Model information sending method and device, data prediction method and device, server and medium

By performing operator type analysis and compression processing of dynamic link library on deep learning model files, the problem of large model storage space is solved and more efficient storage and application efficiency is achieved.

CN119088582APending Publication Date: 2024-12-06BEIJING WODONG TIANJUN INFORMATION TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310664508.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-06-06
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

When the prior art sends a deep learning model to the end-side device, the model occupies a large storage space, resulting in high storage space consumption of the device and affects the application efficiency of the model.

Method used

By parsing the operator type of the model file, a dynamic link library is generated, and compressed, it can reduce duplicate content, generate a compressed dynamic link library, and send it to the model application.

Benefits of technology

It reduces the storage size of model files, reduces the storage space requirements, and improves the application efficiency and sending speed of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088582A_ABST
    Figure CN119088582A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses a model information sending method and device, a data prediction method and device, a server and a medium. A specific embodiment of the method comprises the following steps: performing operator type analysis on a model file to determine the operator type of each operator in a model corresponding to the model file so as to obtain an operator type set; generating a dynamic link library according to the operator type set; compressing the dynamic link library to obtain a compressed dynamic link library, and taking the compressed dynamic link library as a target dynamic link library; and sending the target dynamic link library and the model file to a model application end. The implementation mode is related to artificial intelligence, the model storage size of the model corresponding to the model file is reduced, and model application of a subsequent model application end is facilitated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a model information sending method, a data prediction method, a device, a server, and a medium. Background Art

[0002] With the development of artificial intelligence, the miniaturization of deep learning models and the enhanced computing power of end-side devices (i.e., model application end), it has become possible to deploy artificial intelligence models on the end-side. When sending the model, the method usually adopted is: first, in response to receiving a model request for a business scenario sent by the model application end, send the model for the business scenario to the model application end.

[0003] However, the inventors have found that when the model is sent in the above manner, the following technical problems often occur:

[0004] The model takes up a lot of storage space, which results in the model application end consuming a high amount of storage space.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0006] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0007] Some embodiments of the present disclosure propose a model information sending method, a data prediction method, an apparatus, a server and a medium to solve the technical problems mentioned in the above background technology section.

[0008] In a first aspect, some embodiments of the present disclosure provide a method for sending model information, the method comprising: performing operator type parsing on a model file to determine the operator type of each operator in a model corresponding to the model file, and obtaining an operator type set; generating a dynamic link library based on the operator type set; compressing the dynamic link library to obtain a compressed dynamic link library as a target dynamic link library; and sending the target dynamic link library and the model file to a model application end.

[0009] Optionally, the dynamic link library includes: function information. And compressing the dynamic link library to obtain a compressed dynamic link library includes: deduplicating a function set corresponding to the function information to obtain a deduplicated dynamic link library as the compressed dynamic link library.

[0010] Optionally, the dynamic link library includes: function information and source file information. And compressing the dynamic link library to obtain a compressed dynamic link library includes: using a preset compilation optimization method to compile and optimize the source file corresponding to the source file information and the function set corresponding to the function information to obtain an optimized dynamic link library as the compressed dynamic link library.

[0011] In the second aspect, some embodiments of the present disclosure provide a model information sending device, the device comprising: a first parsing unit, configured to perform operator type parsing on a model file to determine the operator type of each operator in the model corresponding to the above model file, and obtain an operator type set; a generating unit, configured to generate a dynamic link library according to the above operator type set. Among them, the calling function in the above dynamic link library is the operator calling function corresponding to the operator type in the above operator type set; a compression processing unit, configured to perform compression processing on the above dynamic link library, and obtain a compressed dynamic link library as a target dynamic link library; a sending unit, configured to send the above target dynamic link library and the above model file to the model application end.

[0012] Optionally, the dynamic link library includes function information, and the compression processing unit may be configured to perform deduplication processing on the function set corresponding to the function information to obtain a deduplication dynamic link library as the compressed dynamic link library.

[0013] Optionally, the dynamic link library includes function information and source file information, and the compression processing unit can be configured to compile and optimize the source file corresponding to the source file information and the function set corresponding to the function information using a preset compilation optimization method to obtain an optimized dynamic link library as the compressed dynamic link library.

[0014] In a third aspect, some embodiments of the present disclosure provide a data prediction method, the method comprising: performing data preprocessing on the acquired prediction data to obtain processed data. Receive a model file and a target dynamic link library sent by a server. Perform file parsing on the above model file to determine the parameter information corresponding to each operator in the model corresponding to the above model file and the connection relationship between each operator, respectively as an operator parameter information set and operator connection relationship information. Take the above processed data as input, and perform data prediction based on operator reasoning according to the above operator connection relationship information, the above operator parameter information set and the above target dynamic link library to generate a prediction result.

[0015] Optionally, the above-mentioned processed data is used as input, and according to the above-mentioned operator connection relationship information, the above-mentioned operator parameter information set and the above-mentioned target dynamic link library, data prediction based on operator reasoning is performed to generate a prediction result, including: according to the operator order corresponding to each target operator information chain, each operator information corresponding to at least one target operator information chain is added to the operator processing queue in turn to obtain an added operator processing queue, wherein the front-layer operator information set corresponding to the operator information of the first target position in each target operator information chain in the above-mentioned at least one target operator information chain is empty; for each target operator information chain in the at least one target operator information chain, the following result generation steps are performed: according to the above-mentioned operator parameter information set and the above-mentioned target dynamic link library, each operator information associated with the target operator information chain in the added operator processing queue is executed in turn to obtain an operator execution result; and a prediction result is generated according to the obtained at least one operator execution result.

[0016] Optionally, the above-mentioned operation of sequentially executing each operator information associated with the target operator information chain in the added operator processing queue according to the operator parameter information set and the target dynamic link library to obtain the operator execution result includes: for the operator information associated with the target operator information chain in the added operator processing queue, executing the following result generation sub-steps: using the operator parameter information set and the target dynamic link library, determining the calling function corresponding to the operator information; in response to determining that the operator information has at least one corresponding historical previous layer operator information, inputting at least one previous layer prediction result corresponding to the operator information into the above-mentioned calling function; , obtaining the current prediction result; removing the operator information from the corresponding target operator processing queue to obtain a removed operator processing queue, wherein the target operator processing queue does not include at least one operator information in the target operator information chain that is located before the operator information; in response to determining that the operator information is the operator information at the second target position in the associated target operator information chain, determining the current prediction result as the operator execution result; in response to determining that the removed operator processing queue is not empty, determining the next operator information in the target operator information chain that corresponds to the operator information as the operator information, and executing the result generation sub-step again.

[0017] Optionally, before removing the operator information from the corresponding target operator processing queue to obtain the removed operator processing queue, the method further includes: in response to determining that the operator information does not have corresponding at least one historical previous layer operator information, inputting the processed data into the calling function to obtain the current prediction result.

[0018] Optionally, generating a prediction result based on at least one operator execution result obtained includes: in response to determining that a back-layer operator information set corresponding to each target operator information in at least one target operator information is empty, determining at least one operator execution result corresponding to at least one target operator information as the prediction result, wherein the target operator information in the at least one target operator information is the operator information of the second target position in the target operator information chain in at least one target operator information chain.

[0019] Optionally, in response to determining that the above-mentioned operator processing queue is empty after removal, and there is target operator information whose corresponding rear-layer operator information set is not empty in at least one target operator information, the non-empty rear-layer operator information set is determined as the target rear-layer operator information set to obtain at least one target rear-layer operator information set; for each target rear-layer operator information in the above-mentioned at least one target rear-layer operator information set, the following queue generation step is performed: the corresponding target operator information is removed from the front-layer operator information set corresponding to the target rear-layer operator information to obtain the removed front-layer operator information set; in response to determining that the removed front-layer operator information set is empty, the operator information chain corresponding to the target rear-layer operator information is determined as the target operator information chain; according to the operator sequence corresponding to each target operator information chain in the obtained at least one target operator information chain, each operator information corresponding to the at least one target operator information chain is added to the post-removal operator processing queue in turn to obtain the added operator processing queue, so as to perform the above-mentioned result generation step again.

[0020] In a fourth aspect, some embodiments of the present disclosure provide a data prediction device, which includes: performing data preprocessing on the acquired prediction data to obtain processed data; receiving a model file and a target dynamic link library sent by a server; performing file parsing on the above model file to determine the parameter information corresponding to each operator in the model corresponding to the above model file and the connection relationship between each operator, respectively as an operator parameter information set and an operator connection relationship information; taking the above processed data as input, and performing data prediction based on operator reasoning according to the above operator connection relationship information, the above operator parameter information set and the above target dynamic link library to generate a prediction result.

[0021] Optionally, the execution unit may be configured to: take the processed data as input, and perform data prediction based on operator reasoning according to the operator connection relationship information, the operator parameter information set and the target dynamic link library to generate a prediction result, including: according to the operator order corresponding to each target operator information chain, add each operator information corresponding to at least one target operator information chain to the operator processing queue in turn to obtain an added operator processing queue, wherein the front-layer operator information set corresponding to the operator information at the first target position in each target operator information chain is empty; for each target operator information chain in at least one target operator information chain, perform the following result generation steps: according to the operator parameter information set and the target dynamic link library, execute each operator information associated with the target operator information chain in the added operator processing queue in turn to obtain an operator execution result; and generate a prediction result based on the obtained at least one operator execution result.

[0022] Optionally, the execution unit may be configured to: for the operator information in the added operator processing queue that is associated with the target operator information chain, perform the following result generation sub-steps: determine the calling function corresponding to the operator information using the operator parameter information set and the target dynamic link library; in response to determining that the operator information has at least one corresponding historical previous layer operator information, input at least one previous layer prediction result corresponding to the operator information into the calling function to obtain a current prediction result; remove the operator information from the corresponding target operator processing queue to obtain a removed operator processing queue, wherein the target operator processing queue does not include at least one operator information in the target operator information chain that is located before the operator information; in response to determining that the operator information is the operator information at the second target position in the associated target operator information chain, determine the current prediction result as the operator execution result; in response to determining that the removed operator processing queue is not empty, determine the next operator information corresponding to the operator information in the target operator information chain as the operator information, and perform the result generation sub-step again.

[0023] Optionally, the execution unit may be configured to: in response to determining that the operator information does not have corresponding at least one historical previous layer operator information, input the processed data into the calling function to obtain a current prediction result.

[0024] Optionally, the execution unit may be configured to: in response to determining that a set of back-layer operator information corresponding to each target operator information in at least one target operator information is empty, determine at least one operator execution result corresponding to at least one target operator information as a prediction result, wherein the target operator information in the at least one target operator information is the operator information of the second target position in the target operator information chain in at least one target operator information chain.

[0025] Optionally, the execution unit may be configured to: in response to determining that the operator processing queue after removal is empty, and there is target operator information whose corresponding rear-layer operator information set is not empty in at least one target operator information, determine the non-empty rear-layer operator information set as the target rear-layer operator information set, and obtain at least one target rear-layer operator information set; for each target rear-layer operator information in the at least one target rear-layer operator information set, perform the following queue generation step: remove the corresponding target operator information from the front-layer operator information set corresponding to the target rear-layer operator information, and obtain the front-layer operator information set after removal; in response to determining that the front-layer operator information set after removal is empty, determine the operator information chain corresponding to the target rear-layer operator information as the target operator information chain; according to the operator sequence corresponding to each target operator information chain in the obtained at least one target operator information chain, add each operator information corresponding to the at least one target operator information chain to the operator processing queue after removal in turn, and obtain the operator processing queue after addition, so as to perform the result generation step again.

[0026] In a fifth aspect, some embodiments of the present disclosure provide an electronic device, comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first and third aspects above.

[0027] In a sixth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method described in any one of the implementation modes of the first and third aspects is implemented.

[0028] In a seventh aspect, some embodiments of the present disclosure provide a computer program product, including a computer program, which, when executed by a processor, implements the method described in any implementation of the first and third aspects above.

[0029] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the model information sending method of some embodiments of the present disclosure, the model storage size of the model corresponding to the model file is reduced, which is convenient for the model application end of the subsequent model application. Specifically, the reason why the model file corresponding to the model is inconvenient to use is that the model occupies a large storage space, resulting in the model application end model needing to consume a high storage space. Based on this, the model information sending method of some embodiments of the present disclosure first performs operator type parsing on the model file to determine the operator type of each operator in the model corresponding to the above model file, and obtains an operator type set. Here, by parsing the operator type set, it is convenient to determine the dynamic link library corresponding to the operator type set in the subsequent. Then, according to the above operator type set, a dynamic link library is generated. Among them, a dynamic link library is generated by the corresponding function of each operator type in the operator type set, so as to support the call of the function required by the model in the subsequent support. Then, the above dynamic link library is compressed to obtain a compressed dynamic link library as a target dynamic link library. Here, compressing the dynamic link library can reduce the repeated content included in the dynamic link library and reduce the storage space of the dynamic link library, so that the subsequent model application end can directly reason and use it. In addition, due to the reduction of the storage space of the dynamic link library, the sending speed of the dynamic link library can also be improved. Finally, the target dynamic link library and the model file are sent to the model application end, so that the model application end can realize data prediction based on the target dynamic link library and the model file. In summary, the dynamic link library is compressed, and the size of the model storage corresponding to the model file is reduced, so that the model application end can use a lower storage space to store the model, and it is also convenient to use the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0031] Figure 1 is a schematic diagram of an application scenario of the model information sending method according to the present disclosure;

[0032] Figure 2 is a flow chart of some embodiments of the model information sending method according to the present disclosure;

[0033] Figure 3 are flow charts of other embodiments of the model information sending method according to the present disclosure;

[0034] Figure 4 is a flow chart of some embodiments of the data prediction method according to the present disclosure;

[0035] Figure 5 is a schematic diagram of a target operator information chain according to the data prediction method disclosed herein;

[0036] Figure 6 is a schematic diagram of the structure of some embodiments of the model information sending device according to the present disclosure;

[0037] Figure 7 is a schematic diagram of the structure of some embodiments of the data prediction device according to the present disclosure;

[0038] Figure 8 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0039] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0040] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0041] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0042] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one".

[0043] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0044] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0045] Figure 1 It is a schematic diagram of an application scenario of the model information sending method according to some embodiments of the present disclosure.

[0046] exist Figure 1In the application scenario, the server 100 can first parse the model file 101 to obtain the operator type set 102. In this application scenario, the operator type set 102 can include multiple operator types. For example, the operator type can be a convolution operator type. Secondly, according to the operator type set 102, a dynamic link library 103 is generated. Then, the dynamic link library 103 is compressed to obtain a target dynamic link library 104. Finally, the target dynamic link library 104 and the model file 101 are sent to the model application end 105.

[0047] It should be noted that the above-mentioned server 100 can be hardware or software. When the server is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or it can be implemented as a single server or a single terminal device. When the server is embodied as software, it can be installed in the hardware devices listed above. It can be implemented as multiple software or software modules for providing distributed services, for example, or it can be implemented as a single software or software module. No specific limitation is made here.

[0048] It should be understood that Figure 1 The number of servers in is only for illustration. Any number of servers may be provided according to implementation requirements.

[0049] Continue to refer Figure 2 , shows a process 200 of some embodiments of the model information sending method according to the present disclosure. The process 200 of the model information sending method includes the following steps:

[0050] Step 201 , parse the model file to determine the operator type of each operator in the model corresponding to the model file, and obtain an operator type set.

[0051] In some embodiments, the execution subject of the model information sending method (for example, server 101) can perform operator type analysis on the model file to determine the operator type of each operator in the model corresponding to the above model file, and obtain an operator type set. Among them, the model file can be a file storing the relevant content of the model corresponding to the target application scenario. In practice, for the above target application scenario is a face recognition scenario, the model file can be a file storing the face recognition model. The above operator type can be a category that characterizes the operator. In practice, the operator type and can be an operator category set of all operators used by the model corresponding to the model file. For the model using multiple identical operators, there may be repeated operator types between the operator types included in the operator type set here. For example, the operator type can be a convolution operator type, and can also be a Fourier transform operator type. The above model file can be a model file that completes the corresponding application scenario. For example, the above application scenario can be a face recognition scenario, and the model corresponding to the above model file can be a face recognition model. The above model file may include but is not limited to at least one of the following: the number of operators in the corresponding model, and the operator type of the operator in the corresponding model. The operator type in the operator type set may be a type of operator. The operator type in the operator type set may be an operator category type of operator. For example, the operator type in the operator type set may be a convolution operator type.

[0052] Step 202: Generate a dynamic link library according to the operator type set.

[0053] In some embodiments, the execution subject may generate a dynamic link library according to the operator type set. The dynamic link library may be a library that supports calling functions. For example, the functions supported by the dynamic link library may include functions for face recognition scenarios. In practice, the dynamic link library is also a link library for application scenarios.

[0054] It should be noted that by calling relevant functions in the dynamic link library, the model file corresponding to the model can be restored.

[0055] In practice, the above execution entity generates a dynamic link library based on the operator type set, including:

[0056] In the first step, the execution subject may determine the calling function corresponding to each operator type in the operator type set to obtain a calling function set.

[0057] In the second step, the execution subject may package the definition function information and source files corresponding to each calling function in the calling function set to generate a dynamic link library. The definition function information may be information on the definition of the calling function. The source file may be a source file using the calling function. For example, the definition function information may be information on the number of definitions of the calling function.

[0058] Step 203, compressing the dynamic link library to obtain a compressed dynamic link library as a target dynamic link library.

[0059] In some embodiments, the execution entity may compress the dynamic link library to obtain a compressed dynamic link library as a target dynamic link library.

[0060] In practice, the execution subject compresses the dynamic link library to obtain a compressed dynamic link library, including:

[0061] The dynamic link library is compressed using LZMA (Lempel-Ziv-Markov chain-Algorithm) to obtain a compressed dynamic link library.

[0062] In some optional implementations of some embodiments, the dynamic link library includes: function information and source file information; and

[0063] The execution subject may compress the dynamic link library to obtain a compressed dynamic link library, including:

[0064] The execution subject may utilize a preset compilation optimization method to compile and optimize the source file corresponding to the source file information and the function set corresponding to the function information, and obtain an optimized dynamic link library as a compressed dynamic link library. The source file information may be information of a source file. The source file may be a source file that uses a calling function. For example, the source file information may be identification information of a source file. The preset compilation optimization method may be a method for optimizing the function set corresponding to the function information and the source file corresponding to the source file information. For example, the preset compilation optimization method may be a Link Time Optimization (LTO) method.

[0065] Step 204, sending the target dynamic link library and model file to the model application end.

[0066] In some embodiments, the execution subject may send the target dynamic link library and the model file to a model application end, wherein the model application end may be an application device that uses the target dynamic link library and the model file to perform prediction.

[0067] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the model information sending method of some embodiments of the present disclosure, the model storage size of the model corresponding to the model file is reduced, which is convenient for the model application end of the subsequent model application. Specifically, the reason why the model file corresponding to the model is inconvenient to use is that the model occupies a large storage space, resulting in the model application end model needing to consume a high storage space. Based on this, the model information sending method of some embodiments of the present disclosure first performs operator type parsing on the model file to determine the operator type of each operator in the model corresponding to the above model file, and obtains an operator type set. Here, by parsing the operator type set, it is convenient to determine the dynamic link library corresponding to the operator type set in the subsequent. Then, according to the above operator type set, a dynamic link library is generated. Among them, a dynamic link library is generated by the corresponding function of each operator type in the operator type set, so as to support the call of the function required by the model in the subsequent support. Then, the above dynamic link library is compressed to obtain a compressed dynamic link library as a target dynamic link library. Here, compressing the dynamic link library can reduce the repeated content included in the dynamic link library and reduce the storage space of the dynamic link library, so that the subsequent model application end can directly reason and use it. In addition, due to the reduction of the storage space of the dynamic link library, the sending speed of the dynamic link library can also be improved. Finally, the target dynamic link library and the model file are sent to the model application end, so that the model application end can realize data prediction based on the target dynamic link library and the model file. In summary, the dynamic link library is compressed, and the size of the model storage corresponding to the model file is reduced, so that the model application end can use a lower storage space to store the model, and it is also convenient to use the model.

[0068] Continue to refer Figure 3 , shows a process 300 of another embodiment of the method for sending model information according to the present disclosure. The process 300 of the method for sending model information includes the following steps:

[0069] Step 301 , perform operator type parsing on the model file to determine the operator type of each operator in the model corresponding to the model file, and obtain an operator type set.

[0070] Step 302: Generate a dynamic link library according to the above operator type set.

[0071] Step 303, deduplicate the function set corresponding to the above function information to obtain a deduplicated dynamic link library as a compressed dynamic link library.

[0072] In some embodiments, an execution entity (e.g., Figure 1The server 101 shown) can perform deduplication processing on the function set corresponding to the above function information to obtain a deduplicated dynamic link library as a compressed dynamic link library. Among them, there is a one-to-one correspondence between the functions in the above function set and the operator types in the operator type set. The above dynamic link library includes: function information. Among them, the above function information can be information for calling a calling function. The above function information can include but is not limited to at least one of the following: the dynamic link library includes type information of each function in a plurality of functions, and the dynamic link library includes quantity information of functions in a plurality of functions. For example, the above function information can be type information of a loss function.

[0073] Step 304: Send the target dynamic link library and the model file to the model application end.

[0074] In some embodiments, the specific implementation of steps 301-302 and 304 and the technical effects thereof can be referred to in Figure 3 The steps 201-202 and 204 in the corresponding embodiment are not described in detail here.

[0075] from Figure 3 It can be seen that Figure 2 Compared with the description of some corresponding embodiments, Figure 3 The process 300 of the model information sending method in some corresponding embodiments can efficiently and quickly realize the compression of the dynamic link library to reduce the amount of subsequent model information sent, so that the subsequent model application end can use a smaller model storage space to realize subsequent model applications.

[0076] Further references Figure 4 The process 400 of some embodiments of the data prediction method according to the present disclosure is shown. The process 400 of the data prediction method includes the following steps:

[0077] Step 401, preprocessing the acquired prediction data to obtain processed data.

[0078] In some embodiments, the execution subject of data prediction (e.g., the model application end) performs data preprocessing on the acquired prediction data to obtain processed data. The data preprocessing may include but is not limited to at least one of the following: data cleaning and data transformation. The prediction data may be input data for subsequent data prediction.

[0079] Step 402: Receive the model file and target dynamic link library sent by the server.

[0080] In some embodiments, the execution subject may receive a model file and a target dynamic link library sent by a server. The model file may be a model file that completes a corresponding application scenario. The dynamic link library may be a library that supports calling a function.

[0081] Step 403 , parsing the model file to determine parameter information corresponding to each operator in the model corresponding to the model file and the connection relationship between each operator as the operator parameter information set and operator connection relationship information respectively.

[0082] In some embodiments, the execution subject may parse the above model file to determine the parameter information corresponding to each operator in the model corresponding to the above model file and the connection relationship between each operator, respectively as an operator parameter information set and operator connection relationship information. Among them, the operator parameter information in the above operator parameter information set may be the parameter information of the function corresponding to the operator. The corresponding function is the function in the above target dynamic link library. The above operator connection relationship information may be the contextual relationship information between each operator in the model corresponding to the above model file. For example, in Figure 5 In the schematic diagram, the model corresponding operator information set includes: operator information V3, operator information V3, operator information V4, operator information V5, operator information V6, operator information V7. Among them, operator information V1 can be the front-layer operator information of operator information V2, and operator information V2 can be the back-layer operator information of operator information V1. For example, Figure 5 The operator information V1 in may be the information of the starting operator position. The operator information V7 may be the information of the ending operator position. The at least one subsequent layer operator information corresponding to the operator information V1 is the operator information V2 and the operator information V3. The at least one subsequent layer operator information corresponding to the operator information V4, the operator information V5 and the operator information V6 is the operator information V7. The at least one previous layer operator information corresponding to the operator information V7 is the operator information V4, the operator information V5 and the operator information V6. For example, the above-mentioned operator information is the operator information V3, and the at least one previous layer operator information is the operator information V1. The at least one subsequent layer operator is the operator information V6 and the operator information V5.

[0083] Step 404 , using the processed data as input, and performing data prediction based on operator reasoning according to the operator connection relationship information, the operator parameter information set and the target dynamic link library to generate a prediction result.

[0084] In some embodiments, the execution subject may use the processed data as input, and perform data prediction based on operator reasoning according to the operator connection relationship information, the operator parameter information set and the target dynamic link library to generate a prediction result. The prediction result may be a prediction result for the scene corresponding to the model file. For example, when the corresponding scene is animal recognition, the prediction result is the recognized animal result. For example, the animal result may be a giraffe.

[0085] In practice, the above-mentioned processed data is used as input, and according to the above-mentioned operator connection relationship information, the above-mentioned operator parameter information set and the above-mentioned target dynamic link library, data prediction based on operator reasoning is performed to generate prediction results, including:

[0086] In the first step, the execution subject may use the operator connection relationship information, the operator parameter information set and the target dynamic link library to determine the calling function corresponding to each of the multiple operator information corresponding to the operator connection relationship information, and obtain multiple calling functions.

[0087] In the second step, for each of the multiple call functions, perform the following input steps:

[0088] The first sub-step is to determine the operator information corresponding to the above-mentioned calling function as the target operator information.

[0089] The second sub-step, in response to determining that the above-mentioned target operator information is the last operator information among multiple operator information, and at least one previous layer operator corresponding to the above-mentioned target operator information has been executed, at least one previous layer prediction result corresponding to at least one previous layer operator is input into the above-mentioned calling function to obtain the target prediction result as the prediction result.

[0090] In some optional implementations of some embodiments, the above-mentioned processing data is used as input, and according to the above-mentioned operator connection relationship information, the above-mentioned operator parameter information set and the above-mentioned target dynamic link library, the data prediction based on operator reasoning is performed to generate a prediction result, which may include the following steps:

[0091] In the first step, according to the operator sequence corresponding to each target operator information chain, each operator information corresponding to at least one target operator information chain is added to the operator processing queue in turn to obtain the added operator processing queue.

[0092] Among them, the front-layer operator information set corresponding to the operator information at the first target position in each target operator information chain is empty. The above-mentioned first target position may be the position of the first operator information in the target operator information chain. The operator sequence corresponding to the above-mentioned target operator information chain may be the sequence between at least one operator information included in the above-mentioned target operator information chain. The operator processing queue may be in the form of a queue, including the operator information corresponding to each operator to be executed. According to the sequence of each operator information in the operator processing queue, the operator logic corresponding to each operator is executed in sequence. The target operator information chain may be an operator information chain in which the front-layer operator information set corresponding to the operator information at the first target position is empty. The above-mentioned target operator information chain may be a chain with the out-degree and in-degree being the target values. The front-layer operator information set may be in the form of an operator information table. That is, each operator information initially has a corresponding forward operator information table and a backward operator information table. Each forward operator information in the forward operator information table may be an operator information whose execution sequence is located before the operator information and whose backward operator is the operator information. As the number of forward operator information in the forward operator information set decreases, the forward operator executions before the representation operator information are executed in sequence. Similarly, each backward operator information in the backward operator information set is the operator information of the operator to be executed subsequently and connected to the operator information.

[0093] For example, the target value may be "1". Figure 5 , Figure 5 Four target operator information chains are shown, namely, "target operator information chain composed of operator information V1", "target operator information chain composed of operator information V2 and operator information V4", "target operator information chain composed of operator information V3 and operator information V5" and "target operator information chain composed of operator information V6 and operator information V7".

[0094] It should be noted that for the queue writing and deletion of each target operator information chain, a mutex lock is added to ensure the normal writing and deletion of operator information. In addition, for each target operator information chain, a corresponding thread will be assigned to handle the logical execution of each operator information in the operator information chain.

[0095] In practice, according to the operator order corresponding to each target operator information chain, each operator information corresponding to at least one target operator information chain is added to the operator processing queue in sequence to obtain the added operator processing queue, including:

[0096] According to the operator sequence corresponding to each target operator information chain, each operator information corresponding to at least one target operator information chain is added to the operator processing queue in sequence to obtain an added operator processing queue.

[0097] In practice, the above-mentioned front-layer operator information set can be generated by the following steps:

[0098] By using the operator connection relationship information, a front-layer operator information set of each operator information in the plurality of operator information corresponding to the operator connection relationship information is determined.

[0099] In the second step, for each target operator information chain in at least one target operator information chain, the following result generation steps are performed:

[0100] According to the operator parameter information set and the target dynamic link library, each operator information associated with the target operator information chain in the added operator processing queue is executed in sequence to obtain the operator execution result. The operator execution result may be the output result corresponding to the last operator information included in the target operator information chain. Each operator information associated with the target operator information chain may be each operator information included in the target operator information chain.

[0101] In practice, the above-mentioned operator parameter information set and the above-mentioned target dynamic link library are used to sequentially execute each operator information associated with the target operator information chain in the added operator processing queue to obtain the operator execution result, including:

[0102] The first sub-step is to determine a call function corresponding to at least one associated operator information using the operator parameter information set and the target dynamic link library, wherein the at least one associated operator information may be at least one operator information associated with the target operator information chain in the added operator processing queue.

[0103] The second sub-step is, for each associated operator information in at least one associated operator information, in response to determining that the above associated operator information is the penultimate operator information in the target operator information chain, inputting at least one previous layer prediction result corresponding to the associated operator information into the corresponding calling function to obtain the current prediction result as the operator execution result.

[0104] The third step is to generate a prediction result based on the obtained at least one operator execution result.

[0105] In practice, the above method generates a prediction result based on the obtained at least one operator execution result, including:

[0106] In response to determining that the set of subsequent operator information corresponding to each target operator information in at least one target operator information is empty, at least one operator execution result corresponding to the at least one target operator information is screened to obtain the screened operator execution result as the prediction result. The target operator information in the at least one target operator information may be the operator information at the second target position in the target operator information chain in the at least one target operator information chain.

[0107] In some optional implementations of some embodiments, the above-mentioned step of sequentially executing each operator information associated with the target operator information chain in the added operator processing queue according to the operator parameter information set and the target dynamic link library to obtain the operator execution result may include the following steps:

[0108] In the first step, for each operator information in the added operator processing queue that is associated with the above target operator information chain, the following result generation sub-steps are performed:

[0109] The first sub-step: using the above operator parameter information set and the above target dynamic link library, determine the calling function corresponding to the operator information.

[0110] As an example, first, the execution subject may determine the function corresponding to the operator information in the function set corresponding to the function information included in the target dynamic link library as the function to be called. Then, the execution subject may determine the parameters of the function to be called using the operator parameter information corresponding to the operator information in the operator parameter information set, so as to change the parameters of the function to be called and obtain the calling function corresponding to the operator information.

[0111] In the second sub-step, in response to determining that the operator information has at least one corresponding historical previous layer operator information, at least one previous layer prediction result corresponding to the operator information is input into the above-mentioned calling function to obtain the current prediction result. The at least one previous layer prediction result is at least one current prediction result corresponding to the at least one previous layer operator information corresponding to the operator information. The at least one historical previous layer operator information can be the operator information of at least one previous layer operator that has been executed historically and is related to the input of the operator information.

[0112] The third sub-step is to remove the operator information from the corresponding target operator processing queue to obtain a removed operator processing queue. The target operator processing queue does not include at least one operator information in the target operator information chain that is located before the operator information. That is, the target operator processing queue does not include the operator information that is located before the operator information in the chain.

[0113] In a fourth sub-step, in response to determining that the operator information is the operator information at the second target position in the associated target operator information chain, the current prediction result is determined as the operator execution result. The second position may be the position corresponding to the last operator information in the chain.

[0114] In the second step, in response to determining that the operator processing queue is not empty after removal, the execution subject may determine the next operator information corresponding to the operator information in the target operator information chain as the operator information, and execute the result generation sub-step again.

[0115] Optionally, before removing the operator information from the added operator processing queue to obtain the removed operator processing queue, the method further includes:

[0116] In response to determining that the operator information does not have corresponding at least one historical previous layer operator information, the execution entity may input the processed data into the calling function to obtain a current prediction result.

[0117] In some optional implementations of some embodiments, generating a prediction result according to the obtained at least one operator execution result includes:

[0118] In response to determining that the set of subsequent operator information corresponding to each target operator information in at least one target operator information is empty, the execution subject may determine at least one operator execution result corresponding to the at least one target operator information as the prediction result. The target operator information in the at least one target operator information may be the operator information at the second target position in the target operator information chain in at least one target operator information chain.

[0119] Optionally, in response to determining that the operator processing queue is not empty after the removal, determining the next operator information corresponding to the operator information in the target operator information chain as the operator information, and executing the result generation sub-step again, the method further includes:

[0120] In the first step, in response to determining that there is a target operator information in at least one target operator information whose corresponding back-layer operator information set is not empty, the back-layer operator information set that is not empty is determined as a target back-layer operator information set to obtain at least one target back-layer operator information set. The target back-layer operator information set may be a back-layer operator information set that is not empty and corresponds to the target operator information in the at least one target operator information.

[0121] In the second step, for each target back-layer operator information in the at least one target back-layer operator information set, the following queue generation steps are performed:

[0122] In the first sub-step, the corresponding target operator information is removed from the front-layer operator information set corresponding to the target rear-layer operator information to obtain the front-layer operator information set after removal.

[0123] In the second sub-step, in response to determining that the removed front-layer operator information set is empty, the operator information chain corresponding to the target rear-layer operator information is determined as the target operator information chain. The operator information chain corresponding to the target rear-layer operator information may be a chain including the target rear-layer operator information.

[0124] In the third step, according to the operator sequence corresponding to each target operator information chain in the obtained at least one target operator information chain, each operator information corresponding to the at least one target operator information chain is added to the removed operator processing queue in turn to obtain the added operator processing queue, so as to perform the above result generation step again. Among them, the removed operator processing queue can be a queue after the added operator processing queue removes the executed operator information. The above executed operator information can be the operator information input into the corresponding calling function corresponding to at least one previous layer prediction result.

[0125] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the data prediction method of some embodiments of the present disclosure, the prediction results for the prediction data can be accurately and efficiently achieved based on the received model file and the target dynamic link library.

[0126] Further references Figure 6 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a model information sending device, and these device embodiments are Figure 2 Corresponding to the method embodiments shown, the above-mentioned model information sending device can be specifically applied to various electronic devices.

[0127] like Figure 6 As shown, the model information sending device 600 of some embodiments includes: a first parsing unit 601, a generating unit 602, a compression processing unit 603 and a sending unit 604. Among them, the first parsing unit 601 is configured to perform operator type parsing on the model file to determine the operator type of each operator in the model corresponding to the above model file, and obtain an operator type set; the generating unit 602 is configured to generate a dynamic link library according to the above operator type set. Among them, the calling function in the above dynamic link library is the operator calling function corresponding to the operator type in the above operator type set; the compression processing unit 603 is configured to perform compression processing on the above dynamic link library to obtain a compressed dynamic link library as a target dynamic link library; the sending unit 604 is configured to send the above target dynamic link library and the above model file to the model application end.

[0128] In some optional implementations of some embodiments, the dynamic link library includes: function information. And the compression processing unit 603 can be configured to perform deduplication processing on the function set corresponding to the function information to obtain a deduplication dynamic link library as a compressed dynamic link library.

[0129] In some optional implementations of some embodiments, the dynamic link library includes: function information and source file information. And the compression processing unit 603 can be configured to use a preset compilation optimization method to compile and optimize the source file corresponding to the source file information and the function set corresponding to the function information to obtain an optimized dynamic link library as the compressed dynamic link library.

[0130] It can be understood that the units recorded in the model information sending device 600 are similar to the reference Figure 2 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the device 600 and the units included therein, and will not be described in detail here.

[0131] Further references Figure 7 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a data prediction device. These device embodiments are similar to Figure 4 Corresponding to the method embodiments shown, the above data prediction device can be specifically applied to various electronic devices.

[0132] like Figure 7 As shown, the data prediction device 700 of some embodiments includes: a data preprocessing and parsing unit 701, a receiving unit 702, a second parsing unit 703 and an execution unit 704. The data preprocessing unit 701 is configured to perform data preprocessing on the acquired prediction data to obtain processed data; the receiving unit 702 is configured to receive the model file and the target dynamic link library sent by the server; the second parsing unit 703 is configured to perform file parsing on the above model file to determine the parameter information corresponding to each operator in the model corresponding to the above model file and the connection relationship between each operator, respectively as the operator parameter information set and the operator connection relationship information; the execution unit 704 is configured to take the above processed data as input, and perform data prediction based on operator reasoning according to the above operator connection relationship information, the above operator parameter information set and the above target dynamic link library to generate a prediction result.

[0133] In some optional implementations of some embodiments, the execution unit 704 may be configured to: perform the following result generation step according to the operator information in the plurality of operator information corresponding to the operator connection relationship information: in response to receiving an execution instruction for the operator information, determine the calling function corresponding to the operator information using the operator parameter information set and the target dynamic link library. In response to determining that the operator information is not the starting operator position, input at least one previous layer prediction result into the calling function to obtain the current prediction result corresponding to the calling function. Among them, the at least one previous layer prediction result is at least one previous layer prediction result corresponding to at least one previous layer operator information corresponding to the operator information. In response to determining that the operator information is the ending operator position, determine the current prediction result as the prediction result.

[0134] Optionally, the execution unit 704 may be configured to: for the operator information in the added operator processing queue that is associated with the target operator information chain, perform the following result generation sub-steps: determine the calling function corresponding to the operator information using the operator parameter information set and the target dynamic link library; in response to determining that the operator information has at least one corresponding historical previous layer operator information, input at least one previous layer prediction result corresponding to the operator information into the calling function to obtain a current prediction result; remove the operator information from the corresponding target operator processing queue to obtain a removed operator processing queue, wherein the target operator processing queue does not include at least one operator information in the target operator information chain that is located before the operator information; in response to determining that the operator information is the operator information at the second target position in the associated target operator information chain, determine the current prediction result as the operator execution result; in response to determining that the removed operator processing queue is not empty, determine the next operator information corresponding to the operator information in the target operator information chain as the operator information, and perform the result generation sub-step again.

[0135] Optionally, the execution unit 704 may be configured to: in response to determining that the operator information does not have corresponding at least one historical previous layer operator information, input the processed data into the calling function to obtain a current prediction result.

[0136] Optionally, the execution unit 704 may be configured to: in response to determining that a set of back-layer operator information corresponding to each target operator information in at least one target operator information is empty, determine at least one operator execution result corresponding to at least one target operator information as a prediction result, wherein the target operator information in the at least one target operator information is the operator information of the second target position in the target operator information chain in at least one target operator information chain.

[0137] Optionally, the execution unit 704 may be configured to: in response to determining that the operator processing queue after removal is empty, and there is target operator information whose corresponding rear-layer operator information set is not empty in at least one target operator information, determine the non-empty rear-layer operator information set as the target rear-layer operator information set to obtain at least one target rear-layer operator information set; for each target rear-layer operator information in the at least one target rear-layer operator information set, perform the following queue generation step: remove the corresponding target operator information from the front-layer operator information set corresponding to the target rear-layer operator information to obtain a post-removal front-layer operator information set; in response to determining that the post-removal front-layer operator information set is empty, determine the operator information chain corresponding to the target rear-layer operator information as the target operator information chain; according to the operator sequence corresponding to each target operator information chain in the obtained at least one target operator information chain, add each operator information corresponding to the at least one target operator information chain to the operator processing queue after removal in turn to obtain an added operator processing queue, so as to perform the result generation step again.

[0138] It can be understood that the units described in the data prediction device 700 are similar to those described in the reference Figure 4 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the device 700 and the units included therein, and will not be described in detail here.

[0139] Reference below Figure 8 , which shows a structural schematic diagram of an electronic device 800 suitable for implementing some embodiments of the present disclosure. Figure 8 The server shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0140] like Figure 8 As shown, the electronic device 800 may include a processing device 801 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage device 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the electronic device 800 are also stored. The processing device 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0141] Typically, the following devices may be connected to the I / O interface 805: input devices 806 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 808 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 809. The communication device 809 may allow the electronic device 800 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 8 The electronic device 800 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 8 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0142] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 809, or installed from the storage device 808, or installed from the ROM 802. When the computer program is executed by the processing device 801, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0143] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0144] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0145] The computer-readable medium may be included in the device; or it may exist independently without being assembled into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: performs operator type analysis on the model file to determine the operator type of each operator in the model corresponding to the model file, and obtains an operator type set; generates a dynamic link library according to the operator type set; compresses the dynamic link library to obtain a compressed dynamic link library as a target dynamic link library; sends the target dynamic link library and the model file to the model application end. Perform data preprocessing on the acquired prediction data to obtain processed data. Receive the model file and the target dynamic link library sent by the server. Perform file analysis on the model file to determine the parameter information corresponding to each operator in the model corresponding to the model file and the connection relationship between each operator, which are respectively used as the operator parameter information set and the operator connection relationship information. The processed data is used as input, and data prediction based on operator reasoning is performed according to the operator connection relationship information, the operator parameter information set and the target dynamic link library to generate a prediction result.

[0146] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0147] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0148] The units described in some embodiments of the present disclosure may be implemented by software or hardware. The units described may also be provided in a processor, for example, may be described as: a processor including a first parsing unit, a generating unit, a compression processing unit, and a sending unit. The names of these units do not, in some cases, constitute limitations on the units themselves, for example, the first parsing unit may also be described as a "unit for parsing a model file".

[0149] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0150] Some embodiments of the present disclosure further provide a computer program product, including a computer program, which implements any of the above-mentioned model information sending methods when executed by a processor.

[0151] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.

Claims

1. A method for sending model information, comprising: Performing operator type parsing on the model file to determine the operator type of each operator in the model corresponding to the model file, and obtaining an operator type set; Generate a dynamic link library according to the operator type set; Compressing the dynamic link library to obtain a compressed dynamic link library as a target dynamic link library; The target dynamic link library and the model file are sent to the model application end.

2. The method according to claim 1, wherein: The dynamic link library includes: function information; and The step of compressing the dynamic link library to obtain a compressed dynamic link library comprises: The function set corresponding to the function information is deduplicated to obtain a deduplicated dynamic link library as a compressed dynamic link library.

3. The method according to claim 1, wherein: The dynamic link library includes: function information and source file information; and The step of compressing the dynamic link library to obtain a compressed dynamic link library comprises: By using a preset compilation optimization method, the source file corresponding to the source file information and the function set corresponding to the function information are compiled and optimized to obtain an optimized dynamic link library as a compressed dynamic link library.

4. A data prediction method, comprising: Performing data preprocessing on the acquired prediction data to obtain processed data; Receive the model file and target dynamic link library sent by the server; Parsing the model file to determine parameter information corresponding to each operator in the model corresponding to the model file and connection relationships between each operator as an operator parameter information set and operator connection relationship information, respectively; The processed data is used as input, and data prediction based on operator reasoning is performed according to the operator connection relationship information, the operator parameter information set and the target dynamic link library to generate a prediction result.

5. The method according to claim 4, wherein: The processing of the processed data as input, performing data prediction based on operator reasoning according to the operator connection relationship information, the operator parameter information set and the target dynamic link library to generate a prediction result includes: According to the operator sequence corresponding to each target operator information chain, each operator information corresponding to at least one target operator information chain is added to the operator processing queue in turn to obtain an added operator processing queue, wherein the front-layer operator information set corresponding to the operator information of the first target position in each target operator information chain in the at least one target operator information chain is empty; For each target operator information chain in at least one target operator information chain, perform the following result generation steps: According to the operator parameter information set and the target dynamic link library, each operator information associated with the target operator information chain in the added operator processing queue is executed in sequence to obtain an operator execution result; A prediction result is generated according to the obtained at least one operator execution result.

6. The method according to claim 5, wherein: The step of sequentially executing each operator information associated with the target operator information chain in the added operator processing queue according to the operator parameter information set and the target dynamic link library to obtain the operator execution result includes: For the operator information in the added operator processing queue that is associated with the target operator information chain, the following result generation sub-steps are performed: Determine the calling function corresponding to the operator information by using the operator parameter information set and the target dynamic link library; In response to determining that the operator information has at least one historical previous layer operator information corresponding thereto, inputting at least one previous layer prediction result corresponding to the operator information into the calling function to obtain a current prediction result; Removing the operator information from the corresponding target operator processing queue to obtain a removed operator processing queue, wherein the target operator processing queue does not include at least one operator information in the target operator information chain that is located before the operator information; In response to determining that the operator information is the operator information of the second target position in the associated target operator information chain, determining the current prediction result as the operator execution result; In response to determining that the removed operator processing queue is not empty, the next operator information corresponding to the operator information in the target operator information chain is determined as the operator information, and the result generation sub-step is performed again.

7. The method according to claim 6, wherein: Before removing the operator information from the corresponding target operator processing queue to obtain the removed operator processing queue, the method further includes: In response to determining that the operator information does not have corresponding at least one historical previous layer operator information, the processed data is input into the calling function to obtain a current prediction result.

8. The method according to claim 7, wherein: The generating a prediction result according to the obtained at least one operator execution result includes: In response to determining that a set of back-layer operator information corresponding to each target operator information in at least one target operator information is empty, at least one operator execution result corresponding to the at least one target operator information is determined as a prediction result, wherein the target operator information in the at least one target operator information is the operator information of the second target position in the target operator information chain in at least one target operator information chain.

9. The method according to claim 8, wherein: In response to determining that the removed operator processing queue is not empty, determining the next operator information corresponding to the operator information in the target operator information chain as the operator information, and executing the result generation sub-step again, comprising: In response to determining that the removed operator processing queue is empty, and there is target operator information in at least one target operator information whose corresponding back-layer operator information set is not empty, determining the back-layer operator information set that is not empty as the target back-layer operator information set, and obtaining at least one target back-layer operator information set; For each target back-layer operator information in the at least one target back-layer operator information set, the following queue generation steps are performed: Removing the corresponding target operator information from the front-layer operator information set corresponding to the target rear-layer operator information to obtain the removed front-layer operator information set; In response to determining that the removed rear-layer operator information set is empty, determining the operator information chain corresponding to the target rear-layer operator information as the target operator information chain; According to the operator order corresponding to each target operator information chain in the obtained at least one target operator information chain, each operator information corresponding to the at least one target operator information chain is added to the removed operator processing queue in turn to obtain the added operator processing queue, so as to execute the result generating step again.

10. A model information sending device, comprising: A first parsing unit is configured to perform operator type parsing on the model file to determine the operator type of each operator in the model corresponding to the model file, and obtain an operator type set; A generating unit, configured to generate a dynamic link library according to the operator type set; A compression processing unit is configured to compress the dynamic link library to obtain a compressed dynamic link library as a target dynamic link library; The sending unit is configured to send the target dynamic link library and the model file to the model application end.

11. A data prediction device, comprising: A data preprocessing unit is configured to perform data preprocessing on the acquired prediction data to obtain processed data; A receiving unit is configured to receive a model file and a target dynamic link library sent by a server; A second parsing unit is configured to parse the model file to determine parameter information corresponding to each operator in the model corresponding to the model file and a connection relationship between each operator as an operator parameter information set and operator connection relationship information, respectively; The execution unit is configured to take the processed data as input, and perform data prediction based on operator reasoning according to the operator connection relationship information, the operator parameter information set and the target dynamic link library to generate a prediction result.

12. A server, comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 9.

13. A computer readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

14. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 9.