WO2022108527A1

WO2022108527A1 - Model processing method, system and apparatus, medium, and electronic device

Info

Publication number: WO2022108527A1
Application number: PCT/SG2021/050707
Authority: WO
Inventors: 陈程; 周子凯; 余乐乐; 解浚源; 吴良超; 常龙; 张力哲; 刘小兵; 吴迪
Original assignee: 脸萌有限公司
Priority date: 2020-11-18
Filing date: 2021-11-16
Publication date: 2022-05-27
Also published as: CN112418446A; CN112418446B

Abstract

The present disclosure relates to a model processing method, system and apparatus, a medium, and an electronic device. The method comprises: acquiring a plurality of sub-models; stitching the plurality of sub-models to obtain a target model; and when receiving a model acquisition request for the target model from an inference service execution party, sending the target model to the inference service execution party, so that the inference service execution party obtains an inference result by means of the target model.

Description

MODEL PROCESSING METHODS, SYSTEMS, APPARATUS, MEDIA AND ELECTRONIC DEVICES CROSS-REFERENCE TO RELATED APPLICATIONS This application is based on an application with CN application number 202011298789.5 and an application date of November 18, 2020, and claims its priority, the CN application The disclosure of is hereby incorporated into this application in its entirety. FIELD OF THE DISCLOSURE The present disclosure relates to the field of computer technology, and in particular, to a model processing method, system, apparatus, medium, and electronic device. 2. Description of the Related Art Federated machine learning, also known as federated learning and federated learning, has been widely used in the field of machine learning. Federated learning can solve data silos and data privacy issues. Under the requirements of user privacy protection and data security, the federated learning system can effectively help multiple institutions to complete the joint training of models. The federated learning model is usually composed of multiple sub-models. How to ensure the reliability of the inference service is an important issue when the inference service is performed through the federated learning model. SUMMARY This Summary section is provided to introduce in a simplified form concepts that are described in detail in the Detailed Description section that follows. This summary section is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution. In a first aspect, the present disclosure provides a model processing method, the method includes: acquiring multiple sub-models; splicing the multiple sub-models to obtain a target model; after receiving a message sent by an inference service executor for the target model In the case of the model obtaining request, the target model is sent to the inference service executor, so that the inference service executor obtains an inference result through the target model. In a second aspect, the present disclosure provides a model processing method. The method includes: an inference service executor sends a model acquisition request for a target model, where the target model is obtained by splicing multiple sub-models; the inference service executes The party receives the target model, and obtains an inference result through the target model. In a third aspect, the present disclosure provides a model processing system, the system includes a model optimization platform and a model storage platform; the model optimization platform is used for acquiring multiple sub-models, and splicing the multiple sub-models to obtain target model, and send the target model to the model storage platform; the model storage platform is configured to store the target model in the case of receiving a model acquisition request for the target model sent by the inference service executor The model is sent to the inference service executor, so that the inference service executor obtains an inference result through the target model. In a fourth aspect, the present disclosure provides a model processing device, the device comprising: an acquisition module configured to acquire multiple sub-models; and a splicing module configured to splicing the multiple sub-models to obtain a target model; a target model sending module, configured to send the target model to the inference service executor in the case of receiving a model acquisition request for the target model sent by the inference service executor, so that The inference service executor obtains an inference result through the target model. In a fifth aspect, the present disclosure provides a model processing apparatus, the apparatus comprising: an acquisition request sending module, configured to send a model acquisition request for a target model, wherein the target model is to convert the multiple sub-models obtained by splicing; an inference module, configured to receive the target model, and obtain an inference result through the target model. For example, the model processing apparatus may be set at the execution side of the inference service. In a sixth aspect, the present disclosure provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processing apparatus, implements the steps of the method provided in the first aspect of the present disclosure. In a seventh aspect, the present disclosure provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processing apparatus, implements the steps of the method provided in the second aspect of the present disclosure. In an eighth aspect, the present disclosure provides an electronic device, comprising: a storage device on which a computer program is stored; and a processing device for executing the computer program in the storage device, so as to implement the computer program provided in the first aspect of the present disclosure. the steps of the method. In a ninth aspect, the present disclosure provides an electronic device, comprising: a storage device on which a computer program is stored; and a processing device for executing the computer program in the storage device, so as to implement the computer program provided in the second aspect of the present disclosure. the steps of the method. In a tenth aspect, the present disclosure provides a computer program, comprising: instructions that, when executed by a processor, cause the processor to execute the model processing method according to any one of the foregoing embodiments. In an eleventh aspect, the present disclosure provides a computer program product comprising instructions that, when executed by a processor, cause the processor to execute the model processing method according to any one of the foregoing embodiments. Other features and advantages of the present disclosure will be described in detail in the detailed description that follows. BRIEF DESCRIPTION OF THE DRAWINGS The above and other features, advantages and aspects of various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numbers refer to the same or similar elements. It should be understood that the drawings are schematic and that the originals and elements are not necessarily drawn to scale. In the drawings: FIG. 1 is a schematic diagram of a federated learning model in the related art. Fig. 2 is a flow chart of a model processing method according to some exemplary embodiments. Fig. 3 is a schematic diagram of a target model according to some exemplary embodiments. FIG. 4 is a schematic diagram of a model processing system according to some exemplary embodiments. Fig. 5 is a schematic diagram illustrating an inference service executor obtaining an inference result through a target model according to its own model input data, according to some exemplary embodiments. Fig. 6 is a flow chart of a model processing method according to some exemplary embodiments. Fig. 7 is a block diagram of a model processing apparatus according to some exemplary embodiments. Fig. 8 is a block diagram of a model processing apparatus according to some exemplary embodiments. FIG. 9 is a schematic structural diagram of an electronic device according to some exemplary embodiments. DETAILED DESCRIPTION Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While certain embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be construed as limited to the embodiments set forth herein, but rather are provided for the purpose of 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 used for exemplary purposes, and are not intended to limit the protection scope of the present disclosure. It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders, and/or performed in parallel. Furthermore, method embodiments may include additional steps and/or omit performing the illustrated steps. The scope of the present disclosure is not limited in this regard. As used herein, the term "including" and variations thereof are open-ended inclusions, ie, "including but not limited to". The term "based on," is "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment": the term "some embodiments" Represents "at least some embodiments". Relevant definitions of other terms will be given in the following description. It should be noted that concepts such as "first" and "second" mentioned in this disclosure are only used for different devices and modules. or units are not used to limit the order or phase of the functions performed by these devices, modules or units. interdependence. It should be noted that the modifications of "a" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless the context clearly indicates otherwise, they should be understood as "one or a plurality of" The names of the messages or information exchanged between the 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. The federated learning system can combine the data of multiple data owners to train a common federated learning model. The federated learning model is trained by combining the data of multiple data owners, and the training data is more comprehensive. Therefore, the accuracy of the federated learning model is higher. A federated learning model is usually composed of multiple sub-models, and FIG. 1 is a schematic diagram of a federated learning model in the related art. As shown in Figure 1, the federated learning model includes sub-model A and sub-model B. For example, the sub-model A corresponds to the model training participant 1, and the model input data X, Y, and Z of the sub-model A are data owned by the model training participant 1. The sub-model B corresponds to the model training participant 2, and the model input data M and N of the sub-model B are data owned by the model training participant 2. When performing inference services through the federated learning model, each model training participant loads its own sub-models, that is, model training participant 1 loads sub-model A, and model training participant 2 loads sub-model B. As shown in FIG. 1 , the model training participant 1 performs computation through the sub-model A according to the model input data X, Y, and Z. Then model training participant 1 needs to remotely send the data to model training participant 2 through the sending node of sub-model A, so as to transmit the data to the receiving node of sub-model B. The model training participant 2 then obtains an inference result through the sub-model B according to the data received by the receiving node and the model input data M and N. In this way, when the inference service is performed through the federated learning model, remote communication is required between multiple model training participants to complete the entire inference service, that is, the sending node and the receiving node use remote communication to transmit data, and the communication overhead is large. Moreover, long-distance communication is easily affected by factors such as network routing, and is usually not stable enough and has low reliability, which makes the calculation process of the inference service not stable enough. For example, if the sending node cannot transmit the data to the receiving node in time due to network congestion when the data is remotely transmitted to the receiving node, the entire inference service will be affected. In order to solve the problems existing in the related art, the present disclosure provides a model processing method, system, apparatus, medium and electronic device. Fig. 2 is a flow chart of a model processing method according to some exemplary embodiments. As shown in FIG. 2, the method may include S201~S203o In S201, multiple sub-models are acquired. In S202, multiple sub-models are spliced to obtain a target model. Fig. 3 is a schematic diagram of a target model according to some exemplary embodiments. The target model shown in FIG. 3 may be obtained according to the federated learning model shown in FIG. 1 , and there is a connection relationship between the sending node of the sub-model A and the receiving node of the sub-model B. As shown in FIG. 3 , the computing node of the sub-model A connected to the sending node and the computing node of the sub-model B connected to the receiving node can be connected to obtain the target model. The target model is the overall full model obtained by splicing sub-model A and sub-model B together. It is worth noting that the present disclosure takes two sub-models as examples for illustration, which does not constitute a limitation on the implementation of the present disclosure. In practical applications, the number of sub-models may be multiple, which is not specifically limited in the present disclosure. . In S203, in the case of receiving the model acquisition request for the target model sent by the inference service executor, the target model is sent to the inference service executor, so that the inference service executor obtains the inference result through the target model. Inference service can refer to the process by which the server performs computations through the model based on the input data and obtains the result. For example, taking the prediction of the user's shopping intention as an example, according to the user's historical shopping behavior information, the user's current shopping intention can be inferred through a model, and then the user can be provided with an inference result that meets his shopping intention and needs. As another example, taking the prediction of the user's search intent as an example, according to the user's historical click behavior information, the user's current search intent can be inferred through the model, and then the user can be provided with an inference result that conforms to the user's search intent. In an optional implementation manner, one of the model training participants can be used as the inference service executor, load the target model, and obtain the inference result through the target model. The target model is obtained by splicing multiple sub-models, and the inference service executor can directly obtain the inference results through the overall target model, without each model training participant loading its own sub-models, and without the need for multiple model training participants. It can effectively avoid the problem of unstable long-distance communication. It is worth noting that, when it is mentioned in the present disclosure that the inference service executor performs operations of sending, receiving, and processing data, it may be understood that the inference service executor performs these operations through a server device. Through the above technical solution, a target model is obtained by splicing multiple sub-models, and the inference service executor can obtain an inference result through the target model. The target model is obtained by splicing multiple sub-models, and the inference service executor can directly obtain the inference result through the overall target model, without each model training participant loading its own sub-models separately. The entire inference service process can be completed locally on the inference service executor, without the need for remote communication between multiple model training participants to transmit data. In this way, not only the communication overhead can be reduced, but also the problem of unstable remote transmission caused by factors such as network routing can be effectively avoided, the normal operation of the inference service can be ensured, and the reliability of the inference service can be improved. In one embodiment, the model processing method shown in FIG. 2 may be applied to a model processing apparatus including a splicing module. The model processing device may be, for example, a cloud server, the acquisition module in the model processing device acquires multiple sub-models, and the splicing module splices the multiple sub-models to obtain the target model. In another embodiment, the model processing method shown in FIG. 2 can also be applied to a model processing system. FIG. 4 is a schematic diagram of a model processing system according to some exemplary embodiments. As shown in FIG. 4 , the model processing system may include a model optimization platform 401, a model storage platform 402, a model training platform 403, a model metadata storage platform 404, a model training participant 1, and a model training participant 2. The model training platform 403 is used to train each sub-model, such as sub-model A and sub-model B. The model meta information storage platform 404 may be used to store model related meta information. The model training platform 403 can send multiple sub-models to the model optimization platform 401. The model optimization platform 401 can be used to obtain the multiple sub-models sent by the model training platform 403, splicing the multiple sub-models to obtain the target model, and sending the target model to Model storage platform 402 _o The inference service executor may send a model acquisition request for the target model to the model storage platform 402; the model storage platform 402 may send the target model to the inference service executor when receiving the request. The inference service executor may be one of model training participant 1 and model training participant 2, for example. FIG. 4 takes an example of including two model training participants for illustration, which does not constitute a limitation on the embodiments of the present disclosure. In the present disclosure, the step of splicing multiple sub-models in S202 may include: acquiring model meta information, where the model meta information may include a sending node having a sub-model of the sending node and other sub-models that have a connection relationship with the sending node The connection relationship information between the receiving nodes; according to the model meta information, connect the computing node of the sub-model connected to the sending node and the computing nodes of other sub-models connected to the receiving node to connect the multiple sub-models. stitching. Model meta-information may refer to information used to describe a model, and may include connection relationship information between nodes. When the model processing method provided by the present disclosure is applied to the model processing system shown in FIG. 4 , the model optimization platform 401 can obtain model meta information from the model meta information storage platform 404, and according to the model meta information, connect the The computing nodes of the sub-model are connected with the computing nodes of other sub-models connected to the receiving node. In this way, data can be directly transmitted between two computing nodes that originally need to forward data through the sending node and the receiving node. The inference service executor performs the inference service through the target model, and the inference service executor locally The entire inference service process can be completed without the need for remote communication of data, effectively avoiding the problem of unstable remote transmission caused by factors such as network routing, ensuring the normal operation of the inference service, and improving the reliability of the inference service. Several exemplary implementations of determining the inference service executor and obtaining inference results through the target model in the present disclosure are described below. In an optional embodiment, multiple sub-models correspond to multiple model training participants one-to-one, and each model training participant has its own model input data. The inference service executor can be one of multiple model training participants, and the inference service executor can obtain inference results through the target model according to its own model input data. In this embodiment, the inference service performer may be any of a plurality of model training participants. Fig. 5 is a schematic diagram illustrating an inference service executor obtaining an inference result through a target model according to its own model input data, according to some exemplary embodiments. Figure 5 takes the inference service executor as the model training participant 1 as an example. As shown in FIG. 5 , after the model training participant 1 obtains the target model, it can input data X, Y, and Z according to its own model, and obtain an inference result through the target model. For example, the inference service executor may be the party that needs the inference result among the multiple model training participants, that is, the inference result demander. For example, the model training participant 1 needs the final inference result, and the model training participant 1 can be used as the inference service executor to obtain the target model and perform the inference service. Or, if the inference service executor is not the inference result requester, the inference service executor may send the inference result to the inference result requester. For example, model training participant 2 needs an inference result, and model training participant 1 can send the inference result to model training participant 2 . Through the above solution, other model training participants may not transmit model input data to the inference service executor, and the inference service executor can perform inference services according to its own data, with low communication overhead. Moreover, the reasoning service is completed by the inference service executor, and there is no need to carry out the process of remote communication and data transmission between multiple model training participants, which reduces the communication overhead and improves the stability of the inference service. In another optional embodiment, multiple sub-models correspond to multiple model training participants one-to-one, each model training participant has its own model input data, and the inference service executor is one of the multiple model training participants . The inference service executor can obtain the inference results in the following ways: Receive encrypted model input data sent by other model training participants except the inference service executor; According to the inference service executor's own model input data and other model training participants The encrypted model input data is obtained, and the inference result is obtained through the target model. In order to protect data privacy and ensure data security, other model training participants can add model input data. encrypted processing. Send the encrypted model input data to the inference service executor, and the present disclosure does not specifically limit the encryption method. The inference service executor can use its own model input data and the encrypted model input data of other model training participants to obtain inference results through the target model, that is, to perform inference services based on the model input data of each sub-model. Optionally, the inference service executor may be the model training participant that needs to receive the encrypted model input data of other model training participants with the smallest amount of data. For example, the inference service executor determines by: determining the data volume of the encrypted model input data sent by other model training participants required by each model training participant to perform the inference service; Determined to be the inference service executor. For example, if the model training participant 1 performs the inference service, the model training participant 2 needs to send the model input data M and N of the sub-model B to the model training participant 1; if the model training participant 2 performs the inference service, Model training participant 1 needs to send the model input data X, Y, and Z of sub-model A to model training participant 2. If the data volume of the model input data M and N is smaller than the data volume of the model input data X, Y, and Z, the model training participant 1 needs to receive the minimum data volume of the model input data, which can be executed by the model training participant 1 as an inference service. square. Through the above solution, if the inference service executor performs the inference service according to the model input data of each sub-model, the model training participant with the smallest amount of model input data to be received can be used as the inference service executor, which can, to a certain extent, be used as the inference service executor. Reduce communication overhead. In addition, the inference service executor performs the calculation through the target model, and there is no need to carry out the process of remote communication and data transmission among multiple model training participants, thereby improving the stability of the inference service process. In yet another optional embodiment, multiple sub-models correspond to multiple model training participants one-to-one, each model training participant has its own model input data, and the inference service executor is not a model training participant; inference service The executor can obtain the inference result in the following ways: Receive the encrypted model input data sent by each model training participant respectively; Obtain the inference result through the target model according to the encrypted model input data of each model training participant. In this embodiment, the inference service executor may not be a model training participant, for example, it may be a preset cloud server. Each model training participant can send its own model input data to the inference service executor in encrypted form. After the inference service executor obtains the target model, it can train the participants' encrypted model input data according to each model, and obtain the inference result through the target model. Through the above scheme, the inference service executor obtains the inference result through the target model, and no longer needs to perform remote communication of data, which effectively avoids the problem of unstable remote transmission caused by factors such as network routing. Ensure the normal operation of the inference service, thereby improving the stability and reliability of the inference service. Fig. 6 is a flow chart of a model processing method according to some exemplary embodiments. The method can be applied to the inference service executor, that is, the server device of the inference service executor. As shown in FIG. 6, the method may include S501 and S502. In S501, the inference service executor sends a model acquisition request for the target model. The target model is obtained by splicing the multiple sub-models. The inference service executor may send a model acquisition request to the model storage platform, and may also send a model acquisition request to a model processing apparatus including a splicing module, which is not limited in the present disclosure. In S502, the inference service executor receives the target model, and obtains an inference result through the target model. Through the above technical solution, the inference service executor can send a model acquisition request for the target model. The target model is obtained by splicing multiple sub-models, and the inference service executor can obtain the inference result through the target model. Since the target model is obtained by splicing multiple sub-models together, the inference service executor obtains the inference result through the target model. The inference service executor can complete the entire inference service process locally. In this way, there is no need to perform remote communication of data, which not only reduces communication overhead, but also effectively avoids the problem of unstable remote transmission caused by factors such as network routing, ensures the normal operation of inference services, and improves the reliability of inference services. . Optionally, the multiple sub-models are in one-to-one correspondence with multiple model training participants, each of the model training participants has its own model input data, and the inference service executor is one of the multiple model training participants. The step of obtaining the inference result through the target model in S502 may include: inputting data according to the model of the inference service executor itself, and obtaining the inference result through the target model. Optionally, the multiple sub-models are in one-to-one correspondence with multiple model training participants, each of the model training participants has its own model input data, and the inference service executor is one of the multiple model training participants. The step of obtaining the inference result from the target model in S502 may include: receiving encrypted model input data sent by other model training participants except the inference service executor; and according to the model input data of the inference service executor itself , and the encrypted model input data of the other model training participants, and obtain the inference result through the target model. Optionally, the inference service executor is a model training participant that needs to receive the encrypted model input data of other model training participants with the smallest amount of data. For example, the inference service executor determines by: determining the data volume of the encrypted model input data sent by other model training participants required by each model training participant to perform the inference service; Determined to be the inference service executor. Optionally, the multiple sub-models are in one-to-one correspondence with multiple model training participants, each of the model training participants has its own model input data, and the inference service executor is not the model training participant. The step of obtaining the inference result through the target model in S502 may include: respectively receiving encrypted model input data sent by each of the model training participants; according to the encrypted model input data of each of the model training participants, The target model obtains the inference result. Regarding the method applied to the inference service executor in the foregoing embodiment, the specific manner in which each step performs operations has been described in detail in the embodiment of the method applied to the model processing system or the model processing apparatus including the splicing module, here A detailed explanation will not be given. The present disclosure also provides a model processing system, such as the model processing system shown in FIG. 4 , the system may include a model optimization platform and a model storage platform; the model optimization platform is used for acquiring multiple sub-models, and storing the multiple sub-models Splicing is performed to obtain a target model, and the target model is sent to the model storage platform; the model storage platform is configured to, upon receiving a model acquisition request for the target model sent by the inference service executor, Send the target model to the inference service executor, so that the inference service executor obtains an inference result through the target model. Optionally, the model optimization platform is used to obtain model meta information, and the model The meta-information includes connection relationship information between a sending node having a sub-model of the sending node and receiving nodes of other sub-models that have a connection relationship with the sending node; the model optimization platform is configured to, according to the model meta-information, The computing node of the sub-model connected with the sending node is connected with the computing nodes of the other sub-models connected with the receiving node, so as to splicing the multiple sub-models. Regarding the system in the above-mentioned embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be described in detail here. Fig. 7 is a block diagram of a model processing apparatus according to some exemplary embodiments. As shown in FIG. 7 , the model processing apparatus 600 may include: an acquisition module 601, configured to acquire multiple sub-models; a splicing module 602, configured to splicing the multiple sub-models to obtain a target model; The target model sending module 603 is configured to send the target model to the inference service executor in the case of receiving a model acquisition request for the target model sent by the inference service executor, so that all The inference service executor obtains the inference result through the target model. Optionally, the splicing module 602 may include: an obtaining submodule configured to obtain model meta information, where the model meta information includes a sending node having a sub-model of a sending node and a sending node having a connection with the sending node the connection relationship information between the receiving nodes of other sub-models of the connection relationship; the splicing sub-module is configured to connect the computing node of the sub-model connected to the sending node and the computing node of the sub-model connected to the sending node according to the model meta information. The computing nodes of the other sub-models connected to the receiving node are connected to splicing the multiple sub-models. Optionally, multiple sub-models are in one-to-one correspondence with multiple model training participants, each of the model training participants has its own model input data, and the inference service executor is one of the multiple model training participants. First, the inference service executor obtains the inference result through the target model according to its own model input data. Optionally, multiple sub-models are in one-to-one correspondence with multiple model training participants, each of the model training participants has its own model input data, and the inference service executor is one of the multiple model training participants. 1. The inference service executor obtains the inference result in the following ways: Receive encrypted model input data sent by other model training participants except the inference service executor; According to the inference service executor's own model The input data and the encrypted model input data of the other model training participants are used to obtain the inference result through the target model. Optionally, the inference service executor is a model training participant that needs to receive the encrypted model input data of other model training participants with the smallest amount of data. For example, the inference service executor determines by: determining the data volume of the encrypted model input data sent by other model training participants required by each model training participant to perform the inference service; Determined to be the inference service executor. Optionally, multiple sub-models are in one-to-one correspondence with multiple model training participants, each model training participant has its own model input data, and the inference service executor is not the model training participant; The inference service executor obtains the inference result in the following ways: respectively receiving the encrypted model input data sent by each of the model training participants; according to the encrypted model input data of each of the model training participants, through the target The model obtains the inference result. Fig. 8 is a block diagram of a model processing apparatus according to some exemplary embodiments. The model processing apparatus 700 can be applied to an inference service executor. As shown in FIG. 8 , the model processing apparatus 700 may include: an acquisition request sending module 701, configured to send a model acquisition request for a target model, wherein the target model is obtained by splicing the multiple sub-models The inference module 702 is configured to receive the target model and obtain an inference result through the target model. Optionally, multiple sub-models are in one-to-one correspondence with multiple model training participants, each of the model training participants has its own model input data, and the inference service executor is one of the multiple model training participants; The inference module 702 may include: a first inference sub-module, configured to input data according to the model of the inference service executor itself, and obtain the inference result through the target model. Optionally, multiple sub-models are in one-to-one correspondence with multiple model training participants, each of the model training participants has its own model input data, and the inference service executor is one of the multiple model training participants; The inference module 702 may include: a first receiving sub-module, configured to receive encrypted model input data sent by other model training participants except the inference service executor; a second inference sub-module, configured by It is configured to obtain the inference result through the target model according to the model input data of the inference service executor itself and the encrypted model input data of the other model training participants. Optionally, the inference service executor is a model training participant that needs to receive the encrypted model input data of other model training participants with the smallest amount of data. For example, the inference service executor determines by: determining the data volume of the encrypted model input data sent by other model training participants required by each model training participant to perform the inference service; Determined to be the inference service executor. Optionally, multiple sub-models are in one-to-one correspondence with multiple model training participants, each model training participant has its own model input data, and the inference service executor is not the model training participant; the reasoning The module 702 may include: a second receiving sub-module, configured to respectively receive encrypted model input data sent by each of the model training participants; The encrypted model input data of the training participant is used to obtain the inference result through the target model. 9, which shows a schematic structural diagram of an electronic device 800 suitable for implementing an embodiment of the present disclosure. Terminal devices in the embodiments of the present disclosure may include, but are not limited to, mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablets), PMPs (portable multimedia players), vehicle-mounted terminals (eg, Mobile terminals such as car navigation terminals), etc., and stationary terminals such as digital TVs, desktop computers, and the like. The electronic device shown in FIG. 9 is only an example, and should not impose any limitation on the function and scope of use of the embodiments of the present disclosure. As shown in FIG. 9, the electronic device 800 may include a processing device (eg, a central processing unit, a graphics processor, etc.) 801, which may be loaded into random access according to a program stored in a read only memory (ROM) 802 or from a storage device 808 The program in the memory (RAM) 803 executes various appropriate actions and processes. In the RAM 803, various programs and data necessary 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 through a bus 804 . An input/output (I/O) interface 805 is also connected to bus 804 . Typically, the following devices can be connected to the I/O interface 805: Input devices 806 including, for example, a touch screen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; including, for example, a liquid crystal display (LCD), speakers, vibration an output device 807 for a computer, etc.; a storage device 808 including, for example, a magnetic tape, a hard disk, etc.; exchange data. While FIG. 9 shows electronic device 800 having various means, it should be understood that not all of the illustrated means are required to be implemented or available. More or fewer devices may alternatively be implemented or provided. In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a non-transitory computer readable medium, the computer program containing program code for performing the method illustrated in the flowchart. In such an embodiment, the computer program may be downloaded and installed from the network via the communication device 809 , or from the storage device 808 , or from the ROM 802 . When the computer program is executed by the processing device 801, the above-mentioned functions defined in the methods of the embodiments of the present disclosure are executed. It should be noted that, the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination of the above. More specific examples of computer readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable Programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present disclosure, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code therein. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device . The program code embodied on the computer-readable medium may be transmitted by any suitable medium, including but not limited to: wire, optical fiber cable, RF (radio frequency), etc., or any suitable combination of the above. In some embodiments, the client and the server can use any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol) to communicate, and can communicate with digital data in any form or medium Communication (eg, communication network) interconnection. Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), the Internet (eg, the Internet), and peer-to-peer networks (eg, ad hoc peer-to-peer networks), as well as any currently known or future development network of. The above-mentioned computer-readable medium may be included in the above-mentioned electronic device; it may also exist independently without assembled into the electronic device. The computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device causes the electronic device to: acquire multiple sub-models; splicing the multiple sub-models to obtain a target model; In the case of receiving the model acquisition request for the target model sent by the inference service executor, send the target model to the inference service executor, so that the inference service executor obtains the target model through the inference service executor inference result. Alternatively, the computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device causes the electronic device to: send a model acquisition request for a target model, where the target model is obtained by splicing the multiple sub-models; receiving the target model, and obtaining an inference result through the target model. Computer program code for carrying out operations of the present disclosure may be written in one or more programming languages, including but not limited to object-oriented programming languages such as Java, Smalltalk. C++, and This includes conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (eg, using an Internet service provider to via an Internet connection). The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or part of code that contains one or more logic functions for implementing the specified executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by dedicated hardware-based systems that perform the specified functions or operations , or can be implemented using a combination of dedicated hardware and computer instructions. The modules involved in the embodiments of the present disclosure may be implemented in a software manner, and may also be implemented in a hardware manner. Wherein, the name of the module does not constitute a limitation of the module itself under certain circumstances, for example, the splicing module may also be described as a "sub-model splicing module". The functions described herein above 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 Logical Devices (CPLDs) and more. In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), fiber optics, portable compact disk read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. According to one or more embodiments of the present disclosure, Example 1 provides a model processing method, the method includes: acquiring multiple sub-models; splicing the multiple sub-models to obtain a target model; In the case of a model acquisition request for the target model sent by the inference party, the target model is sent to the inference service executor, so that the inference service executor obtains an inference result through the target model. According to one or more embodiments of the present disclosure, Example 2 provides the method of Example 1. The splicing of the multiple sub-models includes: acquiring model meta information, where the model meta information includes a sub-model having a sending node The connection relationship information between the sending node of the sending node and the receiving nodes of other sub-models that have a connection relationship with the sending node; The computing nodes of the other sub-models connected to the receiving node are connected to splicing the multiple sub-models. According to one or more embodiments of the present disclosure, Example 3 provides the method of Example 1, wherein multiple sub-models are in one-to-one correspondence with multiple model training participants, and each model training participant has its own model input data, The inference service executor is one of the multiple model training participants, and the inference service executor obtains the inference result through the target model according to its own model input data. According to one or more embodiments of the present disclosure, Example 4 provides the method of Example 1, where multiple sub-models are in one-to-one correspondence with multiple model training participants, and each model training participant has its own model input data, The inference service executor is one of the multiple model training participants; the inference service executor obtains the inference result in the following manner: Receiving the data sent by other model training participants except the inference service executor Encrypted model input data; obtaining the inference result through the target model according to the model input data of the inference service executor itself and the encrypted model input data of the other model training participants. According to one or more embodiments of the present disclosure, Example 5 provides the method of Example 4, wherein the inference service executor is the model training participant that needs to receive the encrypted model input data of other model training participants with the smallest amount of data. For example, the inference service executor determines by: determining the data volume of the encrypted model input data sent by other model training participants required by each model training participant to perform the inference service; Determined to be the inference service executor. According to one or more embodiments of the present disclosure, Example 6 provides the method of Example 1, wherein multiple sub-models are in one-to-one correspondence with multiple model training participants, and each model training participant has its own model input data, The inference service executor is not the model training participant; the inference service executor obtains the inference result in the following ways: Respectively receive encrypted model input data sent by each of the model training participants; The encrypted model input data of the model training participant is obtained, and the inference result is obtained through the target model. According to one or more embodiments of the present disclosure, Example 7 provides a model processing method, the method comprising: an inference service executor sending a model acquisition request for a target model, where the target model is to convert the multiple obtained by splicing the sub-models; the inference service executor receives the target model, and obtains an inference result through the target model. According to one or more embodiments of the present disclosure, Example 8 provides the method of Example 7, wherein multiple sub-models are in one-to-one correspondence with multiple model training participants, and each model training participant has its own model input data, The inference service executor is one of the multiple model training participants; the obtaining an inference result through the target model includes: inputting data according to the model of the inference service executor itself, and obtaining the inference result through the target model inference result. According to one or more embodiments of the present disclosure, Example 9 provides the method of Example 7, where multiple sub-models are in one-to-one correspondence with multiple model training participants, and each model training participant has its own model input data, The inference service executor is one of the multiple model training participants; the obtaining an inference result through the target model includes: receiving encrypted model inputs sent by other model training participants except the inference service executor data; obtaining the inference result through the target model according to the model input data of the inference service executor itself and the encrypted model input data of the other model training participants. According to one or more embodiments of the present disclosure, Example 10 provides the method of Example 9, where the inference service executor is a model that needs to receive encrypted model input data from other model training participants with the smallest amount of data. Train participants. For example, the inference service executor determines by: determining the data volume of the encrypted model input data sent by other model training participants required by each model training participant to perform the inference service; Determined to be the inference service executor. According to one or more embodiments of the present disclosure, Example 11 provides the method of Example 7, wherein multiple sub-models correspond to multiple model training participants one-to-one, and each model training participant has its own model input data, The inference service executor is not the model training participant; the obtaining the inference result through the target model includes: respectively receiving encrypted model input data sent by each of the model training participants; participating in the model training according to each of the model training participants. The encrypted model input data of the party is used to obtain the inference result through the target model. According to one or more embodiments of the present disclosure, Example 12 provides a model processing system, where the system includes a model optimization platform and a model storage platform; the model optimization platform is configured to acquire multiple sub-models, and store the multiple sub-models The models are spliced to obtain a target model, and the target model is sent to the model storage platform; the model storage platform is used to receive a model acquisition request for the target model sent by the inference service executor , sending the target model to the inference service executor, so that the inference service executor obtains an inference result through the target model. According to one or more embodiments of the present disclosure, Example 13 provides the system of Example 12, and the model optimization platform is configured to obtain model meta information, the model meta information including a sending node having a sub-model of the sending node and a sending node associated with the sending node. the connection relationship information between the receiving nodes of the other sub-models of which the sending node has a connection relationship; the model optimization platform is configured to, according to the model meta-information, connect the computing nodes of the sub-models connected to the sending node and the The computing nodes of the other sub-models connected to the receiving node are connected to splicing the multiple sub-models. According to one or more embodiments of the present disclosure, Example 14 provides a model processing apparatus, the apparatus comprising: an acquisition module configured to acquire a plurality of sub-models; a stitching module configured to A target model is obtained by splicing multiple sub-models; a target model sending module is configured to send the target model to the target model in the case of receiving a model acquisition request for the target model sent by the inference service executor and the inference service executor, so that the inference service executor obtains an inference result through the target model. According to one or more embodiments of the present disclosure, Example 15 provides a model processing apparatus, the apparatus comprising: an acquisition request sending module configured to send a model acquisition request for a target model, wherein the target The model is obtained by splicing the multiple sub-models; the inference module is configured to receive the target model and obtain an inference result through the target model. For example, the model processing means may be provided at the execution side of the inference service. According to one or more embodiments of the present disclosure, Example 16 provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processing apparatus, implements the steps of the method in any one of Examples 1-6 . According to one or more embodiments of the present disclosure, Example 17 provides a computer-readable medium on which a computer program is stored, and when the program is executed by a processing apparatus, implements the steps of the method in any one of Examples 7-11 . According to one or more embodiments of the present disclosure, Example 18 provides an electronic device, including: a storage device, on which a computer program is stored; a processing device, for executing the computer program in the storage device, to The steps of implementing the method of any of Examples 1-6. According to one or more embodiments of the present disclosure, Example 19 provides an electronic device, including: a storage device on which a computer program is stored; and a processing device for executing the computer program in the storage device, to The steps of implementing the method of any of Examples 7-11. The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, and should also cover the technical solutions made of the above-mentioned technical features or Other technical solutions formed by any combination of its equivalent features. For example, a technical solution is formed by replacing the above features with the technical features disclosed in the present disclosure (but not limited to) with similar functions. Additionally, although operations are depicted in a particular order, this should not be construed as requiring that the operations be performed in the particular order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Likewise, although the above discussion contains several implementation-specific details, these should not be construed as limitations on the scope of the present disclosure. Certain features that are described in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Although the subject matter has been described in language specific to structural features and/or logical acts of method, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely example forms of implementing the claims. Regarding the apparatus in the above-mentioned embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be described in detail here.

Claims

Rights request

1. A model processing method, comprising: acquiring multiple sub-models; splicing the multiple sub-models to obtain a target model; in the case of receiving a model acquisition request for the target model sent by an inference service executor, The target model is sent to the inference service executor, so that the inference service executor obtains an inference result through the target model.

2. The model processing method according to claim 1, wherein the splicing the multiple sub-models comprises: acquiring model meta-information, the model meta-information including the sending node of the sub-model having the sending node and the the connection relationship information between the receiving nodes of other sub-models that the sending node has a connection relationship; The computing nodes of the other sub-models are connected to splicing the multiple sub-models.

3. The model processing method according to claim 1, wherein the multiple sub-models are in one-to-one correspondence with multiple model training participants, and each model training participant has its own model input data, and the reasoning The service executor is one of the multiple model training participants, and the inference service executor obtains the inference result through the target model according to its own model input data.

4. The model processing method according to claim 1, wherein the multiple sub-models correspond to multiple model training participants one-to-one, and each model training participant has its own model input data, and the reasoning The service executor is one of the multiple model training participants, and the inference service executor obtains the inference result in the following manner: Receive the encrypted model sent by other model training participants except the inference service executor input data; obtain the inference result through the target model according to the model input data of the inference service executor itself and the encrypted model input data of the other model training participants.

5. The model processing method according to claim 4, wherein the inference service executor is determined by: determining the encrypted model input data sent by other model training participants required by each model training participant to perform the inference service The model training participant with the smallest required data amount is determined as the inference service executor.

6. The model processing method according to claim 1, wherein the multiple sub-models correspond to multiple model training participants one-to-one, and each model training participant has its own model input data, and the reasoning The service executor is not the model training participant, and the inference service executor obtains the inference result in the following ways: respectively receiving encrypted model input data sent by each of the model training participants; training according to each model The encrypted model input data of the participant, and the inference result is obtained through the target model.

7. A model processing method, comprising: an inference service executor sending a model acquisition request for a target model, where the target model is obtained by splicing multiple sub-models; the inference service executor receiving the target model, and The inference result is obtained through the target model.

8. The model processing method according to claim 7, wherein the multiple sub-models are in one-to-one correspondence with multiple model training participants, and each model training participant has its own model input data, and the reasoning The service executor is one of the multiple model training participants; the obtaining the inference result through the target model includes: inputting data according to the model of the inference service executor itself, and obtaining the inference result through the target model .

9. The model processing method according to claim 7, wherein the multiple sub-models are in one-to-one correspondence with multiple model training participants, each of the model training participants has its own model input data, and the inference service executes The party is one of the multiple model training participants; the obtaining the inference result through the target model includes: Receive encrypted model input data sent by other model training participants except the inference service executor; according to the inference service executor's own model input data and the encrypted model input data of the other model training participants , and obtain the inference result through the target model.

10. The model processing method according to claim 9, wherein the inference service executor is determined by: determining the encrypted model input data sent by other model training participants required by each model training participant to perform the inference service The model training participant with the smallest required data amount is determined as the inference service executor.

11. The model processing method according to claim 7, wherein the multiple sub-models are in one-to-one correspondence with multiple model training participants, and each model training participant has its own model input data, and the reasoning The service executor is not the model training participant; the obtaining the inference result through the target model includes: respectively receiving the encrypted model input data sent by each of the model training participants; The encrypted model input data is used to obtain the inference result through the target model.

12. A model processing system, comprising a model optimization platform and a model storage platform, wherein the model optimization platform is used to obtain multiple sub-models, splicing the multiple sub-models to obtain a target model, and sending the target model to The model storage platform, where the model storage platform is configured to send the target model to the inference service executor in the case of receiving a model acquisition request for the target model sent by the inference service executor, to The inference service executor obtains an inference result through the target model.

13. The model processing system according to claim 12, wherein the model optimization platform is used to obtain model meta information, and the model meta information includes a sending node having a sub-model of a sending node and a sending node having a connection with the sending node The connection relationship information between the receiving nodes of other sub-models of the relationship; the model optimization platform is configured to, according to the model meta-information, The computing node of the type is connected with the computing nodes of the other sub-models connected to the receiving node, so as to splicing the multiple sub-models.

14. A model processing device, comprising: an acquisition module, configured to acquire a plurality of sub-models; a splicing module, configured to splicing the plurality of sub-models to obtain a target model; a target model sending module, sent by is configured to send the target model to the inference service executor in the case of receiving a model acquisition request for the target model sent by the inference service executor, so that the inference service executor passes the Describe the target model to get the inference results.

15. A model processing apparatus, comprising: an acquisition request sending module, configured to send a model acquisition request for a target model, wherein the target model is obtained by splicing multiple sub-models; an inference module, configured is used to receive the target model and obtain inference results through the target model.

16. A computer-readable medium on which a computer program is stored, characterized in that, when the program is executed by a processing device, the steps of the method according to any one of claims 1-6 are implemented.

17. A computer-readable medium on which a computer program is stored, characterized in that, when the program is executed by a processing device, the steps of the method according to any one of claims 7-11 are implemented.

18. An electronic device, comprising: a storage device on which a computer program is stored; a processing device for executing the computer program in the storage device to implement the method according to any one of claims 1-6 A step of.

19^ An electronic device, comprising: a storage device on which a computer program is stored; a processing device for executing the computer program in the storage device to implement the method in any one of claims 7-11 A step of.

twenty two

20. A computer program, comprising: instructions that, when executed by a processor, cause the processor to execute the model processing method according to any one of claims 1 to 11.

21. A computer program product comprising instructions that, when executed by a processor, cause the processor to perform the model processing method according to any one of claims 1 to 11.

twenty three