results, and optimizes the task processing model according to the analysis results, wherein the analysis results are used to determine the balance of the data distribution behavior of the data distribution network.

9. A data processing system, characterized in that: It includes client and server, among which, The client is configured to determine a data distribution network included in a task processing model and a plurality of task processing networks corresponding to the data distribution network, obtain distribution behavior data of the data distribution network for distributing task data to the plurality of task processing networks, and send the distribution behavior data to the server, wherein the data distribution network is used to distribute the task data of the target task to the plurality of task processing networks when the task processing model executes the target task, and the plurality of task processing networks are used to process the task data distributed by the data distribution network; The server is configured to analyze the data volume of the distribution behavior data, obtain analysis results, and optimize the task processing model according to the analysis results, wherein the analysis results are used to determine the balance of the data distribution behavior of the data distribution network.

10. A computing device, characterized in that include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium, characterized in that: It stores a computer program / instruction, which implements the steps of the method described in any one of claims 1 to 8 when executed by a processor.

12. A computer program product, characterized in that The method comprises a computer program / instruction which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 8.

Images