CN116382798A

CN116382798A - Method, system and equipment for establishing model service

Info

Publication number: CN116382798A
Application number: CN202310354042.4A
Authority: CN
Inventors: 陆则权
Original assignee: Ping An Bank Co Ltd
Current assignee: Ping An Bank Co Ltd
Priority date: 2023-03-21
Filing date: 2023-03-21
Publication date: 2023-07-04

Abstract

The invention discloses a method, a system and equipment for establishing model service, wherein model identification information, storage address information and service resource information are determined by responding to a model service loading instruction; determining a target container for loading the model according to the service resource information; determining target model data from a preset model warehouse based on storage address information and model identification information; at least one model data and identification information of the model data are stored in the model warehouse; storing address information of an address information characterization model warehouse; and controlling the target container to load the target model data to obtain a target model and a service interface corresponding to the target model, wherein the target model provides model service through the service interface. The invention decouples the model management and the model service, so that the change of the model data can not influence the model service, the dependence is lower, and the flexible management of the model service and the model management is realized.

Description

Method, system and equipment for establishing model service

Technical Field

The invention relates to the technical field of computer processing, in particular to a method, a system and equipment for establishing model service.

Background

Model services refer to the process of prediction using a trained neural network, deep learning model, or expert system, making decisions based on facts and rules contained in the expert system knowledge base, or algorithmic models derived from the deep learning system. The current AI architecture can execute a large amount of computation in the neural network in parallel by means of the AI chip/display card to improve the reasoning efficiency. Model management refers to management of model data, including modification of model data.

However, at present, the processes of model management and model service are coupled together, and centralized management on the model is lacking, in this case, the data change of the model directly affects the model service, and each time the data change of the model means forming a new model service, but there is always a need to keep the original model service when the model data change, and when the model data is damaged or the data is lost, the model service cannot be performed and repair is difficult.

Disclosure of Invention

Based on the research, the invention provides a method, a system and equipment for establishing model service, which can realize decoupling of model management and model service, so that the model service is not affected by the change of model data, the dependency is lower, and flexible management of the model service and the model management is realized.

In a first aspect, an embodiment of the present invention provides a method for establishing a model service, where the method includes:

determining model identification information, storage address information and service resource information in response to a model service loading instruction sent by a model creation end;

determining a target container for loading the model according to the service resource information;

determining target model data from a preset model warehouse based on storage address information and model identification information; at least one model data and identification information of the model data are stored in the model warehouse; storing address information of an address information characterization model warehouse;

the target container is controlled to load the target model data to obtain a target model and a service interface corresponding to the target model;

and sending the interface address of the service interface to the model creation end so that the model creation end calls the target model to provide model service according to the interface address of the service interface.

In one possible implementation manner of the present invention, before the target container is controlled to load the target model data, the method includes:

Determining a model storage address of the target model data in a model warehouse;

writing the model storage address into a target container, and controlling the target container to read target model data from a model warehouse according to the model storage address; or alternatively, the process may be performed,

and writing the target model data into the target container.

In one possible implementation of the present invention, the service resource includes mirror structure information and running memory information; determining a target container for loading the model according to the service resource information, wherein the target container comprises:

matching the mirror image structure information with mirror image resources in a preset mirror image warehouse to obtain target mirror image resources matched with the mirror image structure information;

and creating and obtaining a target container according to the target mirror image resource and the running memory information.

In one possible implementation manner of the present invention, after loading target model data by a control target container to obtain a target model and a service interface corresponding to the target model, the method includes:

responding to a model service updating instruction, and determining model identification information, storage address information and an updated version identification of a model to be updated;

determining a first container for running the model to be updated according to the model identification information of the model to be updated;

Determining model data of each version of the model to be updated from a preset model warehouse based on the storage address information and the model identification information of the model to be updated; the model warehouse stores model data of different versions and identification information of the model data; each version of model data having a version identification

According to the updated version information identification, updated model data with version identification matched with the updated version identification information is obtained from the model data of each version of the model to be updated;

and controlling the first container to load the update model data of the model to be updated to obtain the update model.

responding to a model service rollback instruction, and acquiring model identification information, storage address information and rollback version identification of a model to be rolled back;

determining a second container for running the model to be rolled back according to the model identification information of the model to be rolled back;

determining model data of each version of the model to be rolled back from a preset model warehouse based on the storage address information and the model identification information of the model to be rolled back; the model warehouse stores model data of different versions and identification information of the model data; the model data of each version has a version identification;

And according to the rollback version identification, obtaining rollback model data with version identification matched with the rollback identification from model data of each version of the model to be rollback.

In a second aspect, an embodiment of the present invention provides a method for establishing a model service, where the method includes:

responding to a model creation instruction, and determining model identification information of a model to be created and storage address information of a model warehouse; at least one model data and identification information of the model data are stored in the model warehouse; storing address information of an address information characterization model warehouse;

determining service resource information required by a model to be created;

according to the model identification information, the storage address information of the model warehouse and the service resource information, a model service loading instruction is sent to a model service end, so that the model service end determines the model identification information, the storage address information and the service resource information according to the model service loading instruction, determines a target container for loading a model according to the service resource information, determines target model data from a preset model warehouse based on the storage address information and the model identification information, and controls the target container to load the target model data to obtain a target model and a service interface corresponding to the target model; the method comprises the steps of sending an interface address of a service interface to a model creation end;

And calling the target model to perform model service based on the interface address of the service interface.

In one possible implementation manner of the present invention, after calling a target model to perform a model service based on a service interface, the method includes:

in response to the update instruction, determining update model data, model identification information and update version identification of the model to be updated, and storing the update model data and the update version identification of the model to be updated in a model warehouse;

and sending a model service updating instruction to the model server according to the updating version identification of the model to be updated, the model identification information and the storage address information of the model warehouse.

In one possible implementation manner of the present invention, after providing an interface according to a target model service and making a service call to the target model service, the method includes:

responding to the rollback instruction, determining rollback version identification of the model to be rolled back, model identification information and storage address information of a model warehouse;

and sending a model service rollback instruction to the model server according to the rollback version identification, the model identification information and the storage address information of the model to be rollback.

In a third aspect, an embodiment of the present invention further provides a system for building a model service, where the system includes a model service end, a model creation end, and a model repository:

The model creation end is used for responding to the model creation instruction, determining model identification information of a model to be created, storage address information of a model warehouse and required service resource information, and sending a model service loading instruction to the model service end according to the model identification information, the storage address information of the model warehouse and the service resource information; at least one model data and identification information of the model data are stored in the model warehouse; storing address information of an address information characterization model warehouse;

the model service end is used for responding to the model service loading instruction and determining model identification information, storage address information and service resource information; determining a target container for loading a model according to service resource information, determining target model data from a preset model warehouse based on storage address information and model identification information, controlling the target container to load the target model data to obtain a target model and a service interface corresponding to the target model, and sending an interface address of the service interface to a model creation end;

the model creation end is used for calling the target model to perform model service based on the interface address of the service interface.

In a fourth aspect, embodiments of the present invention also provide a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of any of the methods described above when the computer program is executed by the processor.

According to the method, the system and the equipment for establishing the model service, provided by the embodiment of the invention, the model identification information, the storage address information and the service resource information are determined by responding to the model service loading instruction, the target container for loading the model is determined according to the service resource information, the target model data is determined from the preset model warehouse based on the storage address information and the model identification information, then the target container is controlled to load the target model data, the target model and the service interface corresponding to the target model are obtained, and after the interface address of the service interface is sent to the model creation end, the model creation end can call the target model according to the interface address of the service interface to provide the model service. Therefore, the model warehouse stores the model, and the model data of the related model is called from the model warehouse to create the model service according to the service requirement, so that decoupling of model management and model service is realized, the model warehouse is used for managing the model and loading the service by a container, the model and the service can be flexibly managed, and the service stability is ensured.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic view of an embodiment of a system for creating a model service according to an embodiment of the present invention;

FIG. 2 is a flow diagram of one embodiment of a method of modeling a service provided in the present invention;

FIG. 3 is a flow diagram of one embodiment of a method of modeling a service provided in the present invention;

fig. 4 is a schematic structural diagram of a computer device according to an embodiment of the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.

In the description of the present invention, it should be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more features. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

As in the background art, the processes of model management and model service are coupled together, and centralized management of models is lacking, in which case, data changes of models directly affect model service, each time data changes of models mean that new model service is formed, but there is always a need to maintain original model service when model data changes in business, and when model data is damaged or data is lost, model service cannot be performed and repair is difficult.

Based on the above, the method, the system and the equipment for establishing the model service provided by the embodiment of the invention determine the model identification information, the storage address information and the service resource information by responding to the model service loading instruction, determine the target container for loading the model according to the service resource information, determine the target model data from the preset model warehouse based on the storage address information and the model identification information, then control the target container to load the target model data to obtain the target model and the service interface corresponding to the target model, and after the interface address of the service interface is sent to the model creation end, the model creation end can call the target model to provide the model service according to the interface address of the service interface. Therefore, the model warehouse stores the model, and the model data of the related model is called from the model warehouse to create the model service according to the service requirement, so that decoupling of model management and model service is realized, the model warehouse is used for managing the model and loading the service by a container, the model and the service can be flexibly managed, and the service stability is ensured.

Some terms involved in the embodiments of the present invention will be explained first.

A container: a container is an object used to store and organize other objects, typically located within an application server, for which loading and maintenance is responsible. One container can only exist within one application server, and one application server can build and maintain multiple containers.

Model services refer to the process of prediction using a trained neural network, deep learning model, or expert system, making decisions based on facts and rules contained in the expert system knowledge base, or algorithmic models derived from the deep learning system.

Model management refers to management of model data, including modification of model data.

The method, the system and the equipment for establishing the model service provided by the embodiment of the invention are explained below.

Referring to fig. 1, in the system for establishing a model service provided in the embodiment of the present invention shown in fig. 1, the system for establishing a model service provided in the embodiment of the present invention includes a model creation end 200, a model service end 100 and a model repository 300 that are in communication connection.

Network communication may be implemented by any communication means among the model service end 100, the model creation end 200 and the model repository 300, including, but not limited to, mobile communication based on the third generation partnership project (3rd Generation Partnership Project,3GPP), long term evolution (Long Term Evolution, LTE), worldwide interoperability for microwave access (Worldwide Interoperability for Microwave Access, wiMAX), or computer network communication based on the TCP/IP protocol family (TCP/IP Protocol Suite, TCP/IP), user datagram protocol (User Datagram Protocol, UDP), and the like. The model creation terminal 200 may upload the instruction to the model service terminal 100 and create information such as model service resource information through the above communication method.

The model creation end 200 refers to an end facing to a user side, and a user can use a model service through the model creation end.

The model service end 100 refers to an end providing the model service, and when a user needs to use the model service, an instruction can be sent to the model service end 100 through the model creation end, and the model service end 100 loads the model service according to the instruction.

The model warehouse is a cloud database which is established based on the cloud and is used for storing model data uploaded by users, and the model warehouse can form a catalog according to the addresses of the uploaded model data, so that the model data can be conveniently searched and managed.

The model server 100 may be a terminal device such as a notebook computer, an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, basic cloud computing services such as big data and an artificial intelligent platform, but is not limited thereto.

In the embodiment of the present invention, the model server 100 may be established based on Kubernetes, where the Kubernetes system implements control of containers through a Master component, and each container is stored in a Node component in the Kubernetes system. The Kubernetes system controls nodes storing target containers through a Master component running node controller, and loads target model data in the containers through the node controller controlling target containers under the nodes, so that the containers can provide services. The nodes can be controlled and called through the Master component, interfaces are also arranged among the nodes to transmit the information of the coming and going directions, that is, the loaded container can provide service in the Kubernetes system. Meanwhile, the Kubernetes system can provide an http interface to the outside, so that the service and call provided by the container to the outside can be realized.

In the present invention, the model creation terminal 200 may be a general-purpose computer device or a special-purpose computer device. In a specific implementation, the model creation terminal 200 may be a palm computer, a personal digital assistant (Personal Digital Assistant, PDA), a tablet computer, a wireless terminal device, etc., and the present invention is not limited to the type of the model creation terminal 200 described above.

It will be understood by those skilled in the art that, as shown in fig. 1, only one application scenario adapted to the solution of the present invention is shown, and the application scenario of the solution of the present invention is not limited thereto, and other application scenarios may further include more or fewer model creation terminals 200 than those shown in fig. 1, for example, only 1 model creation terminal 200 is shown in fig. 1, and it will be understood that the system for creating a model service may further include other user terminals communicatively connected to the model service terminal 100, which is not limited thereto.

The system for establishing the model service provided by the embodiment of the invention can realize the following functions:

the model creation terminal 200 is configured to respond to the model creation instruction, determine model identification information of a model to be created, storage address information of a model repository, and required service resource information, and send a model service loading instruction to the model service terminal 100 according to the model identification information, the storage address information of the model repository, and the service resource information; at least one model data and identification information of the model data are stored in the model warehouse; storing address information of an address information characterization model warehouse;

The model server 100 is configured to determine model identification information, storage address information, and service resource information in response to a model service loading instruction, determine a target container for loading a model according to the service resource information, determine target model data from a preset model warehouse based on the storage address information and the model identification information, control the target container to load the target model data, obtain a target model and a service interface corresponding to the target model, and send an interface address of the service interface to the model creation end 200;

the model creation end 200 is configured to invoke a target model to perform model services based on an interface address of the service interface. The model creation end 200 may provide an interactive interface for the user to operate, where the interactive interface displays a model upload button, a model service creation button, and the like. In this embodiment, the model upload button is used to upload a model, and the model service creation button is used to create a model service. When a user has a model uploading requirement, according to an interactive interface provided by the model creation end 200 for the user, the user clicks a model uploading button, and an uploading frame is displayed on the interactive interface and used for the user to drag the uploading model to the uploading frame, so that the model creation end can determine the uploading model. When the user needs to provide the model service through the model, the button model can be built by pressing the model service, the model creation end 200 responds to the operation of the user to display the model identification information of each model, the user can select according to the displayed model identification information, after the user selects the model identification information needing to provide the model service, the model creation end 200 can determine the model to be created according to the model identification information selected by the user, and then the service resource information required by the model to be created is obtained according to the memory size, architecture and other attribute information of the model to be created.

Since the model data of the model to be created is stored in the model repository, the model server 100 needs to obtain the model data of the model to be created from the model repository to provide the model service, and thus the model creation end 200 also needs to send the storage address information of the model repository to the model server 100. Based on this, after obtaining the service resource information and the model identification information of the model to be created, the model creation terminal 200 sends a model service loading instruction to the model service terminal 100 according to the model identification information, the storage address information of the model repository, and the service resource information.

It will be appreciated that the model service load instructions include model identification information, storage address information for the model repository, and service resource information.

After receiving the model service loading instruction, the Master component of the model service end 100 responds to the model service loading instruction, analyzes the model service loading instruction to obtain model identification information, storage address information and service resource information of a model to be created, then determines a target container for loading the model according to the service resource information, determines target model data from a model warehouse based on the storage address information and the model identification information, then controls the target container to load the target model data to obtain a target model and a service interface corresponding to the target model, and the model service end 100 sends a service interface address to the model creation end 200. It will be appreciated that the target model is the model to be created.

After receiving the address of the service interface, the model creation end 200 may call the target model to provide the model service according to the address of the service interface.

The user service presses a call service button on the interactive interface, the creation end 200 calls a target container corresponding to the target model at the model service end 100 according to the service requirement, and controls the target container to call according to the interface provided by the target container.

It can be understood that, after receiving the address of the service interface, the model creation end 200 can call the target model according to the address of the service interface to provide the model service, so as to implement decoupling of model management and model service, and load the service by using the model warehouse to manage the model and using the container, so that the model and the service can be flexibly managed, and the stability of service is ensured.

In this embodiment, each model has unique model identification information, where the model identification information may be represented by a model code, a model number, a model name, or the like.

In this embodiment, the model data refers to the running logic code of the model, model architecture parameters, and the like. After determining the model identification information and the storage address information, the model server 100 can determine the position in the model warehouse according to the storage address information, then match the determined model identification information with the model identification information of model data in the model warehouse, and determine that the model data is target model data if the model identification information of the model data is consistent with the model identification information of the model to be created.

The container is characterized by a lightweight, portable and self-contained software packaging technology, one container is a unit, the framework structure of the container is divided into one, the running environment is stable, and the application and all necessary files of the application are bound into one container running environment, so that a model in the container can be easily moved and run on any operating system in any environment.

After the target container is created and obtained and the target model data is obtained, the target container can be controlled to load the target model data, and before the target container is controlled to load the target model data corresponding to the target container, the target model data needs to be put into the target container. Alternatively, the process of the model server 100 putting the target model data into the target container may be implemented by:

determining a model storage address of the target model data in the model warehouse, writing the model storage address into a target container, and controlling the target container to read the target model data from the model warehouse according to the model storage address; alternatively, the object model data is written to the object container.

The model server 100 closes the container after the target container loads the model, and at this time, the data of the container is lost, so that the data generated by the operation of the target container needs to be persisted, so that the target container can store the related data of the model, such as the model storage address and the data generated by the operation.

A data volume is a special directory that can provide for use of one or more containers, and model storage addresses can be written as data volumes into target containers, based on which target containers can store model-related data, such as model storage addresses and run-generated data, to achieve data persistence for the target container run-generated data.

The data volume (model storage address) is written into the target container, and the target container under the Master main component control node of the model service end 100 reads the target model data from the model warehouse according to the data volume (model storage address), or the target container under the Master main component control node copies and writes the target model data into the target container according to the model storage address.

Or, the model server 100 obtains the target model data according to the model storage address of the target model data in the model warehouse, and the Master main component of the model server 100 inputs the target model data into the target container and controls the target container to load the target model data.

In the embodiment of the invention, the service resource information comprises mirror structure information, running memory information and the like, wherein the mirror structure information refers to data for starting the container to run, and the running memory information refers to data volume in a model.

In the embodiment of the present invention, the model server 100 determines a target container for loading a model according to service resource information, and may be implemented by the following procedures:

and matching the mirror image structure information with the mirror image resources in the preset mirror image warehouse to obtain target mirror image resources matched with the mirror image structure information.

The method comprises the steps that a container is started by mirroring, a running file is executed to generate a container, the mirroring is used for starting the running file of the container, mirror structure information, running memory information and the like are included in service resource information corresponding to a model, a model server 100 needs mirror resources to start the container in the process of creating the container, the running memory information is used for evaluating the magnitude of running memory in the container so as to create the container, tens of thousands of mirrors are stored in a preset storage unit in the model server 100 in advance, the storage unit is named as a mirror warehouse, a Master main component of the model server 100 can perform mirror resource matching from the mirror warehouse according to the mirror structure information of a target model, mirror resources identical to the mirror structure information of the target model are obtained through matching, the Master main component of the model server 100 runs the target mirror resources, and a container matched with the running memory information is created, and the target container can be obtained.

Considering that there may be multiple versions of each model, for each model, different versions of the model have unique version identifications, and the model identification information of the different versions of the model are consistent. Based on this, model data for different versions of each model and version identification for each version may also be stored in the model repository.

In this embodiment, based on the possible need for updating the model in the service, the system for establishing the model service provided in this embodiment may further include the following functions:

the model creation end 200 is configured to determine update model data, model identification information, and update version identification of the model to be updated in response to the update instruction, and store the update model data and the update version identification of the model to be updated in the model repository.

And sending a model service updating instruction to the model service end 100 according to the updating version identification of the model to be updated, the model identification information and the storage address information of the model warehouse.

The model service end 100 is configured to determine model identification information, storage address information, and update version identification of a model to be updated in response to a model service update instruction.

And determining a first container for running the model to be updated according to the model identification information of the model to be updated.

Determining model data of each version of the model to be updated from a preset model warehouse based on the storage address information and the model identification information of the model to be updated; the model warehouse stores model data of different versions and identification information of the model data; each version of model data has a version identification.

And obtaining updated model data with the version identification matched with the updated version identification from the model data of each version of the model to be updated according to the updated version identification.

After updating the model based on the service requirement, the user needs to upload the model to the model warehouse for storage, and can click a model service update button on an interface provided by the model creation end 200, send an update instruction to the model creation end 200, the model creation end 200 responds to the update instruction, display a model data uploading frame, the user can upload the update model data and model identification information of the update model data through the model data uploading frame, the model creation end 200 can obtain the update model data and the model identification information of the model to be updated, after obtaining the update model data and the model identification information, the model creation end 200 can store the update model data of the model to be updated in the model warehouse according to the IP address of the model warehouse, inquire the version identification of the model data with the same model identification information in the model warehouse according to the model identification information of the model to be updated, analyze the version identification of the model data with the same inquired model identification information, and obtain the version identification of the model data obtained by last update of the model to be updated. And obtaining the version identifier obtained by the last update, namely obtaining the updated version identifier of the updated model data of the model to be updated based on the version identifier obtained by the last update.

In the embodiment of the invention, the version identifier may be represented by any number, letter or combination thereof, and it is understood that the version identifiers of different versions are different. In order to facilitate the determination of the version identifier, the embodiment of the invention adopts the digital representation version identifier, and when the updated version identifier of the updated model data of the model to be updated is obtained based on the version identifier obtained by the last update, 1 can be added on the basis of the version identifier obtained by the last update to obtain the updated version identifier of the updated model data of the model to be updated.

For example, the model creation end 200 may store the update model data of the model to be updated in the model repository according to the model repository IP address, query the version identifiers of the model data with the same model identifier information in the model repository according to the model identifier information of the model to be updated, obtain the version identifiers of the model data with the same model identifier information as "00001", "00002", "00003" and "00004", analyze the version identifiers of the model data with the same queried model identifier information, obtain the version identifier "00004" obtained by the last update of the model to be updated, and obtain the update version identifier "00005" of the update model data of the current model to be updated based on the version identifier obtained by the last update.

The model creation end 200 obtains the update model data, the model identification information and the update version identification of the model to be updated, stores the update model data and the update version identification of the model to be updated in the model repository, and sends a model service update instruction to the model server 100 according to the update version identification of the model to be updated, the model identification information and the storage address information of the model repository.

After receiving the model service update instruction, the model service end 100 analyzes the model service update instruction to obtain model identification information, storage address information and update version identification of the model to be updated. The container identification information (name, etc.) of the container of the model to be updated may be related to the model identification information of the model carried by the container, and the searching of the target container may be performed by the model identification information of the model carried by the container at the model server 100, so that according to the model identification information of the model to be updated, the searching of the container with the same container as the model identification information may be performed by the model identification information at the model server 100, and the container is set as the first container, thereby determining the first container running the model to be updated, determining the address of the model warehouse in the memory based on the storage address information, and according to the model identification information of the model to be updated, searching the model to be updated in the model warehouse. Based on the storage address information and the model identification information of the model to be updated, the model server 100 determines update model data of the model to be updated from a model warehouse, and controls the first container to load the update model data of the model to be updated through the Master main component to obtain the update model.

As can be appreciated, in response to the model service loading instruction sent by the model creation end 200, the model service end 100 determines the target model data from the preset model repository based on the storage address information and the model identification information, where there may be multiple versions of the model data with the same model identification information in the model repository, so that the user uploads the model based on the interactive interface of the model creation end 200, the model creation end 200 sets the version identification according to the uploaded model, and when sending the model service loading instruction, the model service end 100 carries the model identification information, the storage address information, the service resource information, and the version identification in the instruction, and can determine the target model data from the model repository based on the storage address information, the model identification information, and the version identification.

In this embodiment, based on the possible need for a rollback model in the service, the system for establishing a model service provided in the embodiment of the present invention may further include the following functions:

the model creation end 200 is configured to determine, in response to the rollback instruction, a rollback version identifier of a model to be rolled back, model identification information, and storage address information of a model repository.

And sending a model service rollback instruction to the model service end 100 according to the rollback version identification, the model identification information and the storage address information of the model to be rolled back.

The model service end 100 is configured to obtain, in response to a model service rollback instruction, model identification information, storage address information, and a rollback version identification of a model to be rolled back.

And determining a second container for running the model to be rolled back according to the model identification information of the model to be rolled back.

Determining model data of each version of the model to be rolled back from a preset model warehouse based on the storage address information and the model identification information of the model to be rolled back; the model warehouse stores model data of different versions and identification information of the model data; each version of model data has a version identification.

And controlling the second container to load the rollback model data so as to rollback the model to be rollback.

The model creation end 200, the user clicks a model service backspacing button on an interface provided by the model creation end 200, a backspacing instruction is sent to the model creation end 200, the model creation end 200 responds to the backspacing instruction, a backspacing version selection frame is displayed, the user selects a model to be backspacing according to model identification information based on the backspacing version selection frame, further, the user enters version selection of the model to be backspacing of the same model identification information, the model creation end 200 can obtain backspacing model data and model identification information of the model to be backspacing, and the model creation end 200 sends the model service backspacing instruction to the model service end 100 according to the version of the model to be backspacing, the identification of the model and storage address information.

The model service end 100 responds to the model service backspacing instruction, receives the model service backspacing instruction, analyzes the model service backspacing instruction to obtain model identification information, storage address information and version information carrying a model to be backspacing, searches a container with the same identification information as the model identification information of the model to be backspacing in the model service end 100 according to the model identification information of the model to be backspacing and the container name of the second container, and sets the container as the second container, so that the second container for running the model to be backspacing can be determined at the model service end 100 according to the model identification information of the model to be backspacing. The models of the identification information of the same model data of the model warehouse are distinguished by different versions, the model versions can be sequentially arranged from small to large according to the uploading time, the model server 100 queries the model warehouse to be capable of determining through a Master main component based on the storage address information and the identification information, the model data of each version of the model to be rolled back is determined, and the models of different versions with the same identification information of the model data of the model warehouse can be obtained according to the initial model data. The model server 100 obtains a model corresponding to version information of the model to be rolled back according to the model warehouse, determines that the version information of the model data of each version is compared with the version information of the model to be rolled back, determines that the version information of the model data is consistent with the version information of the model to be rolled back, obtains the model data to be rolled back, and controls the second container to load the model data to be rolled back through the Master main component, so that rolling back of the version of the model to be rolled back is achieved.

In this embodiment, based on the possible requirement that the service needs to be provided by the multi-model combination in the service, the system for establishing the model service provided in the embodiment of the present invention may further include the following functions:

the model creation terminal 200 is configured to confirm the model service orchestration script information, and send a model service orchestration request to the model service terminal 100 according to the model service orchestration script information.

The model service end 100 is configured to respond to the model service arrangement request, parse the model service arrangement request to obtain model service arrangement script information, analyze the model service arrangement script information to obtain containers corresponding to each model service used for model service arrangement, control each container to sequentially order each container according to the sequence in the model service arrangement script information by the Master component according to the model service arrangement script information, and put the containers into a pod component, provide an interface according to the pod component, send an interface address to the model creation end 200, and be used for invoking the multimode service.

The user may click a model service orchestration button on an interface provided by the model creation end 200, send an update instruction to the model creation end 200, and the model creation end 200 responds to the update instruction to display a model service orchestration script uploading frame, through which the user may upload the model service orchestration script.

Fig. 2 is a schematic flow chart of a method for establishing model services according to an embodiment of the present invention. As shown in fig. 2, it should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than what is shown or described herein.

The method for establishing the model service provided by the embodiment of the invention comprises the following steps of S101 to S104:

s101, determining model identification information, storage address information and service resource information in response to a model service loading instruction sent by a model creation end.

S102, determining a target container for loading the model according to the service resource information.

S103, determining target model data from a preset model warehouse based on storage address information and model identification information; at least one model data and identification information of the model data are stored in the model warehouse; address information characterizing the model repository is stored.

S104, controlling the target container to load the target model data to obtain a target model and a service interface corresponding to the target model.

S105, the interface address of the service interface is sent to the model creation end, so that the model creation end calls the target model to provide model service according to the target model through the interface address of the service interface.

Based on different framework structures of each model, the method is difficult to stably process under the scene of multi-model deployment application or single-model update transplantation, so that the data in the target model are bound into a container running environment, and the effects of flexible management of the model, stable service and the like can be realized.

In an optional implementation manner, the method based on the container management data may directly perform management call on the container model data or perform management call on the model data in the model warehouse by means of a data volume, and the step before the embodiment of the invention controls the target container to load the target model data may further include:

and writing the target model data into the target container.

In this embodiment, after the target container carries the target model data, the target container under the Master main component control node loads the target model data, and at this time, the target container may provide an interface to the outside, and the model server 100 obtains the interface address of the target container and sends the interface address to the target creation end 200, where the target creation end 200 may call the target container to provide a service through interface control. According to the address of the target container loading the target model data, the interface service provided by the corresponding container can be obtained, that is, the target model and the service interface corresponding to the target model can be obtained, the target model can provide service in the model service end 100 through the service interface, according to the interface address of the http interface service provided by the corresponding container, the interface address is fed back to the model creation end 200, and the model creation end 200 can remotely provide service to the outside through the interface control of the target container of the model service end.

In an alternative implementation manner, a container is considered to be started by mirroring, and a mirrored file is executed to generate a container; determining a target container for loading the model from the service resource information may further comprise:

Mirroring refers to a form of file storage, an executable program, by which one or more containers may be created.

In an optional implementation manner, based on the possible need of updating the model in the service, the step after the target container loads the target model data to obtain the target model and the service interface corresponding to the target model in the embodiment of the present invention may further include:

According to the updated version identification, updated model data with the version identification matched with the updated version identification is obtained from the model data of each version of the model to be updated;

In an optional implementation manner, based on the possible need of a rollback model in the service, the step after the target container loads the target model data to obtain the target model and the service interface corresponding to the target model in the embodiment of the present invention may further include:

responding to a model service rollback instruction, and acquiring model identification information, storage address information and rollback version identification information of a model to be rolled back;

According to the rollback version identification, rollback model data with version identification matched with the rollback identification is obtained from model data of each version of the model to be rolled back, and according to version information, rollback model data with version identification matched with the version information is obtained from model data of each version of the model to be rolled back;

Based on the same inventive concept, referring to fig. 3, fig. 3 is a flow chart of a method for establishing a model service according to an embodiment of the invention. As shown in fig. 3, it should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than what is shown or described herein.

The method for establishing the model service provided by the embodiment of the invention comprises the steps of S201 to S204:

s201, responding to a model creation instruction, and determining model identification information of a model to be created and storage address information of a model warehouse; at least one model data and identification information of the model data are stored in the model warehouse; address information characterizing the model repository is stored.

S202, determining service resource information required by the model to be created.

S203, sending a model service loading instruction to a model service end according to model identification information, storage address information and service resource information of a model warehouse, so that the model service end determines model identification information, storage address information and service resource information according to the model service loading instruction, determines a target container for loading a model according to the service resource information, determines target model data from a preset model warehouse based on the storage address information and the model identification information, and controls the target container to load the target model data to obtain a target model and a service interface corresponding to the target model; and sending the interface address of the service interface to the model creation end.

S204, calling a target model to perform model service based on the interface address of the service interface.

In an optional implementation manner, the step after the target model is called to perform the model service according to the service interface in the embodiment of the present invention may further include:

In an optional implementation manner, the step after the service call is performed on the target model service according to the interface provided by the embodiment of the present invention may further include:

In addition, the embodiment of the present invention further provides a computer device, which may be a terminal or a server, as shown in fig. 3, which shows a schematic structural diagram of the computer device according to the embodiment of the present invention, specifically:

The computer device may include one or more processor cores, a power supply 20 for one or more computer-readable storage media, a memory 30, and an input unit 50, among other components. Those skilled in the art will appreciate that the computer device structure shown in FIG. 3 is not limiting of the computer device and may include more or fewer components than shown, or may be combined with certain components, or a different arrangement of components. Wherein:

the processor 10 is a control center of the computer device and connects various parts of the entire computer device using various interfaces and lines, and performs various functions of the computer device and processes data by running or executing software programs and/or modules stored in the memory 30 and calling data stored in the memory 30, thereby performing overall monitoring of the computer device. Optionally, processor 10 may include one or more processing cores; preferably, the processor 10 may integrate an application processor that primarily handles operating systems, user interfaces, applications, etc., with a modem processor that primarily handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 10.

The memory 30 may be used to store software programs and modules, and the processor 10 executes various functional applications and data processing by executing the software programs and modules stored in the memory 30. The memory 30 may mainly include a storage program area that may store an operating system, application programs required for at least one function, and the like, and a storage data area; the storage data area may store data created according to the use of the computer device, etc. In addition, memory 30 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 30 may also include a memory controller to provide access to the memory 30 by the processor 10.

The computer device further includes a power supply 20 for powering the various components, preferably, the power supply 20 is logically connected to the processor 10 via a power management system whereby charge, discharge, and power consumption management functions are performed by the power management system. The power supply 20 may also include one or more of any of a direct current or alternating current power supply, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and the like.

The computer device may further comprise an input unit 50, which input unit 50 may be used for receiving input numerical or character information and for generating keyboard, mouse, joystick, optical or trackball signal inputs in connection with user settings and function control.

Although not shown, the computer device may further include a display unit or the like, which is not described herein. In particular, in this embodiment, the processor 10 in the computer device loads executable files corresponding to the processes of one or more application programs into the memory 30 according to the following instructions, and the processor 10 executes the application programs stored in the memory 30, so as to implement various functions as follows:

determining model identification information, storage address information and service resource information in response to the model service loading instruction;

And controlling the target container to load the target model data to obtain a target model and a service interface corresponding to the target model, wherein the target model provides model service through the service interface.

In summary, the method, the system and the equipment for establishing the model service provided by the embodiment of the invention determine the model identification information, the storage address information and the service resource information by responding to the model service loading instruction, determine the target container for loading the model according to the service resource information, determine the target model data from the preset model warehouse based on the storage address information and the model identification information, then control the target container to load the target model data to obtain the target model and the service interface corresponding to the target model, and after the interface address of the service interface is sent to the model creation end, the model creation end can call the target model to provide the model service according to the interface address of the service interface. Therefore, the model warehouse stores the model, and the model data of the related model is called from the model warehouse to create the model service according to the service requirement, so that decoupling of model management and model service is realized, the model warehouse is used for managing the model and loading the service by a container, the model and the service can be flexibly managed, and the service stability is ensured.

The above describes in detail a method, a system and a device for establishing a model service provided by the embodiment of the present invention, and specific examples are applied to describe the principles and implementations of the present invention, where the description of the above examples is only used to help understand the method and core idea of the present invention; meanwhile, as those skilled in the art will vary in the specific embodiments and application scope according to the ideas of the present invention, the present description should not be construed as limiting the present invention in summary.

Claims

1. A method for establishing a model service, comprising:

determining a target container for loading a model according to the service resource information;

determining target model data from a preset model warehouse based on the storage address information and the model identification information; at least one model data and identification information of the model data are stored in the model warehouse; the storage address information characterizes the address information of the model warehouse;

Controlling the target container to load the target model data to obtain a target model and a service interface corresponding to the target model;

2. The method of claim 1, wherein prior to said controlling said target container to load said target model data, said method further comprises:

determining a model storage address of the target model data in the model warehouse, writing the model storage address into the target container, and controlling the target container to read the target model data from the model warehouse according to the model storage address; or alternatively, the process may be performed,

and writing the target model data into the target container.

3. The method of claim 1, wherein the service resource includes mirror structure information and running memory information; the determining a target container for loading a model according to the service resource information comprises the following steps:

matching the mirror image structure information with a mirror image resource in a preset mirror image warehouse to obtain a target mirror image resource matched with the mirror image structure information;

And creating a target container according to the target mirror image resource and the running memory information.

4. The method according to claim 1, wherein after the target container is controlled to load the target model data to obtain a target model and a service interface corresponding to the target model, the method further comprises:

determining model data of each version of the model to be updated from a preset model warehouse based on the storage address information and the model identification information of the model to be updated; the model warehouse stores model data of different versions and identification information of the model data; the model data of each version has a version identification;

according to the updated version identification, updated model data with version identification matched with the updated version identification is obtained from the model data of each version of the model to be updated;

and controlling the first container to load the update model data of the model to be updated to obtain an update model.

5. The method according to claim 1, wherein after the target container is controlled to load the target model data to obtain a target model and a service interface corresponding to the target model, the method further comprises:

according to the rollback version identification, rollback model data with version identification matched with the rollback identification is obtained from the model data of each version of the model to be rollback;

6. A method for establishing a model service, comprising:

responding to a model creation instruction, and determining model identification information of a model to be created and storage address information of a model warehouse; at least one model data and identification information of the model data are stored in the model warehouse; the storage address information characterizes the address information of the model warehouse;

determining service resource information required by the model to be created;

according to the model identification information, the storage address information of the model warehouse and the service resource information, a model service loading instruction is sent to a model service end, so that the model service end determines model identification information, storage address information and service resource information according to the model service loading instruction, a target container for loading a model is determined according to the service resource information, target model data is determined from a preset model warehouse based on the storage address information and the model identification information, the target container is controlled to load the target model data, a target model and a service interface corresponding to the target model are obtained, and an interface address of the service interface is sent to the model creation end;

7. The method of claim 6, wherein after invoking the object model for model services based on the service interface, the method further comprises:

in response to an update instruction, determining update model data, model identification information and update version identification of a model to be updated, and storing the update model data and the update version identification of the model to be updated in the model warehouse;

8. The method of claim 6, wherein after making a service call to the object model service in accordance with the object model service provisioning interface, the method further comprises:

and sending a model service rollback instruction to a model service end according to the rollback version identification, the model identification information and the storage address information of the model to be rollback.

9. The system for establishing the model service is characterized by comprising a model service end, a model creation end and a model warehouse;

the model creation end is used for responding to the model creation instruction, determining model identification information of a model to be created, storage address information of a model warehouse and required service resource information, and sending a model service loading instruction to the model service end according to the model identification information, the storage address information of the model warehouse and the service resource information; at least one model data and identification information of the model data are stored in the model warehouse; the storage address information characterizes the address information of the model warehouse;

the model service end is used for responding to a model service loading instruction, determining model identification information, storage address information and service resource information, determining a target container for loading a model according to the service resource information, determining target model data from a preset model warehouse based on the storage address information and the model identification information, controlling the target container to load the target model data to obtain a target model and a service interface corresponding to the target model, and sending an interface address of the service interface to the model creation end;

And the model creation end is used for calling the target model to perform model service based on the interface address of the service interface.

10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 5 or the steps of the method according to any one of claims 6 to 8 when the computer program is executed by the processor.