Data processing method, device, electronic device and storage medium

Abstract

An embodiment of the present application provides a data processing method, device, electronic device and storage medium, the method comprising: dividing a data processing model to determine a first sub-model and a second sub-model, wherein the output data of the first sub-model is associated with the input data of the second sub-model; determining first configuration information of the first sub-model and second configuration information of the second sub-model as a configuration information combination; determining the data processing time corresponding to each configuration information combination, and screening out at least one group of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model; and improving the processing efficiency of the model.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method, a data processing device, an electronic device, and a storage medium. Background Art

[0002] Neural Networks (NN) models are complex network systems formed by extensively interconnecting a large number of simple processing units (also known as neurons, operators, computing units, or computing nodes). NN models feature large-scale parallelism, distributed storage and processing, self-organization, self-adaptation, and self-learning capabilities. They are particularly well-suited for handling imprecise and ambiguous information processing problems that require simultaneous consideration of many factors and conditions. For example, they can identify images and determine the target objects contained in them, as well as recognize audio, extract semantic information, and understand the audio content.

[0003] The existing neural network model has a complex structure. When processing data, data needs to be transmitted and converted between different nodes, resulting in low processing efficiency of the neural network model. Summary of the Invention

[0004] The embodiments of the present application provide a data processing method to improve the processing efficiency of a neural network model.

[0005] Correspondingly, an embodiment of the present application also provides a data processing device, an electronic device and a storage medium to ensure the implementation and application of the above system.

[0006] In order to solve the above problems, an embodiment of the present application discloses a data processing method, which includes: dividing the data processing model to determine a first sub-model and a second sub-model, and the output data of the first sub-model is associated with the input data of the second sub-model; determining the first configuration information of the first sub-model and the second configuration information of the second sub-model as a configuration information combination; determining the data processing time corresponding to each configuration information combination, and screening out at least one group of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model.

[0007] In order to solve the above problems, an embodiment of the present application discloses a data processing method, including: dividing an image processing model to determine a first sub-model and a second sub-model, the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit; obtaining an image configuration information table corresponding to the image processing model; based on the image configuration information table, determining the first configuration information of the first sub-model and the second configuration information of the second sub-model as a configuration information combination; determining the data processing time corresponding to each configuration information combination, and filtering out at least one group of target configuration information combinations as the configuration information filtering results of the first sub-model and the second sub-model; dividing and analyzing the first sub-model and the second sub-model as models to be processed until the configuration information filtering results of each processing unit are determined; and determining the model configuration information analysis results of the data processing model based on the configuration information filtering results of each processing unit in the image processing model.

[0008] In order to solve the above problems, an embodiment of the present application discloses a data processing method, including: dividing an audio processing model to determine a first sub-model and a second sub-model, the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit; obtaining an audio configuration information table corresponding to the audio processing model; determining the first configuration information of the first sub-model and the second configuration information of the second sub-model based on the audio configuration information table as a configuration information combination; determining the data processing time corresponding to each configuration information combination, and screening out at least one group of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model; dividing and analyzing the first sub-model and the second sub-model as models to be processed until the configuration information screening results of each processing unit are determined; determining the model configuration information analysis results of the data processing model based on the configuration information screening results of each processing unit in the audio processing model.

[0009] In order to solve the above problems, an embodiment of the present application discloses a data processing method, including: dividing a data processing model, determining a first sub-model and a second sub-model, the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit; determining the first configuration information of the first sub-model and the second configuration information of the second sub-model, and determining the configuration information combination; determining the data processing time corresponding to each configuration information combination, and filtering out the target configuration information combination as the configuration information filtering result of the first sub-model and the second sub-model; dividing and analyzing the first sub-model and the second sub-model as models to be processed until the configuration information filtering result of each processing unit is determined; determining the model configuration information analysis result of the data processing model based on the configuration information filtering result of each processing unit in the data processing model.

[0010] In order to solve the above problems, an embodiment of the present application discloses a data processing method, including: dividing a data processing model, determining a first sub-model and a second sub-model, and associating the output data of the first sub-model with the input data of the second sub-model; determining first configuration information of the first sub-model and second configuration information of the second sub-model, and determining a configuration information combination; determining the data processing time corresponding to each configuration information combination, and screening out a target configuration information combination as the configuration information screening result of the first sub-model and the second sub-model; performing a first optimization process on the first sub-model, and performing a second optimization process on the second sub-model.

[0011] In order to solve the above problems, an embodiment of the present application discloses a data processing device, which includes: a sub-model acquisition module, used to split the data processing model, determine the first sub-model and the second sub-model, and the output data of the first sub-model is associated with the input data of the second sub-model; a configuration information combination acquisition module, used to determine the first configuration information of the first sub-model and the second configuration information of the second sub-model as a configuration information combination; a screening result acquisition module, used to determine the data processing time corresponding to each configuration information combination, and screen out at least one group of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model.

[0012] In order to solve the above problems, an embodiment of the present application discloses an electronic device, including: a processor; and a memory, on which executable code is stored. When the executable code is executed, the processor executes one or more methods described in the above method embodiments.

[0013] In order to solve the above problems, the embodiments of the present application disclose one or more machine-readable media on which executable codes are stored. When the executable codes are executed, the processor executes one or more methods described in the above method embodiments.

[0014] Compared with the prior art, the embodiments of the present application have the following advantages:

[0015] In an embodiment of the present application, the data processing model can be divided into a first sub-model and a second sub-model, and the first configuration information of the first sub-model and the second configuration information of the second sub-model are determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is determined, and based on the data processing time, a large number of configuration information combinations that do not match the sub-model are subtracted to screen out a small number of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. In an embodiment of the present application, in the process of analyzing the data processing model, not only the processing time when the sub-model is configured according to each configuration information is considered, but also the time consumed for data conversion between related sub-models is considered, so that a configuration information combination that is more suitable for the model can be screened out, thereby improving the processing efficiency of the data processing model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1A This is a flow chart of a data processing method according to an embodiment of the present application;

[0017] Figure 1B is a schematic diagram of the data format of an embodiment of the present application;

[0018] Figure 1C This is a schematic diagram of the calculation time of a neural network model according to one embodiment of the present application;

[0019] Figure 2A is a flowchart of a data processing method according to another embodiment of the present application;

[0020] Figure 2B This is a flow chart of a data format analysis method according to an embodiment of the present application;

[0021] Figure 2C is a flow chart of a data format analysis method according to another embodiment of the present application;

[0022] Figure 2D 1 is a flow chart of a data format analysis method according to another embodiment of the present application;

[0023] Figure 3 is a flowchart of a data processing method according to another embodiment of the present application;

[0024] Figure 4 is a flowchart of a data processing method according to another embodiment of the present application;

[0025] Figure 5 is a flowchart of a data processing method according to another embodiment of the present application;

[0026] Figure 6 is a flowchart of a data processing method according to another embodiment of the present application;

[0027] Figure 7 is a flowchart of a data processing method according to another embodiment of the present application;

[0028] Figure 8 is a structural diagram of a data processing device according to an embodiment of the present application;

[0029] Figure 9 is a structural diagram of a data processing device according to another embodiment of the present application;

[0030] Figure 10 is a structural diagram of a data processing device according to another embodiment of the present application;

[0031] Figure 11 is a structural diagram of a data processing device according to another embodiment of the present application;

[0032] Figure 12 is a structural diagram of a data processing device according to another embodiment of the present application;

[0033] Figure 13 It is a structural diagram of a device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0035] The embodiments of the present application can be applied to the field of optimization of neural network models, which can also be called neural networks (NNs), data processing models. Neural network models are network systems formed by interconnecting processing units, wherein the processing units can also be called computing nodes, computing units, etc. The processing units include at least one operator, which is the basic unit that constitutes the neural network model and is the smallest unit for data processing in the neural network model. This embodiment can optimize the configuration of each processing unit in the data processing model. In one example, the processing unit can be an operator, and the input data format and output data format of the operator can be optimized to determine the optimal input data format and output data format of each operator as a whole to improve the processing efficiency of the data processing model; it can also configure a corresponding processor for each operator in the model, and configure a better processor for each operator as a whole to improve the processing efficiency of the data processing model. Among them, the data input or output of the operator can include tensors, vectors, matrices, etc. Tensors can be understood as multidimensional arrays. The purpose of using tensors is to create more dimensional matrices and vectors to provide them to the neural network model for use. Tensor data format can be understood as the format (or order) in which Tensors are stored in memory, i.e., memory layout. Different data formats correspond to different storage orders. For example, Figure 1B As shown, Figure 1B The figure shows the data storage order of two different data formats for the same set of data. When operators process data according to different data formats, they need to obtain data from different locations. Therefore, the different data formats of the data processed by the operators will lead to different operator processing efficiency.

[0036] In the embodiment of this application, Figure 1A In the example, the configuration of the data format of an operator is described as an example (the configuration information adopts the data format), such as Figure 1A As shown, the data processing model can be divided into a first sub-model and a second sub-model, wherein the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one operator. After determining the first sub-model and the second sub-model, the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination, wherein the configuration information can be understood as a configuration related to the data processing speed of the sub-model, such as the configuration of the data format of each processing unit in the sub-model, the hardware (processor) configuration used by the processing unit in the sub-model, etc. Figure 1AIn the example shown, the first output data format of the first sub-model and the second input data format of the second sub-model can be enumerated as a configuration information combination. Then, according to the configuration information combination, the processing time of the first sub-model, the processing time of the second sub-model, and the data conversion time of the data conversion between the first sub-model and the second sub-model are determined to determine the data processing time corresponding to the configuration information combination, and based on the data processing time, multiple groups of target configuration information combinations are screened out as the configuration information screening results of the first sub-model and the second sub-model ( Figure 1A The first and second sub-models can then be segmented and analyzed as models to be processed until the configuration information screening results (data format screening results) for each processing unit (e.g., operator) in the data processing model are determined. The unit configuration information corresponding to each processing unit in the model is then determined based on the configuration information screening results for each processing unit to determine the model configuration information combination for the data processing model. A target model configuration information combination suitable for the data processing model is then screened out as the model configuration information analysis result.

[0037] In an embodiment of the present application, during the analysis of the data processing model, not only the processing time of the sub-model according to the corresponding configuration is considered, but also the time consumed for data conversion between related sub-models is considered. This can screen out a model configuration information combination that is more suitable for the model, thereby improving the optimization effect of the data processing model.

[0038] Taking the optimization of the data format of the processing unit (such as an operator) in the data processing model as an example, a method for configuring the data format of the data processing model is to enumerate the data formats corresponding to each operator in the data processing model, and then analyze the model configuration information analysis results suitable for the data processing model as a whole. However, in this way, the complexity of the analysis is related to the number of operators in the data processing model. The more operators there are, the more complex the calculation increases exponentially. In the method adopted in the embodiment of the present application, the data processing model can be divided to obtain a first sub-model and a second sub-model, and the configuration information combination corresponding to the first sub-model and the second sub-model (such as a data format combination, a processor applying the operator, etc.) is determined for analysis, and then according to the preset screening conditions, most of the configuration information combinations that are not suitable for the first sub-model and the second sub-model are removed (also referred to as pruning, pruning acceleration), so as to obtain a small number of target configuration information combinations to reduce the complexity of the later calculations. After the configuration information of the two sub-models is determined, the further optimization of the internal operators of the two sub-models will not affect each other. Therefore, the first sub-model and the second sub-model can be treated as separate models, and the corresponding configuration information combinations can be further divided and analyzed until the configuration information screening results corresponding to each operator (processing unit) are determined. After the configuration information screening results corresponding to each operator are determined, a small amount of operator configuration information corresponding to each operator can be combined to obtain a model configuration information combination, and the data processing time corresponding to each model configuration information combination can be analyzed to determine the target model configuration information combination suitable for the data processing model as the model configuration information analysis effect. The embodiment of the present application can utilize the method of dividing the data processing model to remove a large amount of configuration information that does not match the divided sub-models. In the process of analyzing the data processing model in the later stage, the amount of configuration information corresponding to each operator can be reduced, thereby reducing the computational complexity and improving the data processing efficiency.

[0039] Taking the configuration information as the data format as an example, this embodiment can enumerate the first output data format of the first sub-model and the second input data format of the second sub-model as a configuration information combination, so as to screen out a small number of target data formats as the data format screening results of the sub-models (or configuration information screening results). Among them, for a data processing model containing multiple operators, before the pruning method is used to reduce the data formats corresponding to the operators, it usually takes ten minutes or twenty minutes or more to perform calculations and obtain the corresponding model configuration information combination. By adopting the method of this solution, a large number of mismatched data formats can be reduced, which greatly reduces the complexity of the calculation. The calculation can usually be completed within a few seconds to obtain the corresponding model configuration information combination. Moreover, for a data processing model with optimized data format, the processing efficiency of the data processing model has been significantly improved, such as Figure 1CAs shown in the figure, the calculation time of several common neural network models before and after optimization is shown. The data processing efficiency of the data processing model after optimization using the method of this embodiment is improved by about 1.75 times on average. Among them, Res Net50 and Res Net101 refer to residual network models with different numbers of layers; Mobile Net, Squeeze Net, and Shuffle Net are lightweight neural network models, which are obtained by compressing the neural network model in different ways and can be applied to mobile terminal devices.

[0040] The data processing method of the embodiment of the present application is to select and optimize the basic level of the data processing model (the data format of the data processed by the processing unit, the processor used by the processing unit). Therefore, the data processing method of the embodiment of the present application can be applied to the data processing model of each scene. For example, it can be applied to the data format optimization scene of the image processing model, audio processing model and other models to improve the efficiency of image processing and audio processing; it can also be applied to the data processing model for data analysis of big data (such as analysis of user consumption behavior). Specifically, the image processing model can be used to complete at least one of the following: image semantic recognition (such as recognition of people, animals, scenery, text, faces, fingerprints, etc.), image depth recognition, image optimization processing (such as recognition of image parameters and adjustment of image parameters), and image key point positioning. For example, the data method of the embodiment of the present application can be applied to a data processing model for image recognition of goods in the e-commerce field, wherein the data processing model can recognize the semantics of goods in the image, the depth of goods, the key point positions of goods, etc.; the data method of the embodiment of the present application can also be applied to a data processing model for entity object recognition in the construction of a logistics network, wherein the data processing model can perform entity object recognition on the relevant information of various logistics events and extract the entity objects therein (such as extracting the shipping place, shipping time, receiving place, etc.); the data processing method of the embodiment of the present application can also be applied to a data processing model for face recognition in a live broadcast scene, wherein the data processing model The model can perform face recognition and facial key point positioning in live videos (in order to beautify the face); the method of the embodiment of the present application can also be applied to a data processing model for recommending financial products in the financial field, wherein the data processing model can identify financial events, determine whether the financial events are positive events or negative events, and then determine the credibility of the financial products, and recommend highly credible financial products to users; the method of the embodiment of the present application can also be applied to a data processing model for group identification, wherein the data processing model can analyze information related to the user's behavioral habits, determine the group to which the user belongs, and then recommend the group (or friends in the group) to the user. The method of the embodiment of the present application can also be applied to a data processing model for image recognition of road surveillance videos, wherein the data processing model can perform image recognition on road surveillance videos to determine whether the vehicle's speed is too fast, whether the vehicle is driving in the wrong direction, and whether the driver has irregular behavior (such as not wearing a seat belt). The audio processing model can be used to complete at least one of the following: speech recognition, speech synthesis, audio filtering, etc., wherein the data processing model can also be a sub-model in a large model. For example, in a speech recognition model, the audio data can be a speech recognition model, or it can be a phoneme recognition model, a syntactic analysis model, etc. in the speech recognition model.For example, the method of this embodiment can be applied to a data processing model for voice recognition in a voice interaction scenario, wherein the data processing model can apply a smart speaker or a mobile terminal to recognize the user's voice data and output the corresponding feedback voice. The method of this embodiment can also be applied to a data processing model for voice recognition in the field of e-commerce live broadcast, wherein the data processing model can recognize the voice data in the e-commerce live broadcast video and determine the corresponding semantic information in order to perform subsequent operations, such as switching the product image displayed in the live broadcast video to the product image corresponding to the voice data, such as switching the price corresponding to the product, etc. The method of the embodiment of the present application can also be applied to a data processing model for recognizing and optimizing audio, wherein the data processing model can recognize the audio data and determine the corresponding audio parameters in order to optimize the audio data. For example, when the host is singing, the data processing model can recognize the audio data in order to optimize the host's audio data.

[0041] The embodiment of the present application provides a data processing method, by which the data format processed by the processing unit of the data processing model and / or the processor of the application processing unit can be selected and configured, thereby improving the data processing efficiency of the data processing model. The method can be executed by a processing end, which can be a training device for training the data processing model or a device for storing and transferring training data (for training the data processing model). Specifically, as shown in Figure 2, the method includes:

[0042] Step 202: Divide the data processing model to determine a first sub-model and a second sub-model, wherein the output data of the first sub-model is associated with the input data of the second sub-model. The first sub-model and the second sub-model are composed of at least one processing unit, and the processing unit includes at least one operator. The operators in the processing unit can be pre-divided for analysis, and the number of operators contained in the processing unit can be set according to the needs. The data processing model can be an image processing model for processing image data, an audio processing model for processing audio data, etc. In the embodiment of the present application, the data processing model can be divided into two sub-models, or the data processing model can be divided into multiple sub-models, and the sub-models can be divided into a first sub-model and a second sub-model according to the input-output relationship between the sub-models, wherein the output data of the first sub-model is associated with the input data of the second sub-model. On the one hand, the output data of the first sub-model can be used as the input data of the second sub-model. On the other hand, the output data of the first sub-model can be converted into the input data of the second sub-model through data format conversion. This embodiment can pre-set the model segmentation method to segment the data processing model. For example, according to the number of operators (such as twenty), two sub-models each containing ten operators can be obtained. The specific model segmentation method can be set according to needs.

[0043] The embodiment of the present application can analyze the relationship between operators and operators in the data processing model to determine topological sorting information so that the sub-models can be divided according to the topological sorting information. Specifically, as an optional embodiment, the data processing model is divided to determine the first sub-model and the second sub-model, including: dividing the data processing model to obtain the sub-models after division; obtaining the topological sorting information corresponding to the data processing model, and dividing the sub-models after division to obtain the first sub-model and the second sub-model, wherein the topological sorting information is determined according to the structure of the data processing model. The topological sorting relationship can be determined based on the relationship between operators and operators in the data processing model (characterizing the order of data transmission between operators). For example, the output data of operator A is associated with the input data of operator B, and the output data of operator B is associated with the input data of operator C. Then, operator A can be determined to be the superior operator of operator B, and operator B is the superior operator of operator C. The topological relationship between operator A and operator B, and between operator B and operator C is obtained, and then the topological sorting relationship between operators in the data processing model is obtained. Then, based on the topological sorting relationship, the operators corresponding to the split points in the split sub-models are classified, the superior-subordinate relationship between the operators at the split points is determined, and then the first sub-model and the second sub-model are divided.

[0044] After determining the first sub-model and the second sub-model, the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined in step 204 as a configuration information combination. The configuration information of the sub-model can be understood as a configuration related to the data processing speed of the sub-model, such as the data format corresponding to each processing unit in the sub-model, the hardware (processor) of the processing unit in the application sub-model, etc. Specifically, as an optional embodiment, the determination of the first configuration information of the first sub-model and the second configuration information of the second sub-model as a configuration information combination includes at least one of the following steps: determining the first output data format of the first sub-model and the second input data format of the second sub-model as a configuration information combination; determining the first processor of the first sub-model and the second processor of the second sub-model as a configuration information combination. For the data format of the configuration processing unit, a data format table storing the data format can be pre-set, and the first output data format of the first sub-model and the second input data format of the second sub-model can be determined according to the data format table to form multiple groups of data format combinations (or configuration information combinations). For the hardware (processor) configuring the application processing unit, the corresponding hardware (processor) can be enumerated for each operator in the sub-model to form multiple groups of configuration information combinations. For example, for the operators in the sub-model, the processors of the application operators may include a central processing unit (CPU), a graphics processing unit (GPU), an embedded neural network processor (NPU), etc. The corresponding processors can be enumerated for the operators at the splitting points respectively to obtain the corresponding configuration information combinations.

[0045] After the configuration information combination is determined, in step 206, the data processing duration corresponding to each configuration information combination can be determined, and at least one group of target configuration information combinations can be screened out as the configuration information screening results of the first sub-model and the second sub-model. The data processing duration includes the duration for the first sub-model and the second sub-model to process data after being configured according to the configuration information, and the conversion duration consumed by the data conversion between the first sub-model and the second sub-model. Specifically, as an optional embodiment, the determination of the data processing duration corresponding to each data format combination includes: determining a first duration for the first sub-model to process data according to the first configuration information; determining a second duration for the second sub-model to process data according to the second configuration information; determining a third duration based on the first configuration information and the second configuration information; and determining the data processing duration based on the first duration, the second duration, and the third duration.

[0046] When the configuration information is a data format, this embodiment can use the model input data of the data processing model as the input data of the first sub-model, and the first output data format in the configuration information combination as the output data of the first sub-model to obtain the first duration for the first sub-model to process the data. The second input data format in the configuration information combination can be used as the input data of the second sub-model, and the model output data of the data processing model can be used as the output data of the second sub-model to obtain the second duration for the second sub-model to process the data. In the process of determining the third duration, it can be determined whether the first output data format and the second output data format are the same. If they are the same, the third duration is determined to be zero. If they are not the same, the third duration consumed by converting the first output data format to the second input data format is determined. The data processing duration is then determined based on the first duration, the second duration, and the third duration. When the configuration information is processor information, the first duration consumed by the first sub-model application for data processing on the first processor can be obtained, the second duration consumed by the second sub-model application for data processing on the second processor can be obtained, and the third duration consumed by data conversion between different processors can be determined, and then the data processing duration can be determined based on the first duration, the second duration, and the third duration. In the embodiment of the present application, during the analysis of the data processing model, not only the processing duration of the sub-model for data processing according to each configuration information is considered, but also the duration consumed by the related sub-models for data conversion according to the configuration information is considered, so that a data format and / or processor that is more suitable for the model can be screened out, thereby improving the data processing efficiency of the data processing model.

[0047] The following uses the optimization of data format as an example to illustrate the data processing method. Figure 2B In the example shown, the data processing model G includes operator i-1, operator i, operator i+1 and operator i+2, then the data processing model can be divided to obtain a first sub-model G1 and a second sub-model G2. For the data processing model G, the time consumed by the data processing model G for data processing is related to the time consumed by the first sub-model G1 for data processing and the time consumed by the second sub-model G2 for data processing, as well as the time consumed for data conversion between the first sub-model and the second sub-model. Specifically, the data processing time of the data processing model G can be determined by the following formula.

[0048] Formula 1:

[0049] opt(G)=opt(G1)+opt(G2)+convert(L1, L2)

[0050] Where opt(G) is the data processing time of data processing model G;

[0051] opt(G1) is the data processing time of the first sub-model G1;

[0052] opt(G2) is the data processing time of the second sub-model G2;

[0053] L1 is the output data format of the first sub-model G1, and L2 is the input data format of the second sub-model G2;

[0054] convert(L1, L2) is the time taken to convert data format L1 to data format L2.

[0055] The data format L1 and the data format L2 may correspond to one data format or multiple data formats. Specifically, Figure 2C As shown, in Figure 2C In the example, the data processing model G includes operator i-1, operator i, operator i+1 and operator i+2, then the data processing model G can be divided to obtain a first sub-model G1 and a second sub-model G2, wherein the first sub-model G1 includes two output Tensors, and the second sub-model G2 includes two input Tensors. The data formats t1 and t2 of the two Tensors of the first sub-model G1 can be enumerated, and the data formats t3 and t4 of the two Tensors of the second sub-model G2 can be enumerated, and the corresponding data processing time can be obtained. Specifically, the method for obtaining the data processing time can be determined according to the following formula.

[0056] Formula 2:

[0057] opt(G)=opt(G1,t1,t2)+opt(G2,t3,t4)+convert(t1,t3)+convert(t2,t4)

[0058] Among them, opt(G) is the data processing time of the data processing model G, which can also be understood as the data processing time corresponding to the data format combination.

[0059] opt(G1, t1, t2) is the first duration consumed by the first sub-model G1 for data processing according to the data formats t1 and t2.

[0060] opt(G2, t3, t4) is the second time duration consumed by the first sub-model G1 for data processing according to the data formats t3 and t4.

[0061] convert(t1, t3) is the third duration consumed for converting the data format from t1 to t3.

[0062] convert(t2, t4) is the third duration consumed for converting the data format from t2 to t4.

[0063] In an optional embodiment, the present embodiment can take the entire sub-model as the analysis object, and analyze the time consumed for data processing according to the enumerated data format. Specifically, a preset fixed data processing format can be configured for other operators except the operator at the split, and then the operators at the split (such as Figure 2B In another optional embodiment, the operator at the split point can be used as the analysis object to analyze the time consumed for data processing according to the enumerated data format. Specifically, Figure 2C As shown, for a data processing model G, given the model input data format and model output data format of model G, after the first split, the first sub-model (including operators i-1, i, and i+1) and the second sub-model (including operator i+2) are obtained. The output data format (model output data format) of the last operator (operator i+2) of model G can be used to determine the output data format (t5) of the first sub-model after the first split and the input data format of the second sub-model. During the second split (splitting the first sub-model obtained by the first split), the first sub-model (including operators i-1 and i) and the second sub-model (including operator i+1) can be obtained. Then, the data format screening result t5 obtained by the first split can be used as the output data format of operator i+1. The time consumed by operator i+1 to process data according to the enumerated input data formats is analyzed, and the data processing time corresponding to different data format combinations is determined. Then, in a continuous forward iterative manner, the data formats corresponding to all operators in the data processing model are determined. The data processing time corresponding to the data format combination of operator i+1 can be determined by the following formula.

[0064] Formula 3:

[0065] opt(i+1,t5)=min[opt(i,t1,t2)+convert(t1,t3)+convert(t2,t4)+node(i+1,t3,t4,t5)], where opt(i+1,t5) is the data processing time of operator i+1 in the first state. The first state refers to the state in which the output data format of the previous operator (i) of operator i+1 is t1 and t2, the input data format of this operator i+1 is t3 and t4, and the output data format is t5.

[0066] opt(i, t1, t2) is the first duration consumed by operator i for data processing according to data formats t1 and t2. The duration consumed by operator i for data processing is related to operator i-1. Therefore, this formula can be used to continuously iterate forward to determine the data format of each operator. During the iterative process, each calculation only calculates the data format between the operators at the split (there is no need to calculate the entire model at once), and the pruning strategy can be used to continuously reduce the data format corresponding to the operator, thereby reducing the amount of calculation.

[0067] convert(t1, t3) is the third duration consumed for converting the data format from t1 to t3.

[0068] convert(t2, t4) is the third duration consumed for converting the data format from t2 to t4.

[0069] node(i+1, t3, t4, t5) is the second duration consumed by operator i+1 for data processing according to input data formats t21 and t22 and output data format t3.

[0070] When analyzing the model format combinations of a data processing model as a whole, the time consumed by each sub-model is cumulative. Therefore, for a set of first and second sub-models, a pruning strategy can be used to remove a large number of configuration information combinations with long data processing times, leaving a small number of target configuration information combinations (with short data processing times) for subsequent analysis. During the pruning process, a screening condition can be used to screen out target configuration information combinations that meet the screening condition, thereby eliminating a large number of mismatched configuration information combinations. Specifically, as an optional embodiment, screening out at least one set of target data format combinations includes: sorting the configuration information combinations according to data processing time; and screening out at least one set of target configuration information combinations that meet a preset screening condition based on the sorted configuration information combinations. The preset screening condition can be a preset screening ratio or a preset threshold. In an optional embodiment, the screening ratio can be pre-set on the processing end, and the processing end can remove configuration information combinations that do not meet the screening ratio based on the preset screening ratio. Specifically, screening out at least one set of target configuration information combinations that meet the preset screening condition includes: screening out at least one set of target configuration information combinations based on the preset screening ratio. For example, the preset screening ratio can be 1% (or 20%), etc. In this embodiment, the configuration information combinations can be sorted according to the data processing time, and then the configuration information combinations with the shortest data processing time are screened out as the configuration information screening results of the first sub-model and the second sub-model.

[0071] In another optional embodiment, a screening threshold can be determined according to the shortest data processing time in the configuration information combination, and the target configuration information combination can be screened out according to the screening threshold. Specifically, the screening out of at least one group of target configuration information combinations that meet the preset screening conditions includes: determining a first configuration information combination whose ranking meets the ranking conditions; determining a screening threshold based on the data processing time of the first configuration information combination; and screening out at least one group of target configuration information combinations that meet the screening threshold. The processing end can screen out the first configuration information combination with the shortest data processing time according to the ranking result of the configuration information combination, and determine the corresponding screening threshold based on the data processing time of the configuration information combination, and then screen out the target configuration information combination. For example, if the shortest data processing time screened out is 10ms, the corresponding screening threshold can be determined to be 15ms, and then the target configuration information combination that meets 15ms can be screened out as the configuration information screening result.

[0072] In an optional example, during the pruning process, the data processing times corresponding to different configuration information can also be compared, and then the configuration information with longer data processing time can be pruned. Taking the data format combination as an example, the input data format of operator i+1 is determined to be t11 and t13. For operator i, it can be determined that the data format combination of its output data includes combination 1 (t11 and t12) and combination 2 (t11 and t13). It can be determined that t12 in combination 1 needs to be converted to t13 before it can be input into operator i+1; combination 2 does not need to be converted and can be input into operator i+1; therefore, the two time lengths of the data conversion time of combination 1 and the calculation time of operator i in the combination 1 format can be combined and compared with the calculation time of the operator in the combination 2 format to obtain the following formula.

[0073] Formula 4: opt(i, t11, t12) + convert(t12, t13) > opt(i, t11, t13)

[0074] Here, opt(i, t11, t12) represents the time consumed by the operator when the output data format is in the t11 and t12 states.

[0075] convert(t12, t13) represents the time it takes to convert data from t12 format to t13 format.

[0076] opt(i, t11, t1) represents the time the operator consumes when the output data format is in the t11 and t12 states.

[0077] From this, we can see that operator i spends more time in the state corresponding to combination 1, so the solution of combination 1 can be cut off and combination 2 can be retained.

[0078] After determining the configuration information screening results corresponding to the first sub-model and the second sub-model obtained after the first segmentation, as an optional embodiment, the first sub-model and the second sub-model can be segmented and analyzed as models to be processed until the configuration information screening results of each processing unit are determined. After determining the data format between the two sub-models (or determining the processor of the application sub-model), the further optimization of the internal operators of the two sub-models will not affect each other. Therefore, it is possible to determine whether any of the first sub-model and the second sub-model contains two or more processing units, and the sub-model containing two or more processing units can be used as a separate model, and the configuration information combination corresponding to the model after the sub-model segmentation is analyzed. According to the iterative analysis method of the segmented model, the number of processing units contained in the sub-model can be continuously reduced until the configuration information screening results corresponding to each processing unit are determined. The embodiment of the present application can use the method of continuously segmenting the data processing model to remove a large amount of configuration information that does not match the segmented sub-model. In the process of analyzing the data processing model in the later stage, the amount of configuration information corresponding to each processing unit can be reduced, thereby reducing the computational complexity and improving the optimization efficiency of the model.

[0079] After determining the configuration information screening results for each processing unit, as an optional embodiment, the model configuration information analysis results of the data processing model can be determined based on the configuration information screening results for each processing unit in the data processing model. When the configuration information includes data formats, the model configuration information analysis results can include the input data formats and output data formats of each processing unit in the model; when the configuration information includes the hardware of the application processing unit, the model configuration information analysis results can include the processors used by each processing unit in the model. The data formats (or corresponding processors) corresponding to each processing unit can be combined to obtain configuration information combinations, and the data processing time of each combination can be determined to obtain the model configuration information analysis results. Specifically, as an optional embodiment, determining the model configuration information analysis results of the data processing model based on the configuration information screening results for each processing unit in the data processing model includes: determining the unit configuration information corresponding to each processing unit according to the configuration information screening results of each processing unit to determine the model configuration information combination of the data processing model; determining the model processing time corresponding to each model configuration information combination, and screening out the target model configuration information combination as the model configuration information analysis result. Unit configuration information refers to configuration information related to the processing speed of the processing unit. For example, when the processing unit is an operator, the unit configuration information may include the operator data format and the processor information configured for the operator, wherein the operator data format includes the operator's input data format and the operator's output data format. In this embodiment, the unit configuration information of adjacent processing units can be combined to determine the model configuration information combination of the corresponding data processing model, and then according to the model processing time corresponding to each model configuration information combination, one or several groups of model format combinations with the shortest model processing time are screened out as the model format analysis result, wherein the model processing time is the time consumed by the entire data processing model for data processing, which may include the time consumed by each processing unit for data processing according to the corresponding configuration and the time consumed for data conversion between each processing unit. Taking the data format of the processing unit configuration as an example, the configuration information screening results corresponding to each processing unit may include four data formats (two inputs and two outputs). Then, the four data formats of the first operator can be matched with the four data formats of the second operator to obtain sixteen matching results. Then, the sixteen matching results can be matched with the four data formats of the third operator to obtain sixty-four matching results, until the last operator is matched to obtain the model configuration information combination corresponding to the entire data processing model. Then, the model processing time corresponding to each model configuration information combination can be analyzed, and then a group (or multiple groups) of model configuration information combinations with the shortest model processing time can be selected as the model configuration information analysis result.

[0080] As an optional embodiment, a comparison table can be pre-set, and the comparison table stores the processing time of the processing unit and the data conversion time. The processing time of the processing unit refers to the time it takes for the processing unit to process according to different data formats (or on different processors), and the data conversion time refers to the time consumed for data format conversion between different data formats (or data conversion between different processors). The model processing time corresponding to the model configuration information combination can be determined through the comparison table, and then screening can be performed. As another optional embodiment, the determination of the model processing time corresponding to each model configuration information combination includes: configuring the data processing model according to the model configuration information combination; performing data processing based on the configured data processing model to determine the model processing time. According to the model configuration information combination, the input data format, output data format, data format conversion between processing units, processors used by processing units, and data conversion between processors of each processing unit (such as an operator) in the data processing model can be configured. Then, the model processing time can be determined according to the time consumed by each processing unit for data processing and the time consumed for data conversion between adjacent processing units, so that the model configuration information combination can be sorted and filtered according to the model processing time. Specifically, as an optional embodiment, the target model configuration information combination is screened, including: sorting the model configuration information combination according to the model processing time corresponding to each model configuration information combination; screening the target model configuration information combination according to the sorted model configuration information combination. The model configuration information combination can be sorted from short to long (or from long to short) according to the model processing time according to a pre-set sorting rule, and then one or more target model configuration information combinations with the shortest model processing time are screened as the model configuration information analysis result. After the model configuration information combinations are determined and filtered out, the model configuration information combinations can also be sent to the user, and the user can select one or more groups of model configuration information combinations as the target model configuration information combination. Specifically, as an optional embodiment, the target model configuration information combination is filtered out based on the sorted model configuration information combinations, including: filtering out at least one group of model configuration information combinations to be confirmed based on the sorted model configuration information combinations, and sending it; receiving feedback information on the model configuration information combinations to be confirmed, and determining the target model configuration information combination. In some scenarios, the better model configuration information combination may not match the user's neural network model database. Therefore, after filtering the sorted model configuration information combinations, the filtered model configuration information combinations can be sent to the user. The user can select the model configuration information combination that meets the data processing model processing requirements (the one with the shortest data processing time or the second shortest data processing time) and upload the feedback information so that the target model configuration information combination can be determined based on the feedback information as the model configuration information analysis result.As an optional embodiment, the method further includes: configuring a data processing model according to the results of the model configuration information analysis. The configured data processing model can be trained, and the trained data processing model can be used to recognize and process images and audio.

[0081] In an embodiment of the present application, a data processing model can be divided into a first sub-model and a second sub-model, and the first configuration information of the first sub-model and the second configuration information of the second sub-model are determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is determined, and based on the data processing time, a large amount of configuration information that does not match the sub-model is removed to screen out a small number of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. Afterwards, the first sub-model and the second sub-model can be divided and analyzed as models to be processed respectively until the configuration information screening results of each processing unit in the data processing model are determined. Then, according to the configuration information screening results of each processing unit, the model configuration information combination of the data processing model is determined, and the target model configuration information combination suitable for the data processing model is screened out as the model configuration information analysis result. In an embodiment of the present application, during the analysis of the data processing model, not only the processing time of the sub-model according to each configuration information is considered, but also the time spent on data conversion between related sub-models according to the configuration information is considered, so that configuration information that is more suitable for the model can be screened out, thereby improving the processing efficiency of the data processing model. In addition, the embodiment of the present application can use the method of splitting the data processing model to reduce a large amount of configuration information that does not match the split sub-model. In the later analysis of the data processing model, the amount of configuration information corresponding to each processing unit can be reduced, thereby reducing the computational complexity and improving data processing efficiency.

[0082] On the basis of the above embodiments, the embodiments of the present application further provide a data processing method that can be executed by a processing end, which can be a training device for training a data processing model or a device for storing and transferring training data for training a data processing model. Figure 3 As shown, the method includes:

[0083] Step 302: Split the data processing model to determine a first sub-model and a second sub-model, wherein the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit, each of which includes at least one operator. As an optional embodiment, step 302 specifically includes: splitting the data processing model to obtain split sub-models; obtaining topological sorting information corresponding to the data processing model, and partitioning the split sub-models to obtain a first sub-model and a second sub-model, wherein the topological sorting information is determined based on the structure of the data processing model.

[0084] Step 304: Determine first configuration information of the first sub-model and second configuration information of the second sub-model as a configuration information combination. As an optional embodiment, step 304 specifically includes at least one of the following steps: determining a first output data format of the first sub-model and a second input data format of the second sub-model as the configuration information combination; and determining a first processor of the first sub-model and a second processor of the second sub-model as the configuration information combination.

[0085] Step 306: Determine the data processing duration corresponding to each configuration information combination, and filter out at least one group of target configuration information combinations as the configuration information filtering results of the first sub-model and the second sub-model. As an optional embodiment, step 306 specifically includes: determining the first duration for the first sub-model to process data according to the first configuration information; determining the second duration for the second sub-model to process data according to the second configuration information; determining the third duration based on the first configuration information and the second configuration information; determining the data processing duration based on the first duration, the second duration and the third duration; sorting the configuration information combinations according to the data processing duration; and filtering out at least one group of target configuration information combinations that meet the preset filtering conditions according to the sorted configuration information combinations as the configuration information filtering results of the first sub-model and the second sub-model. Wherein, as an optional embodiment, filtering out at least one group of target configuration information combinations includes: sorting the configuration information combinations according to the data processing duration. According to the sorted configuration information combinations, filtering out at least one group of target configuration information combinations that meet the preset filtering conditions. As another optional embodiment, screening out at least one group of target configuration information combinations that meet preset screening conditions includes: screening out at least one group of target configuration information combinations according to a preset screening ratio.

[0086] Step 308: Determine whether the number of processing units in the sub-model is greater than one. If so, return to step 302 and segment and analyze the first and second sub-models as the models to be processed until the configuration information screening results for each processing unit are determined. If not, proceed to step 310.

[0087] Step 310: According to the configuration information screening results of each processing unit, the unit configuration information corresponding to each processing unit is determined to determine the model configuration information combination of the data processing model.

[0088] Step 312: Configure the data processing model according to the model configuration information combination.

[0089] Step 314: Perform data processing based on the configured data processing model and determine the model processing time.

[0090] Step 316: Sort the model configuration information combinations according to the model processing time corresponding to each model configuration information combination.

[0091] Step 318: Filter out a target model configuration information combination based on the sorted model configuration information combinations as the model configuration information analysis result.

[0092] Step 320: Configure the data processing model according to the model configuration information analysis results.

[0093] In an embodiment of the present application, a data processing model can be divided into a first sub-model and a second sub-model, and the first configuration information of the first sub-model and the second configuration information of the second sub-model are determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is determined, and based on the data processing time, a large number of configuration information combinations that do not match the sub-model are removed to screen out a small number of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. Afterwards, it is determined whether the number of processing units contained in each sub-model exceeds one. If so, the sub-model is divided and analyzed as a model to be processed. If not, based on the configuration information screening results corresponding to each processing unit, the model configuration information combination corresponding to the data processing model is determined. Then, based on the model processing time corresponding to the model configuration information combination, one or more groups of target model configuration information combinations with the shortest model processing time are screened out as the model configuration information analysis results. Based on the model configuration information analysis results, the data processing model is configured, and then the configured data processing model can be trained using training data so that data processing can be more efficiently performed based on the trained data processing model.

[0094] On the basis of the above embodiments, the embodiments of the present application also provide a data processing method, which can optimize the neural network model related to image processing. The neural network model related to image processing can be used to complete at least one of the following: image semantic recognition (such as recognition of people, animals, scenery, text, etc.), image depth recognition, image optimization processing (such as recognition of image parameters and adjustment of image parameters), and image key point positioning. This embodiment can analyze the data format of the image processing model and screen out image data format combinations suitable for the image processing model (or assign corresponding processors to each processing unit in the image processing model) to improve the processing speed of the image processing-related model for image data. Specifically, Figure 4 As shown, the method includes:

[0095] Step 402: Divide the image processing model to determine a first sub-model and a second sub-model, wherein the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit.

[0096] Step 404: Obtain the image configuration information table corresponding to the image processing model.

[0097] Step 406: Determine, according to the image configuration information table, the first configuration information of the first sub-model and the second configuration information of the second sub-model as a configuration information combination.

[0098] Step 408: Determine the data processing time corresponding to each configuration information combination, and filter out at least one set of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model.

[0099] Step 410: The first sub-model and the second sub-model are divided and analyzed as models to be processed until the configuration information screening result of each processing unit is determined.

[0100] Step 412: Determine the model configuration information analysis result of the data processing model based on the configuration information screening result of each processing unit in the image processing model.

[0101] The implementation of this embodiment is similar to that of the above embodiment. The specific implementation can refer to the specific implementation of the above embodiment, which will not be repeated here.

[0102] In an embodiment of the present application, the image processing model can be divided into a first sub-model and a second sub-model, and an image configuration information table corresponding to the image is obtained. The image configuration information table stores a variety of image data formats and a variety of processors for application processing units. Based on the image configuration information table, the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is determined, and based on the data processing time, a large number of configuration information combinations that do not match the sub-model are subtracted to screen out a small number of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. Afterwards, the first sub-model and the second sub-model can be divided and analyzed as models to be processed respectively until the configuration information screening results of each processing unit in the data processing model are determined. Then, according to the configuration information screening results of each processing unit, the model configuration information combination of the image processing model is determined, and the target model configuration information combination suitable for the image processing model is screened out as the model configuration information analysis result.

[0103] On the basis of the above embodiments, the embodiments of the present application further provide a data processing method that can optimize a neural network model related to audio processing. The neural network model related to audio processing can be used to complete at least one of the following: speech recognition, speech synthesis, audio filtering, etc. This embodiment can analyze the data format of the audio processing model and screen out an audio data format combination suitable for the audio processing model (or assign corresponding processors to operators in the audio processing model) to improve the processing speed of the audio processing-related model for image data. Specifically, Figure 5 As shown, the method includes:

[0104] Step 502: Split the audio processing model to determine a first sub-model and a second sub-model, where the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit.

[0105] Step 504: Obtain the audio configuration information table corresponding to the audio processing model.

[0106] Step 506: Determine the first configuration information of the first sub-model and the second configuration information of the second sub-model according to the audio configuration information table as a configuration information combination.

[0107] Step 508: Determine the data processing time corresponding to each configuration information combination, and filter out at least one set of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model.

[0108] Step 510: The first sub-model and the second sub-model are divided and analyzed as models to be processed until the configuration information screening result of each processing unit is determined.

[0109] Step 512: Determine the model configuration information analysis result of the data processing model based on the configuration information screening result of each processing unit in the audio processing model.

[0110] The implementation of this embodiment is similar to that of the above embodiment. The specific implementation can refer to the specific implementation of the above embodiment, which will not be repeated here.

[0111] In an embodiment of the present application, the audio processing model can be divided into a first sub-model and a second sub-model, and an audio configuration information table corresponding to the audio data can be obtained, wherein the audio configuration information table stores a plurality of audio data formats and a plurality of processors for the application processing unit. Based on the audio configuration information table, the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is determined, and based on the data processing time, a large number of configuration information combinations that do not match the sub-model are subtracted to screen out a small number of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. Afterwards, the first sub-model and the second sub-model can be respectively divided and analyzed as models to be processed until the configuration information screening results of each processing unit in the audio processing model are determined. Then, according to the configuration information screening results of each processing unit, the model configuration information combination of the audio processing model is determined, and the target model configuration information combination suitable for the audio processing model is screened out as the model configuration information analysis result.

[0112] On the basis of the above embodiments, the embodiments of the present application further provide a data processing method that can be executed by a processing terminal, and can select a corresponding data format (or processor) for a processing unit of a data processing model for optimization, wherein the data processing model can be a neural network model for image processing, a neural network model for audio processing, etc. This embodiment can divide the data processing model into multiple sub-models and filter out data formats (or processors) suitable for the sub-models, thereby improving the data processing speed of the data processing model as a whole. Specifically, Figure 6 As shown, the method includes:

[0113] Step 602: Split the data processing model to determine a first sub-model and a second sub-model, where the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit.

[0114] Step 604: Determine the first configuration information of the first sub-model and the second configuration information of the second sub-model, and determine a configuration information combination.

[0115] Step 606: Determine the data processing time corresponding to each configuration information combination, and filter out the target configuration information combination as the configuration information screening result of the first sub-model and the second sub-model.

[0116] Step 608: The first sub-model and the second sub-model are divided and analyzed as models to be processed until the configuration information screening result of each processing unit is determined.

[0117] Step 610: Determine the model configuration information analysis result of the data processing model based on the configuration information screening result of each processing unit in the data processing model.

[0118] The implementation of this embodiment is similar to that of the above embodiment. The specific implementation can refer to the specific implementation of the above embodiment, which will not be repeated here.

[0119] In an embodiment of the present application, the data processing model can be segmented to obtain sub-models, and then the first sub-model and the second sub-model in the sub-model can be determined according to the order of data processing between the sub-models, and then the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination. Then the data processing time corresponding to the configuration information combination is obtained, wherein the data processing time includes the time consumed by the sub-model for data processing according to the corresponding configuration information and the time consumed by data conversion between sub-models. Then, based on the data processing time, a target configuration information combination suitable for the data processing model can be determined as a configuration information screening result. Afterwards, the first sub-model and the second sub-model can be segmented and analyzed as models to be processed respectively until the configuration information screening results of each processing unit in the data processing model are determined. Then, according to the configuration information screening results of each processing unit, the model configuration information combination of the data processing model is determined, and the target model configuration information combination suitable for the data processing model is screened out as a model configuration information analysis result.

[0120] On the basis of the above embodiments, the embodiments of the present application further provide a data processing method that can be executed by a processing terminal, which can segment the data processing model and filter out configuration information suitable for the sub-model, and can further optimize the sub-model using the same or different optimization methods, thereby improving the data processing speed of the data processing model as a whole. Specifically, Figure 6 As shown, the method includes:

[0121] Step 702: Split the data processing model to determine a first sub-model and a second sub-model, where the output data of the first sub-model is associated with the input data of the second sub-model.

[0122] Step 704: Determine the first configuration information of the first sub-model and the second configuration information of the second sub-model, and determine a configuration information combination.

[0123] Step 706: Determine the data processing time corresponding to each configuration information combination, and filter out the target configuration information combination as the configuration information screening result of the first sub-model and the second sub-model.

[0124] Step 708: Perform a first optimization process on the first sub-model and a second optimization process on the second sub-model.

[0125] In the embodiment of the present application, the first optimization process and the second optimization process can be the same or different. Specifically, in the embodiment of the present application, the optimization process of the sub-model can be optimized in the same manner as the optimization method of dividing the data processing model and performing configuration analysis, or it can be optimized by performing configuration analysis on the overall operator of the sub-model, without limitation. In this embodiment, the data processing model can be divided into sub-models, and then the first sub-model and the second sub-model in the sub-model can be determined according to the order of data processing between the sub-models. Then, the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is obtained, where the data processing time includes the time consumed by the sub-model for data processing according to the corresponding configuration information and the time consumed by data conversion between the sub-models. Then, based on the data processing time, a target configuration information combination suitable for the data processing model can be determined as the configuration information screening result. Afterwards, the first sub-model and the second sub-model can be further optimized using the same or different optimization processing methods.

[0126] It should be noted that for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the order of the actions described, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0127] Based on the above embodiment, this embodiment further provides a data processing device, referring to Figure 8 , specifically including the following modules:

[0128] The sub-model acquisition module 802 is used to segment the data processing model to determine a first sub-model and a second sub-model, where the output data of the first sub-model is associated with the input data of the second sub-model.

[0129] The configuration information combination acquisition module 804 is configured to determine first configuration information of the first sub-model and second configuration information of the second sub-model as a configuration information combination.

[0130] The screening result acquisition module 806 is used to determine the data processing time corresponding to each configuration information combination, and screen out at least one set of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model.

[0131] In summary, in the embodiments of the present application, the data processing model can be divided into a first sub-model and a second sub-model, and the first configuration information of the first sub-model and the second configuration information of the second sub-model are determined as a configuration information combination. The data processing duration corresponding to the configuration information combination is then determined, and based on the data processing duration, a large number of configuration information combinations that do not match the sub-models are removed to screen out a small number of target configuration information combinations as the configuration information screening results for the first sub-model and the second sub-model.

[0132] Based on the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0133] A model segmentation processing module is configured to segment the data processing model to determine a first sub-model and a second sub-model, wherein the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one operator. As an optional embodiment, the model segmentation processing module specifically includes: segmenting the data processing model to obtain segmented sub-models; obtaining topological sorting information corresponding to the data processing model, and partitioning the segmented sub-models to obtain a first sub-model and a second sub-model, wherein the topological sorting information is determined based on the structure of the data processing model.

[0134] The data configuration information enumeration processing module is configured to determine first configuration information of the first sub-model and second configuration information of the second sub-model as a configuration information combination. As an optional embodiment, the data configuration information enumeration processing module specifically includes at least one of the following steps: determining a first output data format of the first sub-model and a second input data format of the second sub-model as the configuration information combination; and determining a first processor of the first sub-model and a second processor of the second sub-model as the configuration information combination.

[0135] The data processing time acquisition processing module is used to determine the data processing time corresponding to each configuration information combination, and filter out at least one group of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. As an optional embodiment, the data processing time acquisition processing module specifically includes: determining the first time for the first sub-model to process data according to the first configuration information; determining the second time for the second sub-model to process data according to the second configuration information; determining the third time based on the first configuration information and the second configuration information; determining the data processing time based on the first time, the second time and the third time; sorting the configuration information combinations according to the data processing time; and filtering out at least one group of target configuration information combinations that meet the preset screening conditions according to the sorted configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. Wherein, as an optional embodiment, the data processing time acquisition processing module specifically includes: sorting the configuration information combinations according to the data processing time. Filtering out at least one group of target configuration information combinations that meet the preset screening conditions according to the sorted configuration information combinations. As another optional embodiment, the data processing time acquisition processing module specifically includes: screening out at least one group of target configuration information combinations according to a preset screening ratio.

[0136] The unit quantity determination processing module is used to determine whether the number of processing units in a sub-model is greater than one. If so, the module returns to the model segmentation processing module, where the first and second sub-models are segmented and analyzed as the models to be processed until the configuration information screening results for each processing unit are determined. If not, the module executes the model format combination acquisition processing module.

[0137] The model configuration information combination acquisition processing module is used to filter the results according to the configuration information of each processing unit, determine the operator configuration information corresponding to each processing unit, and thus determine the model configuration information combination of the data processing model.

[0138] The model configuration processing module is used to configure the data processing model according to the model configuration information combination.

[0139] The model processing time acquisition processing module is used to process data according to the configured data processing model and determine the model processing time.

[0140] The model configuration information combination sorting processing module is used to sort the model configuration information combinations according to the model processing time corresponding to each model configuration information combination.

[0141] The model configuration information combination screening processing module is used to screen out the target model configuration information combination according to the sorted model configuration information combinations as the model configuration information analysis result.

[0142] The data processing model configuration processing module is used to configure the data processing model according to the model configuration information analysis results.

[0143] In an embodiment of the present application, a data processing model can be divided into a first sub-model and a second sub-model, and the first configuration information of the first sub-model and the second configuration information of the second sub-model are determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is determined, and based on the data processing time, a large number of configuration information combinations that do not match the sub-model are removed to screen out a small number of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. Afterwards, it is determined whether the number of processing units contained in each sub-model exceeds one. If so, the sub-model is divided and analyzed as a model to be processed. If not, based on the configuration information screening results corresponding to each processing unit, the model configuration information combination corresponding to the data processing model is determined. Then, based on the model processing time corresponding to the model configuration information combination, one or more groups of target model configuration information combinations with the shortest model processing time are screened out as the model configuration information analysis results. Based on the model configuration information analysis results, the data processing model is configured, and then the configured data processing model can be trained using training data so that data processing can be more efficiently performed based on the trained data processing model.

[0144] Based on the above embodiment, this embodiment further provides a data processing device, referring to Figure 9 , specifically including the following modules:

[0145] The sub-model determination module 902 is used to divide the image processing model to determine the first sub-model and the second sub-model, the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit.

[0146] The image configuration table determination module 904 is used to obtain the image configuration information table corresponding to the image processing model.

[0147] The configuration combination determining module 906 is configured to determine, according to the image configuration information table, first configuration information of the first sub-model and second configuration information of the second sub-model as a configuration information combination.

[0148] The screening result determination module 908 is used to determine the data processing time corresponding to each configuration information combination, and screen out at least one set of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model.

[0149] The iterative result determination module 910 is used to segment and analyze the first sub-model and the second sub-model as models to be processed until the configuration information screening result of each processing unit is determined.

[0150] The analysis result determination module 912 is used to determine the model configuration information analysis result of the data processing model based on the configuration information screening result of each processing unit in the image processing model.

[0151] In an embodiment of the present application, the image processing model can be divided into a first sub-model and a second sub-model, and an image configuration information table corresponding to the image is obtained. The image configuration information table stores a variety of image data formats and a variety of processors for application processing units. Based on the image configuration information table, the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is determined, and based on the data processing time, a large number of configuration information combinations that do not match the sub-model are subtracted to screen out a small number of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. Afterwards, the first sub-model and the second sub-model can be divided and analyzed as models to be processed respectively until the configuration information screening results of each processing unit in the data processing model are determined. Then, according to the configuration information screening results of each processing unit, the model configuration information combination of the image processing model is determined, and the target model configuration information combination suitable for the image processing model is screened out as the model configuration information analysis result.

[0152] Based on the above embodiment, this embodiment further provides a data processing device, referring to Figure 10 , specifically including the following modules:

[0153] The sub-model acquisition module 1002 is used to divide the audio processing model to determine the first sub-model and the second sub-model, the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit.

[0154] The audio configuration table obtaining module 1004 is used to obtain the audio configuration information table corresponding to the audio processing model.

[0155] The configuration combination obtaining module 1006 is used to determine the first configuration information of the first sub-model and the second configuration information of the second sub-model according to the audio configuration information table as a configuration information combination.

[0156] The screening result obtaining module 1008 is used to determine the data processing time corresponding to each configuration information combination, and screen out at least one set of target configuration information combination as the configuration information screening result of the first sub-model and the second sub-model.

[0157] The iterative result obtaining module 1010 is used to segment and analyze the first sub-model and the second sub-model as models to be processed until the configuration information screening result of each processing unit is determined.

[0158] The analysis result obtaining module 1012 is used to determine the model configuration information analysis result of the data processing model based on the configuration information screening results of each processing unit in the audio processing model.

[0159] In an embodiment of the present application, the audio processing model can be divided into a first sub-model and a second sub-model, and an audio configuration information table corresponding to the audio data can be obtained, wherein the audio configuration information table stores a plurality of audio data formats and a plurality of processors for the application processing unit. Based on the audio configuration information table, the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is determined, and based on the data processing time, a large number of configuration information combinations that do not match the sub-model are subtracted to screen out a small number of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model. Afterwards, the first sub-model and the second sub-model can be respectively divided and analyzed as models to be processed until the configuration information screening results of each processing unit in the audio processing model are determined. Then, according to the configuration information screening results of each processing unit, the model configuration information combination of the audio processing model is determined, and the target model configuration information combination suitable for the audio processing model is screened out as the model configuration information analysis result.

[0160] Based on the above embodiment, this embodiment further provides a data processing device, referring to Figure 11 , specifically including the following modules:

[0161] The sub-model obtaining module 1102 is used to divide the data processing model to determine the first sub-model and the second sub-model, the output data of the first sub-model is associated with the input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit.

[0162] The configuration combination obtaining module 1104 is configured to determine first configuration information of the first sub-model and second configuration information of the second sub-model, and determine a configuration information combination.

[0163] The screening result obtaining module 1106 is used to determine the data processing time corresponding to each configuration information combination, and screen out the target configuration information combination as the configuration information screening result of the first sub-model and the second sub-model.

[0164] The iterative result acquisition module 1108 is used to segment and analyze the first sub-model and the second sub-model as models to be processed until the configuration information screening result of each processing unit is determined.

[0165] The analysis result acquisition module 1110 is used to determine the model configuration information analysis result of the data processing model based on the configuration information screening result of each processing unit in the data processing model.

[0166] In an embodiment of the present application, the data processing model can be segmented to obtain sub-models, and then the first sub-model and the second sub-model in the sub-model can be determined according to the order of data processing between the sub-models, and then the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination. Then the data processing time corresponding to the configuration information combination is obtained, wherein the data processing time includes the time consumed by the sub-model for data processing according to the corresponding configuration information and the time consumed by data conversion between sub-models. Then, based on the data processing time, a target configuration information combination suitable for the data processing model can be determined as a configuration information screening result. Afterwards, the first sub-model and the second sub-model can be segmented and analyzed as models to be processed respectively until the configuration information screening results of each processing unit in the data processing model are determined. Then, according to the configuration information screening results of each processing unit, the model configuration information combination of the data processing model is determined, and the target model configuration information combination suitable for the data processing model is screened out as a model configuration information analysis result.

[0167] Based on the above embodiment, this embodiment further provides a data processing device, referring to Figure 12 , specifically including the following modules:

[0168] The sub-model acquisition module 1202 is used to segment the data processing model to determine a first sub-model and a second sub-model, where the output data of the first sub-model is associated with the input data of the second sub-model.

[0169] The configuration information combination acquisition module 1204 is configured to determine first configuration information of the first sub-model and second configuration information of the second sub-model, and determine a configuration information combination.

[0170] The screening result collection module 1206 is used to determine the data processing time corresponding to each configuration information combination, and screen out the target configuration information combination as the configuration information screening result of the first sub-model and the second sub-model.

[0171] The optimization result collection module 1208 is used to perform a first optimization process on the first sub-model and a second optimization process on the second sub-model.

[0172] In the embodiment of the present application, the first optimization process and the second optimization process can be the same or different. Specifically, in the embodiment of the present application, the optimization process of the sub-model can be optimized in the same manner as the optimization method of dividing the data processing model and performing configuration analysis, or it can be optimized by performing configuration analysis on the overall operator of the sub-model, without limitation. In this embodiment, the data processing model can be divided into sub-models, and then the first sub-model and the second sub-model in the sub-model can be determined according to the order of data processing between the sub-models. Then, the first configuration information of the first sub-model and the second configuration information of the second sub-model can be determined as a configuration information combination. Then, the data processing time corresponding to the configuration information combination is obtained, where the data processing time includes the time consumed by the sub-model for data processing according to the corresponding configuration information and the time consumed by data conversion between the sub-models. Then, based on the data processing time, a target configuration information combination suitable for the data processing model can be determined as the configuration information screening result. Afterwards, the first sub-model and the second sub-model can be further optimized using the same or different optimization processing methods.

[0173] An embodiment of the present application further provides a non-volatile readable storage medium, which stores one or more modules (programs). When the one or more modules are applied to a device, the device can execute instructions (instructions) of each method step in the embodiment of the present application.

[0174] The present application provides one or more machine-readable media having instructions stored thereon, which, when executed by one or more processors, cause an electronic device to perform one or more of the methods described in the above embodiments. In the present application, the electronic device includes a server, a terminal device, and the like.

[0175] The embodiments of the present disclosure may be implemented as a device configured as desired using any appropriate hardware, firmware, software, or any combination thereof, and the device may include electronic devices such as a server (cluster), a terminal, etc. Figure 13 An exemplary apparatus 1300 that can be used to implement various embodiments described in this application is schematically illustrated.

[0176] For one embodiment, Figure 13An exemplary apparatus 1300 is shown having one or more processors 1302, a control module (chip set) 1304 coupled to at least one of the processor(s) 1302, a memory 1306 coupled to the control module 1304, a non-volatile memory (NVM) / storage device 1308 coupled to the control module 1304, one or more input / output devices 1310 coupled to the control module 1304, and a network interface 1312 coupled to the control module 1304.

[0177] The processor 1302 may include one or more single-core or multi-core processors, and the processor 1302 may include any combination of general-purpose processors or dedicated processors (e.g., graphics processors, application processors, baseband processors, etc.). In some embodiments, the apparatus 1300 can serve as a server, terminal, or other device described in the embodiments of the present application.

[0178] In some embodiments, the apparatus 1300 may include one or more computer-readable media (e.g., memory 1306 or NVM / storage 1308) having instructions 1314 and one or more processors 1302 configured in conjunction with the one or more computer-readable media to execute the instructions 1314 to implement a module to perform the actions described in the present disclosure.

[0179] For one embodiment, the control module 1304 may include any suitable interface controller to provide any suitable interface to at least one of the processor(s) 1302 and / or any suitable device or component in communication with the control module 1304 .

[0180] The control module 1304 may include a memory controller module to provide an interface to the memory 1306. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0181] The memory 1306 can be used, for example, to load and store data and / or instructions 1314 for the device 1300. For one embodiment, the memory 1306 can include any suitable volatile memory, such as a suitable DRAM. In some embodiments, the memory 1306 can include double data rate type four synchronous dynamic random access memory (DDR4 SDRAM).

[0182] For one embodiment, control module 1304 may include one or more input / output controllers to provide interfaces to NVM / storage device 1308 and input / output device(s) 1310 .

[0183] For example, NVM / storage 1308 may be used to store data and / or instructions 1314. NVM / storage 1308 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable non-volatile storage device(s) (e.g., one or more hard disk drives (HDDs), one or more compact disk (CD) drives, and / or one or more digital versatile disk (DVD) drives).

[0184] NVM / storage device 1308 may include storage resources that are part of the device on which apparatus 1300 is installed, or it may be accessible to the device without being part of the device. For example, NVM / storage device 1308 may be accessible over a network via input / output device(s) 1310.

[0185] (One or more) input / output devices 1310 may provide an interface for apparatus 1300 to communicate with any other appropriate devices. Input / output devices 1310 may include communication components, audio components, sensor components, etc. Network interface 1312 may provide an interface for apparatus 1300 to communicate via one or more networks. Apparatus 1300 may wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, for example, accessing a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof for wireless communication.

[0186] For one embodiment, at least one of the processor(s) 1302 may be packaged together with the logic of one or more controllers of the control module 1304 (e.g., a memory controller module). For one embodiment, at least one of the processor(s) 1302 may be packaged together with the logic of one or more controllers of the control module 1304 to form a system-in-package (SiP). For one embodiment, at least one of the processor(s) 1302 may be integrated on the same die with the logic of one or more controllers of the control module 1304. For one embodiment, at least one of the processor(s) 1302 may be integrated on the same die with the logic of one or more controllers of the control module 1304 to form a system-on-chip (SoC).

[0187] In various embodiments, the apparatus 1300 may be, but is not limited to, a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). In various embodiments, the apparatus 1300 may have more or fewer components and / or a different architecture. For example, in some embodiments, the apparatus 1300 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touchscreen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0188] Among them, the main control chip can be used as a processor or control module in the detection device, sensor data, location information, etc. are stored in the memory or NVM / storage device, the sensor group can be used as an input / output device, and the communication interface may include a network interface.

[0189] An embodiment of the present application further provides an electronic device, comprising: a processor; and a memory on which executable code is stored. When the executable code is executed, the processor executes one or more methods described in the embodiments of the present application.

[0190] The embodiments of the present application further provide one or more machine-readable media on which executable codes are stored. When the executable codes are executed, the processor executes one or more methods described in the embodiments of the present application.

[0191] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0192] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0193] The present application embodiment is described with reference to the flow chart and / or block diagram of the method, terminal device (system), and computer program product according to the embodiment of the present application. It should be understood that each process and / or box in the flow chart and / or block diagram and the combination of the process and / or box in the flow chart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device produce a device for realizing the function specified in one process or multiple processes and / or one box or multiple boxes of the flow chart.

[0194] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0195] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce computer-implemented processing, so that the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0196] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0197] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0198] The above is a detailed introduction to a data processing method, a data processing device, an electronic device and a storage medium provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.

Claims

1. A data processing method, characterized in that: The method includes: Splitting the data processing model to determine a first sub-model and a second sub-model, where output data of the first sub-model is associated with input data of the second sub-model, the data processing model including an image processing model and / or an audio processing model; determining first configuration information of the first sub-model and second configuration information of the second sub-model as a configuration information combination; The data processing time corresponding to each configuration information combination is determined, and at least one set of target configuration information combinations is screened out as the configuration information screening results of the first sub-model and the second sub-model.

2. The method according to claim 1, characterized in that The first sub-model and the second sub-model are composed of processing units, and the method further includes: The first sub-model and the second sub-model are divided and analyzed as models to be processed until the configuration information screening results of each processing unit are determined; The model configuration information analysis result of the data processing model is determined based on the configuration information screening result of each processing unit in the data processing model.

3. The method according to claim 1, characterized in that Determining the first configuration information of the first sub-model and the second configuration information of the second sub-model as a configuration information combination includes at least one of the following steps: determining a first output data format of the first sub-model and a second input data format of the second sub-model as a configuration information combination; A first processor of the first sub-model and a second processor of the second sub-model are determined as a configuration information combination.

4. The method according to claim 1, wherein The dividing the data processing model to determine the first sub-model and the second sub-model includes: Split the data processing model to obtain the segmented sub-models; Topological sorting information corresponding to the data processing model is obtained, and the segmented sub-model is divided to obtain a first sub-model and a second sub-model, wherein the topological sorting information is determined according to the structure of the data processing model.

5. The method according to claim 1, wherein Determining the data processing duration corresponding to each configuration information combination includes: Determining a first duration for the first sub-model to perform data processing according to the first configuration information; Determining a second duration for the second sub-model to perform data processing according to the second configuration information; Determining a third duration based on the first configuration information and the second configuration information; A data processing duration is determined based on the first duration, the second duration, and the third duration.

6. The method according to claim 1, characterized in that The step of screening out at least one set of target configuration information combinations includes: Sort the configuration information combinations according to the data processing time; According to the sorted configuration information combinations, at least one group of target configuration information combinations that meets the preset screening conditions is screened out.

7. The method according to claim 6, characterized in that The step of screening out at least one set of target configuration information combinations that meet preset screening conditions includes: At least one set of target configuration information combinations is screened out according to a preset screening ratio.

8. The method according to claim 6, characterized in that The step of screening out at least one set of target configuration information combinations that meet preset screening conditions includes: Determine a first configuration information combination that is ranked to meet the ranking condition; Determining a screening threshold based on a data processing time of the first configuration information combination; At least one set of target configuration information combinations that meets the screening threshold is filtered out.

9. The method according to claim 2, characterized in that The step of screening the configuration information of each processing unit in the data processing model to determine the model configuration information analysis result of the data processing model includes: According to the configuration information screening results of each processing unit, the unit configuration information corresponding to each processing unit is determined to determine the model configuration information combination of the data processing model; Determine the model processing time corresponding to each model configuration information combination, and filter out the target model configuration information combination as the model configuration information analysis result.

10. The method according to claim 9, characterized in that Determining the model processing time corresponding to each model configuration information combination includes: Configure the data processing model according to the model configuration information combination; Perform data processing based on the configured data processing model and determine the model processing time.

11. The method according to claim 9, characterized in that The step of screening out the target model configuration information combination includes: Sort the model configuration information combinations according to the model processing time corresponding to each model configuration information combination; According to the sorted model configuration information combinations, the target model configuration information combination is screened out.

12. The method according to claim 11, characterized in that The step of screening out a target model configuration information combination based on the sorted model configuration information combination includes: Based on the sorted model configuration information combinations, at least one set of model configuration information combinations to be confirmed is screened out and issued; Receive feedback information on the model configuration information combination to be confirmed, and determine the target model configuration information combination.

13. A data processing method, characterized in that: include: Segmenting the image processing model to determine a first sub-model and a second sub-model, where output data of the first sub-model is associated with input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit; Get the image configuration information table corresponding to the image processing model; Determining, according to the image configuration information table, first configuration information of the first sub-model and second configuration information of the second sub-model as a configuration information combination; Determine the data processing time corresponding to each configuration information combination, and screen out at least one set of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model; The first sub-model and the second sub-model are divided and analyzed as models to be processed until the configuration information screening result of each processing unit is determined; The model configuration information analysis result of the data processing model is determined based on the configuration information screening result of each processing unit in the image processing model.

14. A data processing method, characterized in that: include: Splitting the audio processing model to determine a first sub-model and a second sub-model, where output data of the first sub-model is associated with input data of the second sub-model, and the first sub-model and the second sub-model are composed of at least one processing unit; Get the audio configuration information table corresponding to the audio processing model; Determining, according to the audio configuration information table, first configuration information of the first sub-model and second configuration information of the second sub-model as a configuration information combination; Determine the data processing time corresponding to each configuration information combination, and screen out at least one set of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model; The first sub-model and the second sub-model are divided and analyzed as models to be processed until the configuration information screening result of each processing unit is determined; The model configuration information analysis result of the data processing model is determined based on the configuration information screening result of each processing unit in the audio processing model.

15. A data processing method, characterized in that: include: Splitting a data processing model to determine a first sub-model and a second sub-model, where output data of the first sub-model is associated with input data of the second sub-model, the first sub-model and the second sub-model are composed of at least one processing unit, and the data processing model includes an image processing model and / or an audio processing model; Determine first configuration information of the first sub-model and second configuration information of the second sub-model, and determine a configuration information combination; Determine the data processing time corresponding to each configuration information combination, and filter out the target configuration information combination as the configuration information screening result of the first sub-model and the second sub-model; The first sub-model and the second sub-model are divided and analyzed as models to be processed until the configuration information screening result of each processing unit is determined; The model configuration information analysis result of the data processing model is determined based on the configuration information screening result of each processing unit in the data processing model.

16. A data processing method, characterized in that: include: Splitting the data processing model to determine a first sub-model and a second sub-model, where output data of the first sub-model is associated with input data of the second sub-model, the data processing model including an image processing model and / or an audio processing model; Determine first configuration information of the first sub-model and second configuration information of the second sub-model, and determine a configuration information combination; Determine the data processing time corresponding to each configuration information combination, and filter out the target configuration information combination as the configuration information screening result of the first sub-model and the second sub-model; A first optimization process is performed on the first sub-model, and a second optimization process is performed on the second sub-model.

17. A data processing device, characterized in that: The device comprises: a sub-model acquisition module, configured to segment the data processing model to determine a first sub-model and a second sub-model, wherein output data of the first sub-model is associated with input data of the second sub-model, and the data processing model includes an image processing model and / or an audio processing model; a configuration information combination acquisition module, configured to determine first configuration information of a first sub-model and second configuration information of a second sub-model as a configuration information combination; The screening result acquisition module is used to determine the data processing time corresponding to each configuration information combination, and screen out at least one set of target configuration information combinations as the configuration information screening results of the first sub-model and the second sub-model.

18. An electronic device, characterized in that: include: processor; and A memory having executable codes stored thereon, which, when executed, causes the processor to perform the method according to any one of claims 1 to 16.

19. One or more machine-readable media having executable codes stored thereon, which, when executed, cause a processor to perform the method according to any one of claims 1 to 16.

Images