WO2023134275A1 - Model management method, model management service, client end, storage medium, and program - Google Patents

Abstract

Embodiments of the present invention provide a model management method, a model management service, a client end, a storage medium, and a program. The method comprises: if a model management service meets a preset synchronization condition, synchronizing a target model file and target model meta-information of a target model from an object storage service; and in response to a load request of a view resolution service, sending the target model file and the target model meta-information to the view resolution service.

Description

Model management method, model management service, client, storage medium and program

Cross References to Related Applications

The embodiment of the present disclosure is based on the Chinese patent application with the application number 202210044896.8, the application date is January 14, 2022, and the application name is "model management method, model management service, client and storage medium", and requires the Chinese patent application The entire content of this Chinese patent application is hereby incorporated into this disclosure as a reference.

technical field

The embodiment of the present application relates to the field of machine learning, involving but not limited to a model management method, model management service, client, storage medium and program.

Background technique

As the market scale increases, the access scale of view analysis is also increasing. When these models are used on an open vision platform that supports thousands of view analysis services, they occupy a large amount of disk space, and also occupy a large amount of bandwidth and disk space during repeated transmissions, which makes the startup of the intelligent analysis system slow down, and it is very difficult for the storage system and The distributed system brings great pressure to the entire network system.

Contents of the invention

An embodiment of the present application provides a model management method.

The technical scheme of the embodiment of the application is realized in this way:

An embodiment of the present application provides a model management method, which is applied to model management services, and the method includes:

Synchronizing the target model file and target model meta information of the target model from the object storage service when the model management service satisfies a preset synchronization condition;

In response to the loading request of the view resolution service, the target model file and the target model meta information are sent to the view resolution service.

In some embodiments, when the model management service satisfies preset synchronization conditions, before synchronizing the target model file and target model meta information of the target model from the object storage service, the method further includes: receiving the client The upload request of the model sent by the terminal; the model includes: a model file and model meta information; in response to the upload request, upload the model file and the model meta information to the object storage service. In this way, by responding to multiple upload requests from the client, uploading model files and model meta information of multiple models to the object storage service, so that multiple model files and model meta information are stored in the object storage service. Therefore, by separately storing model files and model meta information in the object storage service, the upload speed and transmission speed of the model can be accelerated.

In some embodiments, the uploading the model file and the model meta information to the object storage service in response to the upload request includes: responding to the upload request, uploading the model file and the model The meta information is uploaded to the object storage service according to different storage addresses. In this way, storing model files and model meta information in different locations of the object storage service can realize separate storage of model files and model meta information, saving transmission bandwidth.

In some embodiments, the responding to the upload request, uploading the model file and the model meta information to the object storage service according to different storage addresses includes: based on the model meta information, determining the Whether the object storage service includes the model file; in response to the object storage service not including the model file, upload the model file to the object storage service according to the preset storage address; based on the preset storage address, to update the model meta information; upload the updated model meta information to the object storage service according to the multi-level storage address. In this way, by separately storing the model file and the model meta information in the object storage service, the upload of the model can be completed quickly, and it is convenient for the client to locate the model file based on the updated model meta information.

In some embodiments, the determining whether the model file is included in the object storage service based on the model meta information includes: determining a reference path where the model meta information points to a preset model file; based on the reference The path verifies the model file to obtain a verification result; based on the verification result, it is determined whether the object storage service includes the model file. In this way, through the reference path of the model file pointed to by the model meta information, it can be accurately and conveniently determined whether there is a model file of the model to be uploaded in the object storage service.

In some embodiments, the model management service satisfies preset synchronization conditions, including at least one of the following: the model management service receives an event notification message sent by the object storage service; wherein the event notification message represents the The model change event in the object storage service; the time interval between the historical moment and the current moment when the model management service performs data synchronization reaches the first preset duration; the model management service is not initiated within the second preset duration The data is synchronized, and a model synchronization instruction sent by the client is received; wherein, the second preset duration is longer than the first preset duration.

In some embodiments, in the case that the model management service satisfies the preset synchronization condition, synchronizing the target model file and the target model meta information of the target model from the object storage service includes: determining that in the preset Set a model change event in the object storage service under synchronization conditions; obtain a first model list matching the model change event in the object storage service; determine the target model file in the first model list and the target model meta information; based on the target model file and the target model meta information, synchronize the second model list in the model management service; wherein, the model list includes a plurality of model files, and each model file Corresponding to at least one set of model meta information. In this way, accurate synchronization of the model management service can be achieved by comparing the model list in the model management service with the model list in the object storage service.

In some embodiments, when the model change event includes model addition, determining the target model file and the target model meta information in the first model list includes: In the list, determine the target model file and the target model meta information different from the second model list; based on the target model file and the target model meta information, synchronize the model management service The second model list includes: downloading the target model file and the target model meta information from the object storage service; adding the target model file and the target model meta information to the second model list. In this way, after uploading the target model file in the object storage service, if the model management service obtains the upload message, then the model management service starts the synchronization process; downloads the newly uploaded target model file and the target model meta information from the object storage service to local, and added to the second model list. In this way, model files and model meta-information can be shared between multiple model management services, and data synchronization between multiple model management services and object storage services can be quickly realized.

In some embodiments, when the model change event includes model deletion, determining the target model file and the target model meta information in the first model list includes: receiving the client The model deletion request sent by the end; wherein, the model deletion request carries a third storage address; based on the third storage address, determine the target model element information from the first model list; determine the target model element The target model file pointed to by information in the first model list. In this way, by responding to the client's model deletion request, the records of the target model file and target model meta-information in the object storage service can be deleted through the deletion interface. In this way, the model file and meta-information of the object storage service can be deleted accurately, and the The functions that the model management system can realize.

In some embodiments, the synchronizing the second model list in the model management service based on the target model file and the target model meta information includes: in response to the model deletion request, storing the Delete the target model meta-information and the target model file in the service, so that the object storage service broadcasts the synchronization cleaning instruction; receive the object storage service to delete the target model file and the target model The synchronization cleaning instruction broadcast after the meta information; in response to the synchronization cleaning instruction, deleting the target model file and the target model meta information in the second model list. In this way, after deleting the target model file and target meta-information in the object storage service, the deletion command is broadcast to the model management service, so that multiple model management services can delete the target model file and target model meta-information at the same time, and can quickly realize multiple model management The service synchronizes data with the object storage service.

In some embodiments, the responding to the model deletion request, and deleting the to-be-deleted model meta-information and the to-be-deleted model file in the object storage service includes: responding to the model deletion request , delete the target model meta information in the object storage service; delete the target model file in the object storage service according to a preset time interval. In this way, by setting the interval for regularly cleaning model files, the model management service can clean up the model files to be deleted at a certain interval; in this way, frequent deletion operations can be reduced, and the model management service can also be cleaned regularly.

In some embodiments, the method further includes: in response to receiving the retrieval request of the model meta information sent by the client, determining the index parameter of the multi-level storage address carried in the retrieval request; based on the multi-level storage At least one of the index parameters of the address stores the index parameters of the address, retrieves the model meta information in the object storage service, and obtains the model meta information to be fed back that matches the retrieval request; and stores the model meta information to be fed back feedback to the client. In this way, after the model management service receives the retrieval request, it realizes the fuzzy matching retrieval of the model meta information in the object storage service through at least one level of the index parameters of the multi-level storage address, so that it can quickly retrieve the pending information that matches the retrieval request. Feedback model meta information to enrich the functions of the model management system.

In some embodiments, when the model change event in the object storage service includes model addition, in response to the loading request of the view resolution service, the target model file and the target model meta information are sent to The view parsing service includes: receiving a model download request sent by the view parsing service when the running software toolkit is running; wherein, the running software toolkit includes the interface of the model management service; in response to the A model download request, sending the target model file and the target model meta information to the view parsing service through the interface. In this way, the client software toolkit of the model management service is integrated in the parsing or retrieval service, and the model deployment, distribution, download and use can be completed by calling the relevant interface.

An embodiment of the present application provides a model management method, which is applied to a client. The method includes: sending a loading request of a target model to a model management service; the target model includes: a target model file and target model meta information; receiving the The multi-level storage address in the object storage service of the target model meta-information sent by the model management service based on the loading request; based on the multi-level storage address, determine the target model meta-information and The target model file corresponding to the target model meta information. In this way, the client can synchronously load the meta information of the target model and the target model file from the model management service, so that data synchronization between the client and the model management service can be quickly realized.

In some embodiments, the determining the target model meta-information and the target model file corresponding to the target model meta-information in the object storage service based on the multi-level storage address includes: based on the The multi-level storage address stores the address to locate the target model meta-information in the object storage service; determines the storage address of the target model file carried in the target model meta-information in the object storage service; based on the target The storage address of the model file in the object storage service, and the target model file is located in the object storage service. In this way, the client queries the target model file corresponding to the target model meta information according to the determined storage address of the target model file in the object storage service, so that the client can quickly and accurately query the target model meta from the object storage service Information and target model files.

In some embodiments, the client includes at least one view parsing service, and based on the multi-level storage address, the object storage service determines the target model meta-information and the target model meta-information corresponding After the target model file, the method further includes: caching the target model file and the target model meta information; sharing the target model file and the target model meta information to the at least one view resolution service. In this way, by caching the downloaded target model file and target model meta information on the client side, the target model file and target model meta information cached by the client can be shared between multiple view parsing services on the client side

An embodiment of the present application provides a model management service, the model management service includes:

The first synchronization module is configured to synchronize the target model file and target model meta information of the target model from the object storage service when the model management service meets the preset synchronization conditions; the first response module is configured to respond Sending the target model file and the target model meta information to the view resolution service in response to a loading request of the view resolution service.

An embodiment of the present application provides a client, the client includes: a processor, and a memory storing instructions executable by the processor; wherein the processor is configured to: send the load of the target model to the model management service request; the target model includes: target model file and target model meta information; receive the storage address in the object storage service of the target model meta information sent by the model management service based on the load request; based on the storage address , determining the target model meta-information and the target model file corresponding to the target model meta-information in the object storage service.

Correspondingly, an embodiment of the present application provides a computer storage medium, on which computer-executable instructions are stored. After the computer-executable instructions are executed, the above-mentioned model management method can be implemented.

An embodiment of the present disclosure also provides a computer program, where the computer program includes computer readable codes, and when the computer readable codes run in an electronic device, the processor of the electronic device executes any of the above embodiments The model management method described.

The embodiment of the present application provides a model management method, model management service, client, storage medium, and program. When the model management service meets the preset synchronization conditions, the model management service and the object storage service perform data synchronization, and the data from the object storage The target model file and target model meta information of the target model are synchronized in the service. In this way, storing the target model file and the target model meta-information of a target model separately in the object storage service can increase the transmission bandwidth, thereby improving the startup speed of the intelligent analysis system. After receiving the loading request from the view analysis service, the target model file and target model meta-information are synchronized to the view analysis service through the model management service; in this way, data synchronization through multi-level cache distribution can meet larger scale and greater pressure model requirements.

In order to make the above-mentioned objects, features and advantages of the present disclosure more comprehensible, preferred embodiments will be described in detail below together with the accompanying drawings.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present disclosure more clearly, the following will briefly introduce the accompanying drawings used in the embodiments. The accompanying drawings here are incorporated into the specification and constitute a part of the specification. The drawings show the embodiments consistent with the present disclosure, and are used together with the description to explain the technical solution of the present disclosure. It should be understood that the following drawings only show some embodiments of the present disclosure, and therefore should not be regarded as limiting the scope. For those skilled in the art, they can also make From these drawings other related drawings are obtained.

FIG. 1 is a schematic diagram of the implementation process of the model management method provided by the embodiment of the present application;

FIG. 2 is a schematic flow diagram of another implementation of the model management method provided by the embodiment of the present application;

FIG. 3 is a schematic flow diagram of another implementation of the model management method provided by the embodiment of the present application;

FIG. 4 is a schematic diagram of the implementation flow of the model management method provided by the embodiment of the present application;

FIG. 5 is an implementation architecture diagram of the model management method provided by the embodiment of the present application;

FIG. 6 is an interactive schematic diagram of the implementation process of the object management method provided by the embodiment of the present application;

FIG. 7 is a schematic diagram of the structural composition of the model management service of the embodiment of the present application;

FIG. 8 is a schematic diagram of the composition and structure of the client according to the embodiment of the present application.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the specific technical solutions of the invention will be further described in detail below in conjunction with the drawings in the embodiments of the present application. The following examples are used to illustrate the present application, but not to limit the scope of the present application.

In the following description, references to "some embodiments" describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or a different subset of all possible embodiments, and Can be combined with each other without conflict.

In the following description, the term "first\second\third" is only used to distinguish similar objects, and does not represent a specific ordering of objects. Understandably, "first\second\third" Where permitted, the specific order or sequencing may be interchanged such that the embodiments of the application described herein can be practiced in sequences other than those illustrated or described herein.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are described, and the nouns and terms involved in the embodiments of the present application are applicable to the following explanations.

1) Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. Artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that responds in a manner similar to human intelligence. Research in this field includes robotics, language recognition, image recognition, natural language processing and expert systems, etc. Since the birth of artificial intelligence, the theory and technology have become increasingly mature, and the application fields have also continued to expand. Artificial intelligence can simulate the information process of human consciousness and thinking.

2) Computer vision refers to machine vision that uses cameras and computers instead of human eyes to identify, track and measure targets, and further performs graphics processing to make computer processing images that are more suitable for human eyes to observe or sent to instruments for detection.

3) Object storage, a generic term used to describe methods of addressing and processing discrete units. Objects will no longer have a hierarchical structure in a layer structure, which is characterized by extended metadata.

4) Meta information is information about information, and meta information allows the server to provide information about the data sent; in some embodiments, such as hypertext transfer protocol can improve the object language and objects sent, and can also be implemented with meta information Conditionally request and report transaction completion. The browser receiving the data can determine what content is sent by the server based on the meta-information, what data is expected, whether to receive complete data, and whether there is an error in the process, so that the client can know the type of the transmission object. In this embodiment of the present application, the model meta information is used to describe the detailed information of the model, including the hardware information, software information, and network parameters in the model to which the model is applicable.

The exemplary application of the model management device provided by the embodiment of the present application is described below. The device provided by the embodiment of the present application can be implemented as a notebook computer, a tablet computer, a desktop computer, a mobile device (for example, a personal digital assistant, a dedicated Various types of user terminals such as messaging devices, portable game devices, etc., can also be implemented as servers. Below, an exemplary application when the device is implemented as a terminal or a server will be described.

The method can be applied to electronic equipment, and the functions realized by the method can be realized by calling program codes by the processor in the electronic equipment. Of course, the program codes can be stored in computer storage media. It can be seen that the electronic equipment includes at least a processor and storage medium.

The embodiment of this application provides a model management method, which is applied to model management services, as shown in Figure 1, and will be described in conjunction with the steps shown in Figure 1:

Step S101, if the model management service satisfies the preset synchronization condition, synchronize the target model file and the target model meta information of the target model from the object storage service.

In some embodiments, the preset synchronization condition is used to indicate the condition for the model management service to perform data synchronization with the object storage service; the model management service meeting the preset synchronization condition can be understood as the model management service meeting the condition for data synchronization with the object storage service ; Therefore, when the model management service meets the preset synchronization conditions, the model management service starts the synchronization mechanism to realize data synchronization with the object storage service. The model management service is used to manage the algorithm models uploaded to the model management system. The object storage service is used to receive the model file and model meta information of the model uploaded by the client, and store the model file and model meta information of the model in the storage bucket by means of separate storage. The target model can be any algorithm model, for example, a neural network model for target recognition, or a deep neural network model for semantic segmentation.

In some possible implementations, the model to be uploaded is designed as model meta information and model files, wherein the model meta information describes all information of the model, such as the hardware used by the model, accelerator card, operating environment, model name, etc. The model meta-information may also include description information of hardware, software, or programs added by developers. The model file is the original binary model file, including the implementation program and implementation workflow of the model; the model meta-information and the model file are stored separately, and different model meta-information can point to (or reference) the same model file. In a specific example, taking the target model as an example of a neural network model for target recognition, the model file is the binary data such as the implementation program and algorithm workflow of the neural network model, and the model meta information includes the applicable hardware parameters of the neural network model, Relevant parameters of the accelerator card, operating environment, storage path, description information of the program, and the name of the neural network model, etc.

In some possible implementations, after the client sends a model upload request, both the model meta-information and the model file of the model are uploaded to the object storage service, and the model meta-information and the model file are stored separately in the object storage service; thus, Separating model meta information and model files can save storage space and transmission bandwidth.

Step S102, in response to the loading request of the view resolution service, sending the target model file and the target model meta information to the view resolution service.

In some embodiments, the view parsing service uses a software toolkit provided by the model management service to download a model target model file and the target model meta information. The view analysis service analyzes the input data (for example, image or video data, etc.) in the connected model on the open vision platform of computer vision, so as to realize the function of the model. For example, taking the identification of traffic accidents in traffic images as an example, identify whether there are traffic accidents in traffic videos or images; this process can be divided into two parts. Business logic of accident recognition), input traffic video or image into the model; then, analyze and decode the input video or image, analyze each frame after decoding; finally, input the vehicle detection frame of each frame into In the algorithm related to the algorithm model, the model will be called when the algorithm is started, and the network parameters and operators will be loaded to infer whether there is a traffic accident.

In some possible implementation manners, the model management service synchronizes the target model file and target model meta information from the object storage service, so that the view parsing service downloads the target model file and target model meta information from the model management service according to the reference path. The model management service provides a software toolkit, and the view parsing service integrates the software toolkit and can call functions in the toolkit. The running view parsing service sends a model download request to the model management service to read the target model file and target model meta information through the local disk of the model management service, so as to synchronize the target model file and target model meta information to the view parsing service .

In the embodiment of this application, when the model management service meets the preset synchronization conditions, data synchronization is performed with the object storage service, and the target model file and target model meta information of the target model are synchronized from the object storage service. In this way, storing the target model file and the target model meta-information of a target model separately in the object storage service can save storage space and transmission bandwidth, thereby improving the startup speed of the intelligent analysis system. Synchronize the target model file and target model meta information to the view parsing service through the model management service. In this way, the model management service synchronizes the target model file and target model meta information from the object storage service, and synchronizes the target model file and target model meta information to the view parsing service. In this way, data synchronization through multi-level cache distribution can Meet the needs of models with larger scale and higher pressure.

In some embodiments, by responding to the client's model upload request, the model files and model meta information of multiple models are uploaded to the object storage service, that is, before the above step S101, steps S201 and S202 as shown in FIG. 2 are also included. , in conjunction with the steps shown in Figure 2 for the following description:

Step S201, receiving a model upload request sent by a client.

In some embodiments, the model includes: a model file and model meta information. On the client side, the model to be uploaded is separated into a model file and model meta information, and uploaded. Carry the model file and model meta information of the model in the upload request. The model management service receives the upload request sent by the client, and manages the object storage service in response to the upload request.

Step S202, uploading the model file and the model meta information to the object storage service in response to the upload request.

In some embodiments, the model management service uploads the model file and model meta information carried in the upload request to the object storage service in response to the upload request. After the object storage service receives the uploaded model files and model meta information, it stores the model files and model meta information in different storage paths. In this way, by responding to multiple upload requests from the client, uploading model files and model meta information of multiple models to the object storage service, so that multiple model files and model meta information are stored in the object storage service. In this way, by separately storing model files and model meta-information in the object storage service, the uploading speed and transmission speed of the model can be accelerated.

In some embodiments, the model file and model meta information are stored according to different storage paths, that is, the above step S202 may be to respond to the upload request and upload the model file and the model meta information to the The object storage service described above. In this way, storing model files and model meta information in different locations of the object storage service can realize separate storage of model files and model meta information, saving transmission bandwidth.

In some embodiments, after receiving the upload request, determine whether to upload the model file and meta information of the model to the object storage service by judging whether the model file corresponding to the upload request already exists in the object storage service, that is, the above steps S202 can be realized through the following steps S221 to S224 (not shown in the figure):

Step S221, based on the model meta information, determine whether the object storage service includes the model file.

In some embodiments, after receiving the model upload request from the client, the model management service searches the object storage service for a model file matching the model meta information according to the model meta information of the model to be uploaded; if the object storage If there is a model file matching the model meta information in the service, it is not necessary to upload the model file to the object storage service; if there is no model file matching the model meta information in the object storage service, upload the model file to the object storage service.

In some possible implementations, by checking the reference path of the model meta-information to the preset model file and the model file, it can be determined whether the object storage service includes the model file, that is, the above step S221 can be implemented by the following steps:

The first step is to determine the reference path of the model meta-information pointing to the preset model file.

In some embodiments, the preset model file may be the model file pointed to by the model meta information in the object storage service, and it can also be understood that the preset model file may be the model referenced by the model meta information in the object storage service document. Each model meta information will point to a model file, that is, each model meta information will refer to a model file, and different model meta information can also refer to the same model file; after determining the model meta information of the model to be uploaded, you can Determine the reference path that the model meta information points to the preset model file that matches itself in the object storage service.

In the second step, the model file is verified based on the reference path, and a verification result is obtained.

In some embodiments, according to the reference path, it can be searched in the object storage service whether there is a model file under the reference path, and whether the existing model file is the model file of the model to be uploaded, so as to realize the verification of the model file . Based on this, the verification result is that there is no model file under the reference path, or the model file under the reference path is not the model file of the model to be uploaded, or the model file under the reference path is the model file of the model to be uploaded.

The third step is to determine whether the object storage service includes the model file based on the verification result.

In some embodiments, if the verification result is that the object storage service does not have a model file under the reference path, or the model file under the reference path is not a model file of the model to be uploaded, it is determined that the object storage service does not include the Model file; if the verification result shows that the model file under the reference path is the model file of the model to be uploaded, it is determined that the object storage service includes the model file, that is, the model to be uploaded already exists in the object storage service. In this way, through the reference path of the model file pointed to by the model meta information, it can be accurately and conveniently determined whether there is a model file of the model to be uploaded in the object storage service.

Step S222, in response to the object storage service not including the model file, upload the model file to the object storage service according to a preset storage address.

In some embodiments, if the object storage service does not include the model file, it means that the model to be uploaded corresponding to the upload request has not been uploaded to the object storage service. Therefore, upload the model file to the object storage according to the preset storage address service and stored in the object storage bucket or local disk.

Step S223, based on the preset storage address, update the model meta-information.

In some embodiments, the model management service obtains the storage address of the model file in the object storage service, for example, the model file is stored in a local disk of the object storage service, then the local storage path is used as the preset storage address. The preset storage address is added to the model meta-information to update the model meta-information. In this way, the updated model meta-information includes detailed description information of the model and user-defined meta-information.

Step S224, uploading the updated model meta information to the object storage service according to the multi-level storage address.

In some embodiments, after the model management service updates the model meta information according to the preset storage address, the updated model meta information, that is, the model meta information carrying the preset storage address is uploaded to the object storage service.

In some possible implementations, after the model management service uploads the updated model meta information to the object storage service, it obtains the storage address of the updated model meta information in the object storage service, that is, the multi-level storage address; multi-level The storage address can be the same as or different from the preset storage address. For example, the updated model meta information is stored according to the multi-level storage address, the model file is stored according to the fixed address, the source storage is in the object storage service, and other synchronized data are stored through the multi-level disk cache. The multi-level storage address is the multi-level storage directory, including at least two levels: type→subtype→hardware type→software type→model name. According to at least two levels of "type→subtype→hardware type→software type→model name", the model meta information is stored in the object storage service.

In the embodiment of the present application, by checking whether the model file to be uploaded has been stored in the object storage service, if the model file to be uploaded is not stored in the object storage service, upload the model file to the object storage service; and Update the model meta information through the storage address of the model file, so that the updated model meta information can be stored in the object storage service independently of the model file, and the separate storage of the model file and model meta information in the object storage service can be realized quickly. The upload of the model is completed, and it is convenient for the client to locate the model file based on the updated model meta information.

In some embodiments, when any of the following preset synchronization conditions are met, the model management service starts a synchronization mechanism to synchronize the target model file and target model meta information from the object storage service, wherein the preset synchronization conditions include at least the following one:

Condition 1: The model management service receives the event notification message sent by the object storage service.

Here, the event notification message represents a model change event in the object storage service; the model change event is an event that the number of model files or model meta information stored in the object storage service changes. That is, when the number of models in the object storage service changes, the object storage service feeds back an event notification message to the model management service. For example, if a new model file and model meta-information are added to the object storage service, then the object storage service will feed back a message about the addition of the model to the model management service.

Condition 2: The time interval between the historical moment when the model management service performs data synchronization and the current moment reaches a first preset duration.

Here, the first preset duration is set, so that the model management service performs data synchronization every certain preset duration; therefore, when the time interval between the historical moment of the last data synchronization and the current moment reaches the set second After a preset time period, the next data synchronization is performed.

Condition three: the model management service does not initiate data synchronization within a second preset time period, and receives a model synchronization request sent by the client.

Here, the second preset duration is longer than the first preset duration. If the last time the model management service performed data synchronization has reached the first preset time from the current time, but the model management service still has not started the data synchronization mechanism, then the user can manually trigger the interface of the model management service through the client to start data synchronization. That is, the client sends a model synchronization request to the interface of the model management service, so that the model management service performs data synchronization in response to the model synchronization request.

In the embodiment of this application, no matter which of the above synchronization conditions the model management service meets, data synchronization with the object storage service is performed, and then the model management service synchronizes the data to the view analysis service. From the model management service to the object storage service, the multi-level distribution method can meet the needs of larger-scale and higher-pressure model downloads.

In some embodiments, after the model management service reaches the synchronization condition, the model file and model meta information in the model management service are synchronized according to the model change event in the object storage service under the synchronization condition, that is, the above step S101 can be performed through the The steps shown are implemented:

Step S301, determining a model change event in the object storage service under the preset synchronization condition.

In some embodiments, when the model management service satisfies a preset synchronization condition, an event in which the number of models in the object storage service changes under the preset synchronization condition is determined; for example, a model addition event or a model deletion event.

Step S302, obtaining a first model list in the object storage service that matches the model change event.

In some embodiments, the first model list includes a plurality of model files, and at least one set of model meta information corresponding to each model file. The first model list matching the model change event may be a model list in the object storage service after the model change event occurs, that is, multiple model files stored in the object storage service after the model change event occurs, and the model corresponding to each model file meta information.

Step S303, in the first model list, determine the target model file and the target model meta information.

In some embodiments, if the model change event is an event of adding a model, then in the first model list, determine the target model file and the target model element that are different from the second model list in the model management model information; if the model change event is an event of deleting a model, then the target model file and target model meta information are the model files and model meta information that need to be deleted as indicated in the deletion request.

Step S304, based on the target model file and the target model meta-information, synchronize the second model list in the model management service.

In some embodiments, according to the target model file and target model meta information in the object storage service, the model management service performs data synchronization, that is, updates its own second model list. For example, if the target model file and target model meta information are added to the object storage service, then the model management service downloads the target model file and target model meta information from the object storage service to be added to the second model list synchronously.

In the embodiment of the present application, accurate synchronization of the model management service can be realized by comparing the model list in the model management service with the model list in the object storage service.

In some possible implementations, the model change event is different, and the process of synchronizing the second model list is different. When the model change event includes model addition, by downloading the target model file and target model element from the object storage service information to realize data synchronization; that is, the above steps S303 and S304 can be realized through the following steps S341 to S343 (not shown):

Step S341, in the first model list, determine the target model file and the target model meta information that are different from the second model list.

In some embodiments, if the target model file and target model meta information are uploaded in the object storage service, then the target model file and target model meta information are added to the first model list; thus, by comparing the second model list with The difference between the first model lists can determine the model files and model meta information added in the first model list to obtain the target model file and target model meta information.

Step S342, downloading the object model file and the object model meta information from the object storage service.

In some embodiments, if the client uploads the model file and target meta information to the object storage service, an event notification message may be generated in the object storage service, and the event notification message may be sent to the model management service through the memory message queue to The model management service is notified to perform data synchronization; after receiving the event notification message, the model management service starts a synchronization mechanism and downloads the target model file and the target model meta information from the object storage service.

It may also be that the time interval for data synchronization by the model management service reaches a first preset duration, the model management service starts a synchronization mechanism, and downloads the target model file and the target model meta-information from the object storage service to a local cache.

It may also be that the model management service receives the model synchronization instruction sent by the client to start the synchronization mechanism, and downloads the target model file and the target model meta-information from the object storage service to the local cache.

Step S343, adding the target model file and the target model meta information to the second model list.

In some embodiments, the target model file and the meta information of the target model are cached locally, and the notification is added to the second model list to realize data synchronization of the second model list; here, each model management service performs real-time Synchronization, or each model management service shares the second model list, which can reduce the space occupancy rate of data storage.

In this embodiment of the application, after the target model file is uploaded in the object storage service, if the model management service obtains the upload message, the model management service starts the synchronization process; the newly uploaded target model file and the target The model meta information is downloaded locally and added to the second model list. In this way, model files and model meta-information can be shared between multiple model management services, and data synchronization between multiple model management services and object storage services can be quickly realized.

In some embodiments, when the model change event includes model deletion, the model file and model meta information to be deleted in the first model list are the target model file and target model meta information, that is, the above step S303, can be Realized by the following steps S331 to S333 (not shown):

Step S331, receiving a model deletion request sent by the client.

In some embodiments, the model deletion request carries a third storage address. The third storage address is used to indicate the storage address of the model meta information to be deleted in the object storage service. When the client needs to delete the model meta information of any model, it can send a deletion request carrying the third storage address to the model management service, so that the model management service can delete the corresponding model meta information in response to the deletion request.

Step S332, based on the third storage address, determine the target model meta information from the first model list.

Here, according to the third storage address, the model meta information to be deleted under the address is searched from the model meta information stored in the object storage service, that is, the target model meta information is obtained.

Step S333, determining the target model file pointed to by the target model meta information in the first model list.

In some embodiments, since the model meta information has a uniquely referenced model file, the storage address can be obtained according to the path of the model file referenced by the model meta information; the storage address is the model file referenced by the target model meta information Therefore, the target model file pointed to by the target metadata in the object storage service can be obtained.

In this embodiment of the application, by responding to the client’s model deletion request, the records of the target model file and target model meta information in the object storage service can be deleted through the deletion interface, and the model file and meta information of the object storage service can be deleted accurately. It can also enrich the functions that the model management system can realize.

In some embodiments, when the model change event includes model deletion, by responding to the deletion instruction broadcast by the object storage service, the target model file and the target model meta information are deleted to achieve data synchronization; that is, the above step S304 can be performed through the following steps S344 to S346 (illustration not shown) realize:

Step S344, in response to the model deletion request, delete the target model meta-information and the target model file in the object storage service, so that the object storage service broadcasts the synchronization cleaning instruction.

In some embodiments, after receiving the model deletion request sent by the client, the target model file and target model meta information to be deleted are determined. First, delete the target model file and target model meta-information in the object storage service; then, after deleting the target model file and target model meta-information, the object storage service broadcasts a synchronous cleaning command to each model management service, so that each model management service Do a sync cleanup. In the object storage service, the target model file and the target model meta information can be deleted at the same time; the target model file and the target model meta information can also be deleted asynchronously. For example, in the object storage service, the target model meta information is deleted first, and after a certain interval Delete the target model file.

Step S345, receiving a synchronous cleaning instruction broadcast by the object storage service after deleting the target model file and the target model meta information.

In some embodiments, the client sends a model deletion request to the model management service, and the model management service sends the model deletion request to the object storage service, so as to delete the target model file and object specified in the model deletion request in the object storage service Model meta information; while the object storage service deletes the target model file and target model meta information, it broadcasts a synchronous cleanup command to each model management service to prompt each model management service to perform synchronous cleanup.

Step S346, in response to the synchronization cleaning instruction, delete the target model file and the target model meta information in the second model list.

In some embodiments, the model management service receives the synchronous cleaning instruction broadcast by the object storage service, deletes the target model file and the target model meta information, and simultaneously updates the second model list.

In this embodiment of the application, after deleting the target model file and target meta-information in the object storage service, the deletion command is broadcast to the model management service, so that multiple model management services can delete the target model file and target model meta-information at the same time, which can quickly Realize data synchronization between multiple model management services and object storage services.

In some possible implementations, by setting the timing cleaning duration, the model management service can regularly delete the target model files, that is, the above step S344 can be realized through the following steps:

In the first step, in response to the model deletion request, delete the meta information of the target model in the object storage service.

In some embodiments, after receiving the model deletion request sent by the client, the model management service immediately deletes the meta information of the target model in the object storage service in response to the model deletion request.

The second step is to delete the target model file in the object storage service according to a preset time interval.

In some embodiments, the model management service sets a preset time interval, and whenever the preset time interval is reached, the model management service determines whether there is a target model file that needs to be deleted in the object storage service; if there is a need in the object storage service If the target model file is deleted, the model management service will clean up the target model file in the object storage service; at the same time, the object storage service will broadcast a synchronous cleaning command to the model management service according to the preset time interval. Therefore, the model management service can clean up the model files and model meta information to be deleted in the second model list according to the preset time interval.

In the embodiment of this application, by setting the interval for cleaning model files at regular intervals, the model management service can clean up the model files to be deleted at a certain interval; in this way, frequent deletion operations can be reduced, and model management can also be performed regularly The service cleans up.

In some embodiments, the model management system also has a fast retrieval function, through the multi-level reference path, realizes the fast retrieval of fuzzy matching, and retrieves the required model meta-information, which can be realized through the following steps:

In the first step, in response to receiving the retrieval request of the model meta information sent by the client, determine the index parameter of the multi-level storage address carried in the retrieval request.

In some embodiments, the retrieval request is a request sent by the client to the model management service to retrieve model meta information. The index parameter of the multi-level storage address may be a multi-level storage path where the meta information is stored in the object storage service. The index parameter of the multi-level storage address corresponds to the multi-level storage address, that is, the index parameter of the multi-level storage address can also be: type→subtype→hardware type→software type→model name.

The second step is to query the model element information in the object storage service based on the index parameters of at least one level of storage address among the index parameters of the multi-level storage address, and obtain the model element to be fed back that matches the retrieval request information.

In some embodiments, the index parameter of at least one level of storage address may be the first level or the last level of the index parameters of the multilevel storage address, or any level or level of the index parameters of the multilevel storage address. multi-level etc.

In some possible implementations, the index parameters of at least one level of storage address include the first level and the last level of the index parameters of the multi-level storage address, then according to the index parameters corresponding to the first level and the last level in the object storage In the service, fast retrieval of fuzzy prefix and suffix matching is performed on the model meta information, and the retrieved model meta information is returned, that is, the model meta information to be fed back.

The third step is to feed back the meta information of the model to be fed back to the client.

In some embodiments, the model meta information returned by the retrieval function is fed back to the client, so that the client can cache the model meta information locally so that the model meta information can be shared among multiple view parsing services.

In this embodiment of the application, after the model management service receives the retrieval request, it realizes the fuzzy matching retrieval of the model meta information in the object storage service through at least one level of the index parameters of the multi-level storage address, so that the matching can be quickly retrieved. The model meta-information to be fed back in the retrieval request enriches the functions of the model management system.

In some embodiments, when the model change event in the object storage service includes a model addition, by integrating the software toolkit (Software Development Kit, SDK) of the model management service in the model analysis service, thereby calling the The interface in the SDK can realize the distribution and download of the model, that is, the above step S102 can be realized through the following steps S121 and S122 (not shown in the figure):

Step S121, receiving a model download request sent by the view parsing service when the software toolkit is running.

In some embodiments, the SDK includes the interface of the model management service; or, the SDK is obtained by integrating the interface of the model management service. After uploading the target model file and target model meta information to the object storage service, integrate the client SDK of the model management service in the view analysis service; when the view analysis service is running the SDK, if data synchronization is required, it will send the model management service The service sends a model download request.

Step S122, in response to the model download request, sending the target model file and the target model meta information to the view resolution service through the interface.

In some embodiments, in response to the model download request, the model management service can transmit the target model file and target model meta information to the view parsing service by calling a relevant interface in the SDK. In this way, the client SDK of the model management service is integrated in the parsing or retrieval service, and the model deployment, distribution, download and use can be completed by calling the relevant interface.

The embodiment of the present application provides a model management method, which is applied to the client. The client can locate and query the model meta information through the storage address of the model meta information, as shown in FIG. 3B . The following description will be made in conjunction with the steps shown in FIG. 3B:

Step S321, sending a loading request of the target model to the model management service.

Here, the target model includes: target model file and target model meta information.

Step S322, receiving the multi-level storage address of the target model meta information in the object storage service sent by the model management service based on the loading request.

Here, the storage address of the target model meta-information in the object storage service is a multi-level storage address, and the target model meta-information carries the multi-level storage address of the target model file in the object storage service.

Step S323, based on the multi-level storage address, determine the target model meta-information and the target model file corresponding to the target model meta-information in the object storage service.

Here, the model management service sends the updated storage address of the model meta-information in the object storage service to the client, so that the client can read the target model meta-information based on the multi-level storage address, and based on the The preset storage address of the model file, locate and query the target model file in the object storage service. In this way, by feeding back the storage address of the target model meta information in the object storage service to the client, the client can accurately locate the target model file corresponding to the target model meta information by querying the target model meta information.

In some embodiments, the client locates the target model meta information in the object storage service according to the multi-level storage address sent by the model management service, and queries the target model file corresponding to the target model meta information, that is, the above step S323 can be passed The following steps are implemented:

The first step is to locate the target model meta information in the object storage service based on the multi-level storage address.

Here, according to the multi-level storage address, the client can query and locate the meta information of the target model in the object storage service.

The second step is to determine the storage address of the target model file carried in the target model meta-information in the object storage service.

Here, when storing the model file and model meta information, the storage address of the model file in the object storage service is added to the model meta information, so after querying the target model meta information, it can be determined in the target model meta information The storage address of the target model file in the object storage service.

The third step is to locate the target model file in the object storage service based on the storage address of the target model file in the object storage service.

Here, the client can accurately locate and query the target model file corresponding to the target model meta information according to the determined storage address of the target model file in the object storage service.

In some embodiments, the client can also download the model, and cache the downloaded model file and model meta information locally for sharing by other analysis services, which can be achieved through the following steps:

The first step is to cache the target model file and the target model meta information.

In some embodiments, after downloading the target model file and target model meta information from the object storage service, the client may cache the target model file and target model meta information locally.

In the second step, the target model file and the target model meta information are shared with the at least one view resolution service.

In some embodiments, the client caches the target model file and target model meta information locally, so that the target model file and target model meta information cached by the client can be shared among multiple view parsing services of the client.

In the embodiment of the present application, by caching the downloaded target model file and target model meta-information on the client side, it can be easily shared with other analysis services on the model management service, thereby saving the storage space of the client side.

In the following, an exemplary application of the embodiment of the present application in an actual application scenario will be described, taking an object storage-based model distribution management system as an example for illustration.

The artificial intelligence algorithm generates a large number of model files in the process of continuous iteration, and the algorithm parameters are more and more, and the model is getting bigger and bigger; as the market scale increases, the access scale of view analysis is also increasing. When these models are used on the open vision platform that supports thousands of view analysis services, they will be packaged into the image along with the application, occupying a large amount of disk space, and also occupying a large amount of bandwidth and disk input/output during repeated transmission. On the one hand Slowing down the startup of the intelligent analysis system, on the other hand, puts great pressure on the storage system and the entire network system of the distributed system.

The embodiment of the present application provides a model management method. The management system provided by the embodiment of the present application is based on the idea of object storage system and multi-level disk cache distribution, which can decouple the algorithm model and the intelligent analysis program, and realize efficient transmission and storage of the algorithm model. This model management method can be implemented through the following steps:

The first step is to implement algorithm model upload and back-end storage functions.

In some embodiments, the implementation process of algorithm model uploading and back-end storage functions is shown in the following steps 1.1 to 1.4, wherein:

Step 1.1, design the algorithm model as model meta information (meta) and model file.

Here, the model meta-information describes the information of the algorithm model, for example, describes which hardware the model is used for, which accelerator card, the environment in which the model runs, the name of the model, and so on. The model file is the original binary model file; the model meta-information and the model file are stored separately, and different meta-information can point to (reference) the same model file. For the application, the model file can be located through the meta information, and the model file cannot be queried and located without the meta information.

Step 1.2: Provide an upload service interface to upload the algorithm model. After verifying the model file and reference path, if the algorithm model does not exist on the server, upload the algorithm model to the back-end object storage service.

Here, the process of uploading the algorithm model is shown in Figure 4, and Figure 4 is a schematic diagram of the implementation flow of the model management method provided by the embodiment of the present application, and the following description is made in conjunction with the steps shown in Figure 4:

Step S401, start the process of uploading the algorithm model.

Step S402, judging whether the algorithm model already exists in the object storage service.

Here, if the algorithm model already exists in the object storage service, there is no need to upload the algorithm model. If the algorithm model does not exist in the object storage service, go to step S403. If the algorithm model exists in the object storage service, end the whole process.

Step S403, upload the model file of the algorithm model to the object storage service through the model management service.

Step S404, upload the meta information of the algorithm model to the object storage service, if the upload fails, perform data rollback.

Here, if uploading the meta information of the algorithm model to the object storage service fails, the data will be rolled back.

Step S405, end the process of uploading the algorithm model.

Step 1.3, provide the delete function in the model management system, that is, delete the interface and delete the record of the algorithm model in the object storage service, execute the delete operation asynchronously in the actual algorithm model in the background, and delete the disk cache data on the model management service.

In step 1.4, the model management system provides a fast search function for fuzzy matching of prefixes and suffixes of the algorithm model.

Here, the fast retrieval function of algorithm model suffix and suffix fuzzy matching is realized through the multi-level reference path; the retrieval function can return model meta information, which includes model detailed information and user-defined meta information.

In the embodiment of this application, the model management system can store the models separately without packaging them into the program image, so as to ensure that the sharing and reference of the same model does not generate additional space occupation and transmission time bandwidth costs.

The second step is to implement the algorithm model distribution mechanism.

In some embodiments, in a large-scale view parsing platform, multi-level caching and model synchronous distribution are used to solve the problem of model pull, which can be achieved through the following process:

First, synchronize the object storage service data to the model management service in real time.

Second, the view resolution service downloads the model from the model management service to the view resolution service.

Finally, all analysis services of the view analysis service can share and use model data to cope with the download pressure of large-scale model files.

As shown in Figure 5, Figure 5 is an implementation architecture diagram of the model management method provided by the embodiment of the present application, wherein the minio-service 51 (minio-service) can be applied to other object storage services that support the S3 interface; the model management service 52 includes : model management 1 (model-manager-1), model management 2 (model-manager-2), model management n (model-manager-n) refers to n model management service nodes (corresponding to the model management in the above-mentioned embodiment service); node cache 501 to node cache 503 (node-cache) refer to the local cache of the view resolution service. As can be seen in Figure 5, first, the system includes multiple view resolution services: node 0 (node-0), node x (node-x) node n (node-n); wherein, each node includes multiple A worker process (worker) that processes subtasks in parallel, for example, node-0 includes 8 parallel worker processes, from worker process 0 to worker process 7, and caches the data processed by worker process 0 to worker process 7 in the node cache 501 middle; the node-x includes a plurality of parallel working processes, and caches the data processed by the multiple working processes in the node cache 502; the node-n includes m parallel working processes, wherein m is an integer greater than 1 ; Work process 0 to work process m, and cache the data processed by work process 0 to work process m in the node cache 503 ; Then, send the cached data in each node cache to the model management service; for example, the data in node caches 501 and 502 are sent to model management 1, and the data in node cache 503 is sent to model management 2 and model management n . Finally, send the data received in each model management service to the object storage service 504 . In this way, the back-end storage of the model management and distribution system is developed based on a compatible S3 interface, supports various object storage systems, and provides model management (upload, download, enumeration, and deletion, etc.) functions;

In some embodiments, the algorithm model distribution mechanism can be implemented through the following steps 2.1 to 2.4, wherein:

Step 2.1, realizing real-time synchronization of node model data.

Here, the event notification mechanism of the object storage service is used to realize the real-time synchronization of the node model data by listening to the write event of a specific bucket (bucket). If the algorithm model file is uploaded to the object storage service, the event is notified to the model management service through the memory message queue; the model management service starts the synchronization process after listening to the event; the model management service compares the model list in the model management service with the object management The difference in the current model file list in the service (for example, the object management service includes the model information list 1, the model management service has the model information list 2, and the model management service compares the model information list 1 with the model information list 2), and these The newly uploaded algorithm model is downloaded from the object storage service to the local cache.

Step 2.2, if any model management service receives a model deletion command, delete the model on the object storage service first, and broadcast to each model storage service through the event mechanism. After these model storage services receive the notification, the background starts to delete the corresponding model .

Here, the model files and model meta files stored in each model management service are the same.

In step 2.3, in order to reduce the omission of event notifications or untimely processing, a timing synchronization process is added in the background. After the timer expires, a synchronization is performed to synchronize the alignment model list from the object storage service.

Here, if the timeout synchronization mechanism is still invalid, a synchronization command can also be manually issued through the synchronization trigger interface.

In step 2.4, in order to reduce the frequent delete operations, only the model meta information is deleted when receiving the model delete command, and the binary model files are not deleted immediately, but are regularly cleaned up in the background.

Here, a sentinel object is set in the object storage service. When these models that are no longer referenced are cleaned up, the sentinel object is updated, and all model management services are notified to perform the cleaning operation synchronously. In this way, the efficient model upload and synchronization is guaranteed through the event notification and timing trigger mechanism, and can be distributed to the model management service and cached in real time. The view analysis service downloads and caches the target model file and target model meta information from the model management service. All view analysis services The usage model can be shared; thus, the storage space can be saved and the synchronization of multiple view parsing services can be realized.

In the embodiment of the present application, model pull is realized through multi-level caching and model synchronous distribution, as shown in Figure 6, which is a schematic diagram of the implementation process interaction of the object management method provided in the embodiment of the present application, combined with the The steps are described below:

Step S601, the client uploads the model to the first model management service.

Here, uploading the model to the first model management service by the client includes uploading the model file and model meta information.

Step S602, the first model management service uploads the model to the object storage service.

Step S603, the object storage service feeds back the notification information of upload completion to the first model management service.

Step S604, the first model management service feeds back the notification information of upload completion to the client.

Step S605, the object storage service feeds back the generated new model event notification to the first model management service, and at the same time sends the new model event notification to the second model management service.

Step S606, the first model management service pulls the model from the object storage service.

In step S607, the object storage service feeds back model pull completion information to the first model management service.

Step S608, the second model management service synchronously pulls the model from the object storage service.

In step S609, the object storage service feeds back model pull completion information to the second model management service.

Here, step S606 and step S608 are executed synchronously. That is, after the object storage service sends a new model event notification to each model management node, each model management node pulls the model from the object storage service synchronously.

In the third step, the view analysis service shares and uses the model uploaded to the object storage service.

In some embodiments, after the model is distributed to the model management service, after the view analysis service integrates the client software development kit of the model management service interface, the SDK is installed, and when the running view analysis service downloads the model, it will pass the corresponding model management service The local disk reads the model; after the client node downloads the model, it also caches the model locally so that it can be shared with other parsing services on the node to use the model. In this way, model management and application usage are facilitated by providing client tools and SDKs.

In the embodiment of this application, when facing large-scale access and hundreds of gigabytes of model files on the view analysis platform, users can choose to deploy model management services to the system; and separate view analysis or retrieval programs and model files, through Use the client SDK of the model management system to upload the model separately, and integrate the client SDK of the model management system in the parsing or retrieval service, call the relevant interface to complete the model deployment, distribution and downloading are applicable. In this way, by separating the models, the image size of the view parsing service is greatly reduced, saving storage space and transmission bandwidth; this model distribution cache method can meet the needs of larger-scale and higher-pressure model downloads, and the model management service can be elastically scaled; through structured The model meta information describes the algorithm model, manages the model meta information and the model file separately, and the same model can be referenced by different paths to realize asynchronous cleaning.

The embodiment of the present application provides a model management device. FIG. 7 is a schematic diagram of the structure and composition of the model management service in the embodiment of the present application. As shown in FIG. 7, the model management service 700 includes:

The first synchronization module 701 is configured to synchronize the target model file and target model meta information of the target model from the object storage service when the model management service meets the preset synchronization conditions; the first response module 702 is configured In response to the loading request of the view resolution service, the target model file and the target model meta information are sent to the view resolution service.

In some embodiments, when the model management service satisfies preset synchronization conditions, the model management service further includes: a first receiving module configured to receive a model upload request sent by a client; The model includes: a model file and model meta information; a first upload module configured to respond to the upload request and upload the model file and the model meta information to the object storage service.

In some embodiments, the first upload module is further configured to, in response to the upload request, upload the model file and the model meta information to the object storage service according to different storage addresses.

In some embodiments, the first uploading module includes: a first determining submodule configured to determine whether the object storage service includes the model file based on the model meta information; the first uploading submodule , configured to upload the model file to the object storage service according to a preset storage address in response to the object storage service not including the model file; the first update submodule is configured to based on the preset The storage address is set to update the model meta-information; the second uploading submodule is configured to upload the updated model meta-information to the object storage service according to the multi-level storage address.

In some embodiments, the first determining submodule includes: a first determining unit configured to determine the reference path of the model meta-information pointing to a preset model file; a first checking unit configured to Verifying the model file with the reference path to obtain a verification result; the second determination unit is configured to determine whether the object storage service includes the model file based on the verification result.

In some embodiments, the model management service satisfies preset synchronization conditions, including at least one of the following: the model management service receives an event notification message sent by the object storage service; wherein the event notification message represents the The model change event in the object storage service; the time interval between the historical moment and the current moment when the model management service performs data synchronization reaches the first preset duration; the model management service is not initiated within the second preset duration The data is synchronized, and a model synchronization instruction sent by the client is received; wherein, the second preset duration is longer than the first preset duration.

In some embodiments, the first synchronization module 701 includes: a second determination submodule configured to determine a model change event in the object storage service under the preset synchronization condition; a first acquisition submodule , configured to obtain a first model list in the object storage service that matches the model change event; a third determining submodule, configured to determine the target model file and the target model file in the first model list The target model meta information; the first synchronization submodule is configured to synchronize the second model list in the model management service based on the target model file and the target model meta information; wherein the model list includes a plurality of models files, and at least one set of model meta information corresponding to each model file.

In some embodiments, when the model change event includes model addition, the third determining submodule includes: a third determining unit configured to determine, in the first model list, the The target model file and the target model meta information are different in the second model list; the first synchronization submodule includes: a first download unit configured to download the target model file from the object storage service and the target model meta information; a first adding unit configured to add the target model file and the target model meta information to the second model list.

In some embodiments, when the model change event includes model deletion, the third determining submodule includes: a first receiving unit configured to receive a model deletion request sent by the client; wherein, The model deletion request carries a third storage address; the fourth determination unit is configured to determine the target model meta information from the first model list based on the third storage address; the fifth determination unit is configured To determine the target model file pointed to by the target model meta information in the first model list.

In some embodiments, the first synchronization submodule includes: a first deletion unit configured to, in response to the model deletion request, update the target model meta information and the target The model file is deleted, so that the object storage service broadcasts the synchronization cleanup instruction; the second receiving unit is configured to receive the object storage service broadcast after deleting the target model file and the target model meta information The synchronization cleaning instruction; a second deletion unit configured to delete the target model file and the target model meta information in the second model list in response to the synchronization cleaning instruction.

In some embodiments, the first deletion unit includes: a first deletion subunit configured to delete the target model meta information in the object storage service in response to the model deletion request; the second deletion The subunit is configured to delete the target model file in the object storage service according to a preset time interval.

In some embodiments, the model management service further includes: a second determining module configured to, in response to receiving a retrieval request for model meta information sent by the client, determine the multi-level storage address carried in the retrieval request index parameters; the first retrieval module is configured to retrieve the model meta information in the object storage service based on the index parameters of at least one level of storage addresses in the index parameters of the multi-level storage addresses, and obtain the matching The model meta information to be fed back of the retrieval request; a feedback module configured to feed back the model meta information to be fed back to the client.

In some embodiments, when the model change event in the object storage service includes a model addition, the first response module 702 includes: a first receiving submodule configured to receive the The model download request sent under the condition of running the running software toolkit; wherein, the running software toolkit includes the interface of the model management service; the first calling unit is configured to respond to the model download request through the The interface sends the target model file and the target model meta information to the view parsing service.

An embodiment of the present application provides a client, the client includes: a processor, and a memory storing instructions executable by the processor; wherein the processor is configured to: send the load of the target model to the model management service request; the target model includes: target model file and target model meta information; receive the multi-level storage address in the object storage service of the target model meta information sent by the model management service based on the loading request; based on the A multi-level storage address, determining the target model meta-information and the target model file corresponding to the target model meta-information in the object storage service.

In some embodiments, the processor is further configured to locate the target model meta-information in the object storage service based on the storage address; determine that the target model file carried in the target model meta-information is in A storage address in the object storage service; based on the storage address of the target model file in the object storage service, locate the target model file in the object storage service.

In some embodiments, the processor is further configured to cache the target model file and the target model meta information; share the target model file and the target model meta information to the at least one view resolution Serve.

It should be noted that the description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding.

It should be noted that, in the embodiment of the present application, if the above-mentioned model management method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for Make an electronic device (which may be a terminal, a server, etc.) execute all or part of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, sports hard disk, read only memory (Read Only Memory, ROM), magnetic disk or optical disk, and various media that can store program codes. Thus, embodiments of the present application are not limited to any specific combination of hardware and software.

Correspondingly, the embodiment of the present application further provides a computer program product, the computer program product includes computer-executable instructions, and after the computer-executable instructions are executed, the steps in the model management method provided in the embodiments of the present application can be realized.

An embodiment of the present disclosure also provides a computer program, the computer program includes computer readable codes, and when the computer readable codes run in the electronic device, the processor of the electronic device executes the model management method described in any of the above embodiments .

Correspondingly, the embodiment of the present application further provides a computer storage medium, the computer storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the model management method provided by the above-mentioned embodiments is implemented. step.

Correspondingly, the embodiment of the present application provides a client. FIG. 8 is a schematic diagram of the composition and structure of the client in the embodiment of the present application. As shown in FIG. 8 , the client 800 includes: a processor 801, at least one communication bus, a communication Interface 802, at least one external communication interface and memory 803. Wherein, the communication interface 802 is configured to realize connection and communication between these components. Wherein, the communication interface 802 may include a display screen, and the external communication interface may include a standard wired interface and a wireless interface. Wherein the processor 801 is configured to execute the image processing program in the memory, so as to realize the steps of the model management method provided in the above embodiments.

The descriptions of the above embodiments of the model management system, electronic equipment, and storage media are similar to the descriptions of the above method embodiments, and have similar technical descriptions and beneficial effects as the corresponding method embodiments. records, so I will not repeat them here. For the technical details not disclosed in the embodiments of the model management system, electronic equipment and storage media of the present application, please refer to the description of the method embodiments of the present application for understanding.

The device involved in the embodiments of the present disclosure may be at least one of a system, a method, and a computer program product. A computer program product may include a computer readable storage medium having computer readable program instructions thereon for causing a processor to implement various aspects of the present disclosure.

A computer readable storage medium may be a tangible device that can retain and store instructions for use by an instruction execution device. A computer readable storage medium may be, for example, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. Examples of computer-readable storage media (a non-exhaustive list) include: portable computer disks, hard disks, Random Access Memory (RAM), Read-Only Memory (ROM), erasable Electrical Programmable Read Only Memory (EPROM) or flash memory, Static Random-Access Memory (Static Random-Access Memory, SRAM), Portable Compact Disc Read-Only Memory (CD-ROM), Digital Video Discs (DVDs), memory sticks, floppy disks, mechanically encoded devices such as punched cards or raised structures in grooves with instructions stored thereon, and any suitable combination of the foregoing. As used herein, computer-readable storage media are not to be construed as transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., pulses of light through fiber optic cables), or transmitted electrical signals.

The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device over at least one of a network, such as the Internet, a local area network, a wide area network, and a wireless network. . The network may include at least one of copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and edge servers. A network adapter card or a network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing/processing device .

Computer program instructions for performing the operations of the present disclosure may be assembly instructions, Industry Standard Architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or in one or more source or object code written in any combination of programming languages, including object-oriented programming languages—such as Smalltalk, C++, etc., and conventional procedural programming languages, such as the “C” language or similar programming languages. Computer readable program instructions 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 implement. In cases involving 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 it may be connected to an external computer (for example, using Internet Service Provider to connect via the Internet). In some embodiments, electronic circuits, such as programmable logic circuits, FPGAs, or programmable logic arrays (Programmable Logic Arrays, PLAs), can be customized by using state information of computer-readable program instructions, which can execute computer-readable Read program instructions, thereby implementing various aspects of the present disclosure.

It should be understood that reference throughout the specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the present application. Thus, appearances of "in one embodiment" or "in an embodiment" in various places throughout the specification are not necessarily referring to the same embodiment. Furthermore, the particular features, structures or characteristics may be combined in any suitable manner in one or more embodiments. It should be understood that, in various embodiments of the present application, the sequence numbers of the above-mentioned processes do not mean the order of execution, and the execution order of the processes should be determined by their functions and internal logic, and should not be used in the embodiments of the present application. The implementation process constitutes any limitation. The serial numbers of the above embodiments of the present application are for description only, and do not represent the advantages and disadvantages of the embodiments.

It should be noted that, in this document, the term "comprising", "comprising" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, It also includes other elements not expressly listed, or elements inherent in the process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising that element.

In the several embodiments provided in this application, it should be understood that the disclosed devices and methods may be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or May be integrated into another system, or some features may be ignored, or not implemented. In addition, the coupling, or direct coupling, or communication connection between the components shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be electrical, mechanical or other forms of.

The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed to multiple network units; Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be used as a single unit, or two or more units can be integrated into one unit; the above-mentioned integration The unit can be realized in the form of hardware or in the form of hardware plus software functional unit. Those of ordinary skill in the art can understand that all or part of the steps to realize the above method embodiments can be completed by hardware related to program instructions, and the aforementioned programs can be stored in computer-readable storage media. When the program is executed, the execution includes: The steps of the foregoing method embodiments; and the foregoing storage media include: removable storage devices, read-only memory (Read Only Memory, ROM), magnetic disks or optical disks and other media that can store program codes.

Alternatively, if the above-mentioned integrated units of the present application are realized in the form of software function modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for Make an electronic device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the methods described in the various embodiments of the present application. The aforementioned storage medium includes various media capable of storing program codes such as removable storage devices, ROMs, magnetic disks or optical disks. The above is only a specific implementation of the application, but the scope of protection of the application is not limited thereto. Anyone familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the application. Should be covered within the protection scope of this application. Therefore, the protection scope of the present application should be determined by the protection scope of the claims.

Industrial Applicability

An embodiment of the present application provides a model management method, a model management service, a client, and a storage medium, wherein the method includes: synchronizing the target from the object storage service when the model management service meets the preset synchronization conditions The target model file and target model meta information of the model; in response to the loading request of the view parsing service, sending the target model file and the target model meta information to the view parsing service.

Claims (20)

1. A model management method applied to a model management service, the method comprising:

   Synchronizing the target model file and target model meta information of the target model from the object storage service when the model management service satisfies a preset synchronization condition;

   In response to the loading request of the view resolution service, the target model file and the target model meta information are sent to the view resolution service.
2. The method according to claim 1, wherein when the model management service satisfies preset synchronization conditions, before synchronizing the target model file and target model meta information of the target model from the object storage service, the method Also includes:

   receiving the upload request of the model sent by the client; the model includes: model file and model meta information;

   In response to the upload request, upload the model file and the model meta information to the object storage service.
3. The method according to claim 2, wherein, in response to the upload request, uploading the model file and the model meta information to the object storage service comprises:

   In response to the upload request, upload the model file and the model meta information to the object storage service according to different storage addresses.
4. The method according to claim 3, wherein in response to the upload request, uploading the model file and the model meta information to the object storage service according to different storage addresses comprises:

   Based on the model meta information, determine whether the object storage service includes the model file;

   In response to the object storage service not including the model file, uploading the model file to the object storage service according to a preset storage address;

   updating the model meta information based on the preset storage address;

   Upload the updated model meta information to the object storage service according to the multi-level storage address.
5. The method according to claim 4, wherein the determining whether the object storage service includes the model file based on the model meta information comprises:

   Determine the reference path of the model meta-information pointing to the preset model file;

   Verifying the model file based on the reference path to obtain a verification result;

   Based on the verification result, it is determined whether the object storage service includes the model file.
6. The method according to any one of claims 1 to 5, wherein the model management service satisfies preset synchronization conditions, including at least one of the following:

   The model management service receives an event notification message sent by the object storage service; wherein, the event notification message represents a model change event in the object storage service;

   The time interval between the historical moment and the current moment when the model management service performs data synchronization reaches a first preset duration;

   The model management service does not initiate data synchronization within a second preset duration, and receives a model synchronization instruction sent by the client; wherein, the second preset duration is longer than the first preset duration.
7. The method according to any one of claims 1 to 6, wherein, in the case that the model management service satisfies preset synchronization conditions, synchronizing the target model file and target model meta information of the target model from the object storage service ,include:

   determining a model change event in the object storage service under the preset synchronization condition;

   Obtain a first model list matching the model change event in the object storage service;

   In the first model list, determine the target model file and the target model meta information;

   Synchronize the second model list in the model management service based on the target model file and the target model meta information; wherein, the model list includes a plurality of model files, and at least one set of model meta information corresponding to each model file .
8. The method according to claim 7, wherein, in the case that the model change event includes a model addition, in the first model list, determining the target model file and the target model meta information includes :

   In the first model list, determining the target model file and the target model meta information different from the second model list;

   The synchronizing the second model list in the model management service based on the target model file and the target model meta information includes:

   downloading the object model file and the object model meta information from the object storage service;

   Adding the target model file and the target model meta information to the second model list.
9. The method according to claim 7, wherein, in the case that the model change event includes model deletion, determining the target model file and the target model meta information in the first model list includes :

   receiving a model deletion request sent by the client; wherein, the model deletion request carries a third storage address;

   determining the target model meta information from the first model list based on the third storage address;

   Determine the target model file pointed to by the target model meta information in the first model list.
10. The method according to claim 9, wherein the synchronizing the second model list in the model management service based on the target model file and the target model meta-information comprises:

    In response to the model deletion request, delete the target model meta information and the target model file in the object storage service, so that the object storage service broadcasts the synchronization cleaning instruction;

    receiving the synchronization cleanup instruction broadcast by the object storage service after deleting the target model file and the target model meta-information;

    In response to the synchronization cleanup instruction, the target model file and the target model meta information are deleted in the second model list.
11. The method according to claim 10, wherein, in response to the model deletion request, deleting the meta information of the model to be deleted and the model file to be deleted in the object storage service comprises:

    deleting the target model meta information in the object storage service in response to the model deletion request;

    The target model file is deleted in the object storage service according to a preset time interval.
12. The method according to any one of claims 2 to 11, wherein the method further comprises:

    In response to receiving a retrieval request for model meta information sent by the client, determine an index parameter of a multi-level storage address carried in the retrieval request;

    Retrieving the model meta information in the object storage service based on the index parameters of at least one level of storage addresses among the index parameters of the multi-level storage addresses, and obtaining the model meta information to be fed back that matches the retrieval request;

    Feedback the meta information of the model to be fed back to the client.
13. The method according to any one of claims 1 to 12, wherein, in the case that the model change event in the object storage service includes a model addition, in response to the load request of the view resolution service, the target model The file and the target model meta information are sent to the view parsing service, including:

    receiving a model download request sent by the view parsing service when running a software toolkit; wherein, the running software toolkit includes an interface of the model management service;

    In response to the model download request, sending the target model file and the target model meta information to the view parsing service through the interface.
14. A model management method, wherein, applied to a client, the method includes:

    Sending a loading request of the target model to the model management service; wherein, the target model includes: target model file and target model meta information;

    receiving the multi-level storage address in the object storage service of the target model meta information sent by the model management service based on the loading request;

    Based on the multi-level storage address, determine the target model meta-information and the target model file corresponding to the target model meta-information in the object storage service.
15. The method according to claim 14, wherein, based on the multi-level storage address, determining the target model meta-information and the target model file corresponding to the target model meta-information in the object storage service, include:

    locating the target model meta information in the object storage service based on the multi-level storage address;

    determining the storage address of the target model file carried in the target model meta-information in the object storage service;

    The target model file is located in the object storage service based on the storage address of the target model file in the object storage service.
16. The method according to claim 14 or 15, wherein the client includes at least one view resolution service, and based on the multi-level storage address, the target model meta-information and the After the target model file corresponding to the target model meta information, the method further includes:

    caching the target model file and the target model meta information;

    Sharing the target model file and the target model meta information to the at least one view resolution service.
17. A model management service, wherein the model management service includes:

    The first synchronization module is configured to synchronize the target model file and the target model meta information of the target model from the object storage service when the model management service satisfies a preset synchronization condition;

    The first response module is configured to send the target model file and the target model meta information to the view resolution service in response to the load request of the view resolution service.
18. A client, wherein the client includes: a processor and a memory storing instructions executable by the processor; wherein the processor is configured to:

    Send a loading request of the target model to the model management service; the target model includes: target model file and target model meta information;

    receiving the multi-level storage address in the object storage service of the target model meta information sent by the model management service based on the loading request;

    Based on the multi-level storage address, determine the target model meta-information and the target model file corresponding to the target model meta-information in the object storage service.
19. A computer-readable storage medium, on which a program is stored and applied to a computer device, wherein, when the program is executed by a processor, the method according to any one of claims 1 to 13 is implemented, or, the When the program is executed by the processor, the method according to any one of claims 14 to 16 is realized.
20. A computer program, said computer program comprising computer readable code, in the case of said computer readable code running in an electronic device, a processor of said electronic device executes to implement any one of claims 1 to 13 The method described in claim 1, or, when the program is executed by a processor, realize the method described in any one of claims 14 to 16.

Images