Model deployment methods, equipment, storage media, and software products

Abstract

This application provides a model deployment method, device, storage medium, and program product. In this application, a novel model deployment method is provided, applied to a model application platform. This platform provides workflow configuration capabilities and is coupled with a cloud resource platform that provides automatic model deployment capabilities. By integrating the model deployment process into the workflow creation process and combining it with the automatic model deployment capabilities provided by the cloud resource platform, integrated deployment of model deployment and application is achieved. Workflow creation is equivalent to the model application process, simultaneously integrating the model deployment process. This eliminates the need for users to manually switch between different platforms, simplifying model deployment operations, improving model deployment efficiency, and providing users with a seamless model application experience.

Description

Technical Field

[0001] This application relates to the field of cloud computing technology, and in particular to a model deployment method, device, storage medium, and program product. Background Technology

[0002] With the development of artificial intelligence technology, the application of neural network models (referred to as model application) is becoming increasingly widespread. Model application refers to the process of using deployed models to solve practical problems in a specific domain; and the prerequisite for model application is model deployment. Model deployment refers to the process of converting a trained model into an executable form and integrating it into a production environment.

[0003] In practical applications, users can deploy models in the cloud and allow local applications to initiate network calls to use the model's capabilities. This approach fully combines the powerful computing capabilities of cloud computing with the flexibility of local applications, enabling developers to build high-performance, scalable applications.

[0004] The above methods involve a series of operations such as deploying the model in the cloud, developing the model's calling interface, and developing the interface between the application and the calling interface. The model deployment operation is relatively cumbersome, resulting in low model application efficiency and failing to provide users with a smooth and seamless model application experience. Summary of the Invention

[0005] This application provides a model deployment method, device, storage medium, and program product to improve model application efficiency and provide users with a consistent and smooth model application experience.

[0006] This application provides a model deployment method, comprising: displaying a first page, the first page including a workflow associated with a target application, the workflow including at least a target workflow node for model invocation, the target workflow node being associated with a candidate neural network model; in response to a selection operation of choosing a target neural network model from the candidate neural network models, displaying a second page for deploying the target neural network model, the second page being associated with a cloud resource platform having automated model deployment capabilities; in response to a configuration operation on the second page, obtaining resource description information required for deploying the target neural network model, sending the resource description information to the cloud resource platform, the resource description information being used to instruct the cloud resource platform to deploy the target neural network model and return invocation information of the target neural network model; configuring the invocation information into the target workflow node, so as to trigger the target workflow node to invoke the target neural network model based on the invocation information during the operation of the target application.

[0007] This application also provides an electronic device, including: a memory and a processor; the memory for storing a computer program; and the processor, coupled to the memory, for executing the computer program to implement the steps in the model deployment method provided in this application.

[0008] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the steps in the model deployment method provided in this application.

[0009] This application also provides a computer program product, including a computer program / instructions, which, when executed by a processor, cause the processor to implement the steps in the model deployment method provided in this application.

[0010] In this embodiment, a novel model deployment method is provided, applied to a model application platform. This platform provides workflow configuration capabilities and is coupled with a cloud resource platform that provides automatic model deployment capabilities. By integrating the model deployment process into the workflow creation process and combining it with the automatic model deployment capabilities provided by the cloud resource platform, integrated deployment of model deployment and application is achieved. Workflow creation is equivalent to the model application process, while simultaneously integrating the model deployment process. Users do not need to manually switch between different platforms, simplifying model deployment operations, improving model deployment efficiency, and providing users with a seamless model application experience. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0012] Figure 1a A flowchart illustrating a model deployment method provided for an exemplary embodiment of this application;

[0013] Figure 1b A flowchart illustrating another model deployment method provided for an exemplary embodiment of this application;

[0014] Figure 1c A schematic diagram illustrating the arrangement of a model and application provided for an exemplary embodiment of this application;

[0015] Figure 1d A flowchart illustrating another model deployment method provided for an exemplary embodiment of this application;

[0016] Figure 2 A flowchart illustrating a model deployment method provided for an exemplary embodiment of this application;

[0017] Figure 3a A schematic diagram of a first page provided for an exemplary embodiment of this application;

[0018] Figure 3b A schematic diagram illustrating the display of a second page on a first page, provided as an exemplary embodiment of this application;

[0019] Figure 3c A schematic diagram of another first page provided for an exemplary embodiment of this application;

[0020] Figure 3d A schematic diagram of a third page provided for an exemplary embodiment of this application;

[0021] Figure 4 A schematic diagram illustrating the creation of a workflow on a model application platform, provided as an exemplary embodiment of this application;

[0022] Figure 5 A schematic diagram of the structure of a model deployment device provided for an exemplary embodiment of this application;

[0023] Figure 6 A schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0026] The various models involved in this application (including but not limited to language models or large models) comply with relevant laws and standards.

[0027] Currently, the traditional approach involves a series of operations, such as deploying the model in the cloud or offline, developing the model's API, developing the interface between the application and the API, and producing and using the model. Figure 1a As shown. Among them, Figure 1aThe model deployment process shown is rather cumbersome, resulting in low model application efficiency and failing to provide users with a smooth and consistent model application experience.

[0028] To address the low efficiency of model application in traditional methods, this application provides a solution involving an application integration platform. This platform can distribute neural network models to various applications, such as distributing them to social applications as intelligent customer service, or distributing them to group chats within social applications as Q&A bots. The process of distributing the neural network model is essentially a workflow generation process where the model application platform connects and orchestrates the application and the neural network model.

[0029] A workflow refers to a series of interconnected workflow nodes that are executed according to certain rules and order to achieve a specific goal. Workflow nodes are the basic building blocks of a workflow, representing specific tasks or decision points performed within the workflow. For example... Figure 1b The diagram illustrates one example of a workflow, but is not limited to it. A workflow can have one or more nodes, such as three, five, or ten. The following example illustrates a workflow with three nodes. The first workflow node is associated with application B1, the second with neural network model B2, and the third with application B3. Applications B1 and B3 can be the same application or different applications. During workflow operation, the first user can submit user input to application B1 of the application integration platform via a user interface or application programming interface (API) to trigger the workflow. Neural network model B2 generates a response based on the user input and provides it to application B3, which then provides it to the second user. The first and second users can be the same user or different users.

[0030] The prerequisite for generating a workflow is model deployment. In one application scenario, when a user wants to replace a neural network model with a privately deployed neural network model, it often requires manual deployment in the cloud or offline, or deployment through an automated deployment platform. Private deployment refers to deploying the neural network on infrastructure exclusively owned or controlled by the user, giving the user a high degree of control over the deployed neural network model.

[0031] like Figure 1cThe diagram illustrates an example of deploying a model on a model deployment platform. A workflow is created on the application integration platform, accessing the model deployment platform during the creation process. The neural network model is deployed on the model deployment platform. Once deployment is complete, the application integration platform is accessed again to establish a connection between the application and the deployed neural network model, thus completing the workflow configuration and enabling model application. However, this approach separates model deployment and application, requiring users to switch between different platforms for both processes. This is relatively cumbersome and fails to provide a seamless and consistent model application experience.

[0032] To address the cumbersome operational issues in the aforementioned model deployment process, this application provides a novel model deployment method applied to a model application platform. This platform offers workflow configuration capabilities and is coupled with a cloud resource platform that provides automatic model deployment capabilities. By integrating the model deployment process into the workflow creation process and combining it with the automatic model deployment capabilities provided by the cloud resource platform, integrated deployment of model deployment and application is achieved. Workflow creation is equivalent to the model application process, while simultaneously incorporating the model deployment process. This eliminates the need for users to manually switch between different platforms, simplifying model deployment operations, improving model deployment efficiency, and providing users with a seamless and smooth model application experience.

[0033] Among them, such as Figure 1d As shown, a flowchart illustrates a one-stop solution for model application and deployment on a model application platform. This solution supports the following operations in one stop on the model application platform: creating a workflow, deploying a neural network model during workflow creation, and configuring the workflow after the neural network model deployment is complete, thereby enabling model application. The following will combine... Figures 2-4 The illustrated embodiments provide a detailed description of the one-stop solution for model application and model deployment provided in this application.

[0034] Figure 2 This is a flowchart illustrating a model deployment method provided as an exemplary embodiment of this application. Figure 2 As shown, the method includes:

[0035] 201. Display the first page, which includes the workflow associated with the target application. The workflow includes at least the target workflow node for model invocation, and the target workflow node is associated with the candidate neural network model.

[0036] 202. In response to the selection operation of choosing the target neural network model from the candidate neural network models, a second page for deploying the target neural network model is displayed. The second page is associated with a cloud resource platform with automated model deployment capabilities.

[0037] 203. In response to the configuration operation on the second page, obtain the resource description information required to deploy the target neural network model, send the resource description information to the cloud resource platform, and use the resource description information to instruct the cloud resource platform to deploy the target neural network model and return the calling information of the target neural network model.

[0038] 204. Configure the call information into the target workflow node so that the target workflow node can call the target neural network model based on the call information during the execution of the target application.

[0039] In this embodiment, a model application platform is provided, which can provide workflow creation services for any application. For ease of distinction and description, the following description will focus on creating a workflow for a target application. The target application can be any type of application. For example, the target application can be an application deployed in the cloud or on a remote server, which users can access through a web browser without installing the application locally on their terminal. For example, the target application can include, but is not limited to: Customer Relationship Management (CRM) applications, office applications, Enterprise Resource Planning (ERP) applications, etc. Another example is that the target application is a mobile application designed for a terminal device. These applications are typically developed for a specific operating system and can be downloaded through the app store on the terminal device. For example, the target application can include, but is not limited to: social applications, video applications, game applications, and shopping applications, etc.

[0040] In addition, the target application can also be a special type of application. For example, the target application can be implemented as a sub-application that relies on the content publishing and service interaction of a certain type of application, such as a WeChat Official Account. Another example is that the target application can be implemented as a lightweight mobile application, such as a mini-program. Yet another example is that the target application can be embedded in other applications as a web page component.

[0041] In this embodiment, a neural network model can be used to provide services to the target application. The neural network model is not limited in this embodiment. From the perspective of model scale, the neural network model in this embodiment can be a model with a large parameter scale, such as a large language model, or a traditional neural network model with a relatively small parameter scale; there is no limitation in this regard. The services that the neural network model can provide include, but are not limited to, natural language processing services and computer vision services. Natural language processing services include, but are not limited to, text classification, translation, question answering systems, text generation, and speech recognition and synthesis. Computer vision services include, but are not limited to, image classification, object detection, image segmentation, and pose estimation.

[0042] In this embodiment, the model application platform can display a first page to the user, which includes a workflow associated with the target application. Optionally, the user can initiate a workflow creation operation for the target application on the first page. Specifically, the workflow creation operation can be the creation of a workflow node. The first page includes a workflow node adding control; in response to a user operation on this control, a workflow node is created. If there are multiple workflow nodes, multiple user operations on the adding control can be initiated to create multiple workflow nodes, thereby generating a workflow for the target application. Figure 3a As shown, an add control is displayed on the first page. When the add control is triggered, the model application platform can display one or more candidate neural network models on the first page, allowing the user to select the target neural network model and create the corresponding workflow node for the target neural network model.

[0043] The user operations for adding controls can include, but are not limited to, various interactive operations such as touch operations, gesture operations, voice operations, head movements, and eye movements. Touch operations include, but are not limited to, click operations, double click operations, long press operations, swipe operations, pinch operations, or mouse hover operations. Swipe operations include, but are not limited to, straight swipes and curved swipes.

[0044] In this embodiment, the model application platform can respond to the creation operation on the first page and create a workflow associated with the target application on the first page. The workflow includes at least a target workflow node for model invocation, and the number of target workflow nodes is not limited. For example, the number of target workflow nodes in the workflow can be one or more, such as two, three, or five, etc., without limitation. The target workflow node is associated with candidate neural network models, and the number of candidate neural network models is also not limited; it can be one or more, such as five, ten, or twenty, etc.

[0045] In this embodiment, the user can select a target neural network model from candidate neural network models. In response to this selection, the model application platform displays a second page for deploying the target neural network model. This second page is associated with a cloud resource platform that has automated model deployment capabilities.

[0046] In this embodiment, the cloud resource platform can provide users with a one-stop automated model deployment service. It can offer various cloud resources, such as computing resources, storage resources, and network resources, and has the capability to automatically deploy models on these cloud resources. It can be various types of resource platforms or products, including, but not limited to, resource platforms that help users efficiently deploy, manage, and operate cloud applications. The cloud resource platform can create cloud computing instances for users and deploy neural network models on these instances to enable the neural network models to run. The specific implementation of the cloud computing instance is not limited; for example, a cloud computing instance can include, but is not limited to, virtual machines or containers (Docker).

[0047] The implementation method for associating the second page with the cloud resource platform is not limited. For example, if the cloud resource platform provides a model deployment page, through which it creates cloud computing instances and deploys neural network models for users, the model deployment page can be directly embedded into the model application platform as the second page to establish a connection between the second page and the cloud resource platform. Alternatively, resource files for the model deployment page can be obtained. These resource files are necessary for rendering the model deployment page and may include, but are not limited to, HyperText Markup Language (HTML) information, Cascading Style Sheets (CSS) information, and JavaScript files. JavaScript is a widely used programming language for web development that enables web pages to be interactive. The model application platform renders these resource files to obtain and display the second page.

[0048] The display method of the second page is not limited; furthermore, it can be displayed using inline frames. An inline frame is an HTML element that allows a second page to be embedded within a first page. By using... <iframe>The label can display the second page in the first page without leaving the first page. For example, the second page can be displayed through inline frame technology, and the display methods of the second page can include but are not limited to: pop-up box, drop-down menu, sidebar, collapsible box, etc.

[0049] In this embodiment, the user can initiate a configuration operation on the second page, and this configuration operation is used for the user to provide the resource description information required for deploying the target neural network model. For example, an upload control is included on the second page, and the user uploads the resource description information required for deploying the target neural network model through this upload control. The configuration operation can be implemented as an upload operation for the upload control. Another example is that at least one configuration item corresponding to the resource description information is included on the second page, and the user performs an editing operation on at least one configuration item to initiate a configuration operation on the second page. Among them, in Figure 3b, taking the second page being implemented as a pop-up box on the first page and configuring the computing resources, memory resources, and bandwidth resources of the cloud computing instance on the second page as an example for illustration. Optionally, the second page can also provide information on paying for the computing instance, so that the cloud resource platform can deploy the cloud computing instance when the user completes the payment.

[0050] In this embodiment, when a configuration operation is initiated on the second page, the model application platform responds to the configuration operation on the second page, obtains the resource description information required for deploying the target neural network model, and sends the resource description information to the cloud resource platform. The resource description information is used to instruct the cloud resource platform to deploy the target neural network model and return the call information of the target neural network model.

[0051] Among them, the resource description information can include but is not limited to: description information on the type and quantity of resources. For example, the types of resources required for deploying a cloud computing instance can include but are not limited to: computing resources, memory resources, and network resources, etc.Computing resources include, but are not limited to: Central Processing Unit (CPU), Graphics Processing Unit (GPU), Data Processing Unit (DPU), Tensor Processing Unit (TPU), Cloud Infrastructure Processing Unit (CIPU), Application Specific Integrated Circuit (ASIC), and Field Programmable Gate Array (FPGA). Memory resources include, but are not limited to: Random Access Memory (RAM) and Cache. Network resources include, but are not limited to: Network Interface Card (NIC) or bandwidth. For example, resource description information may include, but is not limited to: 4 CPUs, 4GB of memory, and 1024M bandwidth.

[0052] The cloud resource platform can create a target cloud computing instance based on the resource description information, and run the target neural network model on the target cloud computing instance based on the source files of the target neural network model. The source files of the model refer to the files used to define and save the structure, configuration, and trained parameters (weights) of the target neural network model. These files typically contain the information required to create the target neural network model. The source files of the target neural network model can come from the cloud resource platform or a third-party resource platform.

[0053] Here, the model invocation information refers to the network address information required to interact with the target neural network model when it is deployed. For example, the invocation information can be an application programming interface (API) endpoint, accessible via Hypertext Transfer Protocol / Hypertext Transfer Protocol Secure (HTTP / HTTPS). For instance, the invocation information can be the IP (Internet Protocol) address information of a cloud computing instance.

[0054] In this embodiment, the calling information of the target neural network model can be configured in the target workflow node. Generally, one target workflow node is configured with one target neural network model. Of course, one target workflow node can also be configured with multiple target neural network models, which is not limited. In the subsequent process, the workflow is distributed to the target application. During the operation of the target application, when it reaches the target workflow node, the target workflow node is triggered to call the target neural network model based on its configured calling information, and the target neural network model provides services to the target application. Among them, the services provided by the target neural network model to the target application are different depending on the function of the target neural network model. For example, if the target neural network model is implemented as an image classification model, then the image classification model can provide image classification services to the target application. If the target neural network model is implemented as a large language model, then the large language model can provide intelligent question answering services to the target application.

[0055] In this embodiment of the application, a new model deployment method is provided, which is applied to a model application platform. The model application platform provides workflow configuration capabilities and is coupled with a cloud resource platform that provides automatic model deployment capabilities. By integrating the model deployment process into the workflow creation process and combining the automatic model deployment capabilities provided by the cloud resource platform, the integrated deployment of model deployment and application is achieved. The workflow creation is equivalent to the model application process, and the model deployment process is integrated at the same time. Users do not need to manually switch between different platforms, which simplifies the model deployment operation, improves the model deployment efficiency, and provides users with a smooth and seamless model application experience.

[0056] In an optional embodiment, the implementation of displaying a second page for deploying the target neural network model in response to a selection operation of choosing a target neural network model from candidate neural network models is not limited. Considering the user experience, deployment confirmation information for the target neural network model is provided to the user so that the user can choose whether to deploy the target neural network model. Based on this, an implementation of displaying a second page for deploying the target neural network model in response to a selection operation of choosing a target neural network model from candidate neural network models includes: displaying a third page in response to the selection operation of choosing a target neural network model from candidate neural network models, the third page including deployment confirmation information; and displaying the second page in response to a confirmation operation of the deployment confirmation information. Providing a confirmation step before actually deploying the target neural network model provides users with more choices and freedom, improving the user experience.

[0057] Exemplarily, in Figure 3c, a "connection credential" selection control is included on the first page, where the connection credential refers to the connection credential required to connect to the target neural network model. The "connection credential" selection control is implemented as a drop-down menu. When the "connection credential" selection control is triggered, it can be determined whether the target neural network model has been deployed. If the target neural network model has not been deployed, the deployed target neural network models can be displayed in the drop-down menu for the user to select. Further, regardless of whether the target neural network model has not been deployed, the confirmation information for adding a new credential can be displayed. In Figure 3c, taking the deployment confirmation information being implemented as an "add new credential control" as an example for illustration. When the "add new credential control" is triggered, it is determined to initiate a confirmation operation for the confirmation information for adding the new credential, and the model application platform displays the third page. Figure 3d exemplarily shows a third page, which includes deployment confirmation information. The implementation manner of the deployment confirmation information is not limited. For example, on the third page in Figure 3d, taking the deployment confirmation information being implemented as a deployment control as an example, in response to the trigger operation for the deployment control, it is determined to initiate a confirmation operation for the deployment confirmation information.

[0058] In an optional embodiment, the model application platform maintains information on the neural network models that have been deployed, and determines whether the target neural network model has been deployed based on the information on the neural network models that have been deployed. The following is an exemplary introduction.

[0059] For example, when the target neural network model has been deployed, the user can initiate a selection operation for the already deployed target neural network model. In response to this selection operation, the model application platform obtains the call information of the already deployed target neural network model and configures this call information into the target workflow node, realizing the reuse of the deployed model and avoiding the waste of resources caused by repeated deployment.

[0060] Further, when the target neural network model has been deployed and the user wants to deploy a new target neural network model, the user can initiate a confirmation operation for the deployment confirmation information on the third page. In response to this confirmation operation, the model application platform displays the second page.

[0061] For another example, when the target neural network model has not been deployed, the user can initiate a confirmation operation for the deployment confirmation information on the third page. In response to this confirmation operation, the model application platform displays the second page.

[0062] For yet another example, when the target neural network model has not been deployed, the model application platform can directly display the second page, simplifying the user operation and enhancing the user experience.

[0063] Among them, the above examples provide the user with the choice of whether to deploy the target neural network model in combination with whether the target neural network model has been deployed, enabling the user to have more control over the entire process and increasing the user's trust and satisfaction.

[0064] In an optional embodiment, the workflow node includes a connector for connecting different neural network models and applications, enabling data to flow between different applications. The connector includes authentication settings, triggers, and execution actions, etc. Among them, the connector needs to be able to interact with the application securely, which usually involves some form of authentication or authorization. For example, OAuth2.0, API keys, usernames or passwords, etc. OAuth 2.0 is an open standard authorization protocol or framework. The authentication settings can ensure that only authorized applications or users can access specific data (such as the target neural network model). The trigger is used to define when a certain operation should be executed or the workflow node should be started. The trigger can be a scheduled task, event-driven (such as when a new record is inserted into the database) or any other condition that can trigger an action. The trigger can also automatically respond to changes according to the preset logic. The execution action refers to the task that the connector actually needs to execute, such as reading data from the target application, writing data to the target application, calling an API, etc. The connector can receive some input parameters and obtain an output result by calling the target neural network model for subsequent processing.

[0065] Among them, the model application platform can configure connectors to distribute the model call results to any application through the connectors. Among them, some application connectors are maintained on the model application platform. The connectors include the call interfaces of the applications. In addition, for the connectors not maintained by the model application platform, the model application platform supports users to customize application interfaces, create connectors, and configure the interface information into the connectors to distribute the model call results to the corresponding applications through the connectors.

[0066] Exemplarily, the interface information configured in the connector can be the call information of the application. For example, the Uniform Resource Locator (URL) address information of the application. Optionally, in this embodiment, a schematic diagram of the process of creating, model deployment, model application, and model distribution from workflow nodes is provided. As shown in Figure 4, in the process of creating a workflow on the model application platform, create and configure the first workflow node. The first workflow node includes a trigger. For example, the trigger can be a chatbot in a group chat. When a message is provided to the chatbot, the trigger condition is satisfied; create and configure the second workflow node, select a target neural network model from the candidate neural network models, and determine whether the target neural network model has been deployed. If not, display the second page and deploy the target neural network model at one click on the second page; if so, create and configure the third workflow node to distribute the target neural network model. Among them, the third workflow node is used to provide the output result of the target neural network model to the distributed application.

[0067] Among them, in the process of creating the second workflow node, generate the call information of the first workflow node, and configure the call information into the trigger so that when the trigger condition is satisfied, information can be sent to the first workflow node through the call information. Exemplarily, one implementation method of the call information can be the Webhook URL. When a specific event occurs, the Webhook URL can send a notification message from the trigger to the workflow node. More specifically, the Webhook URL can be implemented as an HTTP POST request in the form of an HTTP callback. The request includes the data sent to the specified URL when the trigger is triggered. This URL is the receiving endpoint used to process the incoming data. Among them, POST is a request method in the HTTP protocol, mainly used to submit data to the server.In the process of configuring the third workflow node, the connectors of the target application or other applications can be configured into the third workflow node so that the output information of the target neural network model can be sent to the target application or other applications through the connector.

[0068] In an optional embodiment, the cloud resource platform may maintain source files of some neural network models. For ease of distinction and description, the neural network models whose source files are maintained by the cloud resource platform can be referred to as the first type of neural network model, and the neural network models whose source files are not maintained by the cloud resource platform can be referred to as the second type of neural network model. The candidate neural network models associated with the target workflow nodes may include the first type of neural network model, the second type of neural network model, or both.

[0069] Wherein, when the target neural network model belongs to the first type of neural network model, the model application platform can also provide the identification information of the target neural network model and the resource description information for deploying the target neural network model to the cloud resource platform. The cloud resource platform obtains the resource description information for deploying the target neural network model and the identification information of the target neural network model; creates a cloud computing instance based on the resource description information; obtains the source file of the target neural network model from the local machine based on the identification information; and runs the target neural network model in the cloud computing instance based on the source file.

[0070] When the target neural network model belongs to the second type of neural network model, the model application platform can also obtain the link information of the source files from the third-party platform and send the link information to the cloud resource platform. The link information instructs the cloud resource platform to obtain the source files from the third-party platform. In addition, the model application platform provides resource description information to the cloud resource platform, which creates a cloud computing instance and runs the target neural network model in the cloud computing instance based on the source files obtained from the third-party platform. By obtaining the link information of the source files from the third-party platform, a wider range of model sources can be supported, improving interoperability and flexibility.

[0071] The execution order of the two steps, "obtaining the link information of the source files from the third-party platform" and "obtaining the resource description information required for deploying the target neural network model in response to the configuration operation on the second page," is not limited. Optionally, a fourth page may be displayed before displaying the second page. The fourth page includes configuration items corresponding to the link information, and the link information of the source files from the third-party platform is obtained in response to the configuration operation on the configuration items. Furthermore, the fourth page also includes a resource description information configuration control, and the second page is displayed in response to the trigger operation on the resource description information configuration control.

[0072] In an optional embodiment, the implementation method for obtaining the resource description information required for deploying the target neural network model in response to the configuration operation on the second page is not limited. A specific implementation method is provided below, in which the instance specification information of the virtualization instance carrying the target neural network model is obtained as the resource description information required for deploying the target neural network model in response to the configuration operation on the second page; wherein, the resource description information is used by the cloud resource platform to create the target virtualization instance and deploy the target neural network model in the target virtualization instance according to the source files of the target neural network model. By allowing users to customize the instance specifications, it can be ensured that the most suitable computing resources are provided for each model, avoiding resource waste or insufficiency. At the same time, it allows users to adjust the deployment configuration according to actual needs, increasing flexibility and user control.

[0073] The instance specification information describes the specific parameters required to deploy a cloud computing instance. These parameters determine the performance and characteristics of the cloud computing instance created by the cloud resource platform. For deploying a target neural network model, appropriate instance specification information ensures that the target neural network model obtains sufficient computing resources during runtime, thereby guaranteeing its performance and stability. Instance specification information may include, but is not limited to, hardware resource parameters such as the number of computing unit cores, memory size, and storage capacity, as well as software environment parameters such as the operating system type. These ensure that the created cloud computing instance can meet the requirements for running the target neural network model.

[0074] In an optional embodiment, the method for displaying the second page is not limited. The user pre-creates a resource account on the model application platform. The resource account information includes, but is not limited to: nickname, identifier (ID), authentication and authorization information, and contact information. The user can use this resource account to create workflows. When it is necessary to display the second page, the user can send the resource account information used to create the workflow to the cloud resource platform for authentication. If the authentication is successful, the cloud resource platform will return the resource file for the model deployment page; the resource file will then be rendered to display the second page on the first page. For a detailed description of the model deployment page and resource file, please refer to the foregoing embodiments; further details will not be repeated here.

[0075] In this way, by sending the resource account information to the cloud resource platform for authentication, it is ensured that the authenticated resource account can access the cloud resource platform, which greatly enhances security and prevents unauthorized access. At the same time, if the authentication is successful, there is no need to jump to the cloud resource platform, nor is there a need to perform complex login, verification, and target neural network model location operations on the cloud resource platform. Users can access the cloud resource platform model deployment page on the model application platform, realizing the integration of model deployment and application, and simplifying user operations.

[0076] The authentication process of the cloud resource platform is not limited. For example, after receiving a request, the cloud resource platform parses and verifies the provided resource account information. For example, it checks whether the username / password, API key, etc. are correct; and confirms whether the user has the right to perform the required operation, such as creating a virtualization instance or deploying a model. If the authentication is successful, the cloud resource platform will return a success response; otherwise, it will return an error message.

[0077] Wherein, if the authentication is successful, the cloud resource platform can directly return the resource files of the model deployment page, or the cloud resource platform can return a notification message of successful authentication to the model application platform. After receiving the notification message of successful authentication, the model application platform can send a request to the cloud resource platform to obtain the resource files required for the model deployment page. After receiving the request, the cloud resource platform can return the resource files of the model deployment page.

[0078] In an optional embodiment, the implementation of configuring the call information into the target workflow node is not limited. The target workflow node includes initial metadata information, which at least includes a first attribute field describing the model call information. One implementation of configuring the call information into the target workflow node includes: configuring the call information of the target neural network model as the attribute value of the first attribute field to obtain target metadata.

[0079] For example, the first attribute field can be a default field or an address field, without limitation. The first attribute field and its attribute value can be represented in key-value form. For example, if the address is implemented as a URL, it could be {Default: URL}, or {Address: URL}.

[0080] By configuring the calling information of the target neural network model as the attribute value of the first attribute field in the initial metadata information, this method significantly improves the flexibility of the workflow, development efficiency, and user experience. It not only simplifies the model integration process but also enhances the security and controllability of the system, promoting the development of automation and intelligence.Overall, this approach provides users with an efficient and easy-to-use platform to manage and deploy neural network models, thereby accelerating the transition from development to production.

[0081] Optionally, for the target workflow node including initial metadata information, at least one of the following configuration operations can also be performed:

[0082] Example F1, Version Configuration Operation: The source file of the target neural network model may change, such as through upgrades or improvements. That is, the version information of the target neural network model may change, and different versions of the target neural network model may have different API interface designs, functional characteristics, or configuration options. Therefore, as the target neural network model is upgraded and improved, the format and structure of the output call information may also change. In order to enable the target metadata of the target workflow node to support multiple version information of the target neural network model, at least two version configuration information can be generated for the target neural network model on the cloud resource platform. The version configuration information is used to record the call information of the target neural network model. Different version configuration information includes different call information, and the version range of the target neural network model corresponding to different version configuration information is different. For example, version configuration information E1 corresponds to version range 1-3, and version configuration information E1 corresponds to version range 3-latest version. For example, taking the call information as a URL as an example, one example of version configuration information E1 is: version 1-3, call information is: {URL}, and another example of version configuration information E1 is: version > 3, {call information is: "URL"}.

[0083] Based on this, the model application platform can obtain at least two version configuration information of the target neural network model from the cloud resource platform; configure at least two version configuration information into the target metadata so that when the target neural network model is called, the calling information can be obtained from the corresponding version configuration information according to the current version information of the target neural network model, so that the target metadata of the target workflow node supports multiple version information of the target neural network model.

[0084] Example F2, Network Configuration Operation: Configure the network address of the target neural network model or cloud resource platform. The cloud resource platform and the workflow creation platform may be located in the same local area network (LAN) or different LANs. If the cloud resource platform and the workflow creation platform (e.g., the model application platform) are located in the same LAN, the model application platform can access the cloud resource platform through the intranet without consuming the user's network resources, providing a faster and more secure network environment. If the cloud resource platform and the workflow creation platform (e.g., the model application platform) are located in different LANs, the model application platform can access the cloud resource platform through the internet.

[0085] An implementation structure of the target metadata of the target workflow node is provided below, but the implementation structure of the target metadata is not limited to this.

[0086]

[0087] Among them, Metadata describes the metadata information of the workflow node and contains some key information about the target neural network model. The following is an explanation of these fields: Service Identification (ServiceId), which is the identification information of the cloud computing instance where the target neural network model is deployed and is used to distinguish different cloud computing instances. Service Name (ServiceName): Helps to identify and manage the services provided by the model. Output is used to describe the relevant information output to the model application platform. Default Output, where the version number in the version upgrade is 3, indicating that when the version information of the target neural network model is less than 3, the call information is parsed from the default output; Internal Network Information (Internal), indicating that the cloud resource platform is accessed through this internal network address; Upgrade List represents another version configuration information different from the default output; Version Number, indicating that when the version number is greater than 3, the call information is parsed from the keyword (key) field; Keyword (key), which is used to record the call information; Internal Network Information (Internal) indicates that the cloud resource platform is accessed through this internal network address.

[0088] In the embodiment of the present application, combining the characteristics of the model application platform and the cloud resource platform, making full use of various advantages on the cloud, a one-stop solution from model deployment to model application call is provided for users, and all the needs of users from neural network model deployment to actual production use can be completed on one platform without multiple platform jumps.

[0089] Regarding the detailed implementation manners and beneficial effects of each step in the method shown in Figure 2 provided in the embodiment of the present application, they have been described in detail in the foregoing embodiments and will not be elaborated here.

[0090] It should be noted that the execution subject of each step of the method provided in the above embodiment can be the same device, or the method can also be executed by different devices as the execution subject.For example, the execution subject of steps 201 to 203 can be device A; for another example, the execution subject of steps 201 and 202 can be device A, and the execution subject of step 203 can be device B; and so on.

[0091] In addition, in some of the processes described in the above embodiments and accompanying drawings, multiple operations appear in a specific order. However, it should be clearly understood that these operations can be executed not in the order in which they appear in this article or in parallel. The operation numbers such as 201, 202, etc. are only used to distinguish different operations, and the numbers themselves do not represent any execution order. In addition, these processes can include more or fewer operations, and these operations can be executed in sequence or in parallel. It should be noted that the descriptions such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., do not represent a sequence, and do not limit that "first" and "second" are of different types.

[0092] Figure 5 is a schematic structural diagram of a model deployment device provided by an exemplary embodiment of the present application. As shown in <a href='javascript:;' class="see-img-anchor" img-id="HDA00052740996700

[0096] The configuration module 54 is used to configure the call information into the target workflow node so as to trigger the target workflow node to call the target neural network model based on the call information during the operation of the target application.

[0097] In an optional embodiment, the display module is specifically used to: display a third page in response to a selection operation of selecting a target neural network model from candidate neural network models, the third page including deployment confirmation information, and display a second page in response to a confirmation operation of the deployment confirmation information.

[0098] In an optional embodiment, the device further includes: a judgment module; the judgment module is used to determine whether the target neural network model has been deployed based on the information of the deployed neural network model; and a display module is used to perform the operation of displaying a second page if the target neural network model has not been deployed.

[0099] In an optional embodiment, the candidate neural network model includes a first type of neural network model with source files on the cloud resource platform and a second type of neural network model without source files; when the target neural network model belongs to the second type of neural network model, the acquisition module is used to acquire the link information of the source files from the third-party platform and send the link information to the cloud resource platform. The link information is used to instruct the cloud resource platform to acquire the source files from the third-party platform.

[0100] In an optional embodiment, the acquisition module is specifically used to: in response to the configuration operation on the second page, acquire the instance specification information of the virtualization instance carrying the target neural network model as the resource description information required for deploying the target neural network model; wherein, the resource description information is used for the cloud resource platform to create the target virtualization instance, and deploy the target neural network model in the target virtualization instance according to the source file of the target neural network model.

[0101] In an optional embodiment, the display module is specifically used to: send the resource account information used to create the workflow to the cloud resource platform so that the cloud resource platform can authenticate the resource account information and return the resource file of the model deployment page if the authentication is successful; and render the resource file to display the second page on the first page.

[0102] In an optional embodiment, the target workflow node includes initial metadata information, which includes at least a first attribute field for describing model call information; the configuration module is specifically used to: configure the call information of the target neural network model as the attribute value of the first attribute field to obtain the target metadata.

[0103] Optionally, the configuration module is further configured to perform at least one of the following operations: version configuration operation: obtaining at least two version configuration information of the target neural network model from the cloud resource platform, where the version ranges of the target neural network models corresponding to different version configuration information are different, and the call information included in different version configuration information is different; configuring the at least two version configuration information into the target metadata, so as to obtain call information from the corresponding version configuration information according to the current version information of the target neural network model when the target neural network model is called; network configuration operation: configuring the network address of the target neural network model or the cloud resource platform, where the cloud resource platform and the workflow creation platform are located in the same local area network or different local area networks.

[0104] The detailed implementation manners and beneficial effects of each step in the device shown in Figure 4 provided in the embodiments of the present application have been described in detail in the foregoing embodiments, and will not be elaborated here.

[0105] Figure 6 is a schematic structural diagram of an electronic device provided in an exemplary embodiment of the present application. As shown in Figure 6, the device includes: a memory 64 and a processor 65.

[0106] The memory 64 is used to store computer programs and can be configured to store various other data to support operations on the electronic device. Examples of these data include instructions for any application program or method for operating on the electronic device, etc.

[0107] The processor 65 is coupled to the memory 64 and is used to execute the computer program in the memory 64 for: executing the model deployment method described in the foregoing embodiments. For details, please refer to the descriptions in the foregoing embodiments, and will not be elaborated here again.

[0108] Regarding the detailed implementation manners and beneficial effects of each step in the device shown in Figure 6 provided in the embodiments of the present application, they have been described in detail in the foregoing embodiments, and will not be elaborated here.

[0109] Furthermore, as shown in Figure 6, the electronic device further includes: other components such as a communication component 66, a display 67, a power supply component 68, an audio component 69, etc. Only some components are schematically shown in Figure 6, which does not mean that the electronic device only includes the components shown in Figure 6. In addition, the components within the dashed box in Figure 6 are optional components, rather than mandatory components, and can be determined according to the product form of the electronic device. The electronic device in this embodiment can be implemented as a desktop computer, a laptop computer, a smart phone, or an IOT (Internet of Things) device, as well as a smart wearable device (such as a smart watch, a smart bracelet), etc. terminal devices, or can also be a server-side device such as a conventional server, a cloud server, or a server array.If the electronic device in this embodiment is implemented as a terminal device such as a desktop computer, a laptop computer, or a smartphone, it may include the components within the dashed box in Figure 6; if the electronic device in this embodiment is implemented as a server device such as a conventional server, a cloud server, or a server array, it may not include the components within the dashed box in Figure 6.

[0110] The detailed implementation manners and beneficial effects of the electronic device provided in the embodiments of this application have been described in detail in the foregoing embodiments, and will not be elaborated here.

[0111] The above-mentioned memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0112] The above-mentioned communication component is configured to facilitate communication between the device where the communication component is located and other devices in a wired or wireless manner. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component further includes a near field communication (NFC) module to facilitate short-range communication.

[0113] The above-mentioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can not only sense the boundaries of touch or swipe actions, but also detect the duration and pressure associated with the touch or swipe operation.

[0114] The above-mentioned power supply component provides power for various components of the device where the power supply component is located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device where the power supply component is located.

[0115] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), which is configured to receive external audio signals when the device where the audio component is located is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals can be further stored in the memory or transmitted via the communication component. In some embodiments, the audio component further includes a speaker for outputting audio signals.

[0116] Correspondingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to be able to implement each step executable by an electronic device in the method embodiment shown in the above Figure 2.

[0117] Among them, the computer-readable storage medium can be implemented by volatile or non-volatile or a combination thereof, and can be removable or non-removable.Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technologies, CD-ROM (compact disc read-only memory), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices or any other non-transmission medium.

[0118] Correspondingly, an embodiment of the present application further provides a computer program product, including a computer program / instructions, which, when executed by a processor, cause the processor to implement each step executable by an electronic device in the method embodiment shown in Figure 2 above.

[0119] It should be understood that each process or the combination of multiple processes in the above method flow can be implemented by a computer program or instructions. In addition, these computer programs or instructions can be applied to the processors of general-purpose computers, special-purpose computers, embedded processors or other programmable data processing devices, so that the processors of general-purpose computers, special-purpose computers, embedded processors or other programmable data processing devices can be used as devices to implement the corresponding functions in the above method embodiments.

[0120] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, commodity or device including the element.

[0121] The above are only embodiments of the present application and are not used to limit the present application.Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.< / iframe>

Claims

1. A model deployment method, characterized in that, include: The first page is displayed, which includes a workflow associated with the target application. The workflow includes at least a target workflow node for model invocation, and the target workflow node is associated with a candidate neural network model. In response to a selection operation that selects a target neural network model from the candidate neural network models, a second page for deploying the target neural network model is displayed, the second page being associated with a cloud resource platform with automated model deployment capabilities; In response to the configuration operation on the second page, the resource description information required to deploy the target neural network model is obtained, and the resource description information is sent to the cloud resource platform. The resource description information is used to instruct the cloud resource platform to deploy the target neural network model and return the calling information of the target neural network model. The invocation information is configured into the target workflow node so that the target workflow node can be triggered to invoke the target neural network model based on the invocation information during the operation of the target application.

2. The method according to claim 1, characterized in that, In response to a selection operation that chooses a target neural network model from the candidate neural network models, a second page for deploying the target neural network model is displayed, including: In response to a selection operation that chooses a target neural network model from candidate neural network models, a third page is displayed, which includes deployment confirmation information. In response to a confirmation operation of the deployment confirmation information, a second page is displayed.

3. The method according to claim 1, characterized in that, Before displaying the second page, it also includes: Based on the information of the already deployed neural network models, determine whether the target neural network model has been deployed; If the target neural network model is not deployed, then the operation of displaying the second page is performed.

4. The method according to claim 1, characterized in that, The candidate neural network models include a first type of neural network model with source files and a second type of neural network model without source files on the cloud resource platform. If the target neural network model belongs to the second type of neural network model, the method further includes: Obtain the link information of the source file from the third-party platform, and send the link information to the cloud resource platform. The link information is used to instruct the source file to be obtained from the third-party platform.

5. The method according to claim 1, characterized in that, In response to the configuration operation on the second page, obtain the resource description information required to deploy the target neural network model, including: In response to the configuration operation on the second page, the instance specification information of the virtualization instance carrying the target neural network model is obtained as the resource description information required to deploy the target neural network model; The resource description information is used by the cloud resource platform to create a target virtualization instance and deploy the target neural network model in the target virtualization instance according to the source file of the target neural network model.

6. The method according to claim 1, characterized in that, The second page is displayed, including: The resource account information used to create the workflow is sent to the cloud resource platform so that the cloud resource platform can authenticate the resource account information and return the resource files of the model deployment page if the authentication is successful. The resource file is rendered to display the second page on the first page.

7. The method according to claim 1, characterized in that, The target workflow node includes initial metadata information, which includes at least a first attribute field for describing model call information; Configuring the invocation information into the target workflow node includes: The calling information of the target neural network model is configured as the attribute value of the first attribute field to obtain the target metadata.

8. The method according to claim 7, characterized in that, It also includes at least one of the following configuration operations: Version configuration operation: Obtain at least two version configuration information of the target neural network model from the cloud resource platform. The version range of the target neural network model corresponding to different version configuration information is different, and the calling information included in different version configuration information is different. Configure the at least two version configuration information into the target metadata so that when calling the target neural network model, the calling information can be obtained from the corresponding version configuration information according to the current version information of the target neural network model. Network configuration operation: Configure the network address of the target neural network model or the cloud resource platform, wherein the cloud resource platform and the workflow creation platform are located on the same local area network or different local area networks.

9. An electronic device, characterized in that, include: Memory and processor; The memory is used to store a computer program; the processor, coupled to the memory, is used to execute the computer program to implement the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method according to any one of claims 1-8.

11. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1-8.

Images