An AI algorithm deployment running method, device and medium

Abstract

The embodiments of the present specification disclose an AI algorithm deployment and running method, device and medium, comprising: when a specified AI algorithm is deployed, determining a deployment mode corresponding to the specified AI algorithm; deploying the specified AI algorithm to a host server through the deployment mode corresponding to the specified AI algorithm to form a specified AI algorithm model; determining a running environment corresponding to the specified AI algorithm model; and when the specified AI algorithm model is running in the running environment, performing state monitoring and start-stop control on the specified AI algorithm model. The embodiments of the present specification solve the problem that algorithms and dependent environments cannot be deployed simultaneously by determining the deployment mode corresponding to the specified AI algorithm, and solve the problem that the start-stop mode of algorithms is different after differentiated deployment by determining the running environment corresponding to the specified AI algorithm model.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a method, device, and medium for deploying and running an AI algorithm. Background Technology

[0002] With the development of AI technology, various related technologies have been applied, which, while promoting the development of artificial intelligence, has also created obstacles to the unified management and maintenance of AI algorithms. AI algorithms developed using different graphics processing frameworks, deep learning frameworks, and machine learning frameworks cannot run in the same environment. Summary of the Invention

[0003] This specification provides one or more embodiments of an AI algorithm deployment and operation method, device, and medium to solve the technical problems raised in the background art.

[0004] One or more embodiments of this specification employ the following technical solutions:

[0005] This specification provides a method for deploying and running an AI algorithm through one or more embodiments, including:

[0006] When deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm;

[0007] The specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm, thereby forming the specified AI algorithm model;

[0008] Determine the operating environment corresponding to the specified AI algorithm model;

[0009] When the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is subject to status monitoring and start / stop control.

[0010] This specification provides one or more embodiments of an AI algorithm deployment and operation device, comprising:

[0011] At least one processor; and,

[0012] A memory communicatively connected to the at least one processor; wherein,

[0013] The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to:

[0014] When deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm;

[0015] The specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm, thereby forming the specified AI algorithm model;

[0016] Determine the operating environment corresponding to the specified AI algorithm model;

[0017] When the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is subject to status monitoring and start / stop control.

[0018] This specification provides one or more embodiments of a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows:

[0019] When deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm;

[0020] The specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm, thereby forming the specified AI algorithm model;

[0021] Determine the operating environment corresponding to the specified AI algorithm model;

[0022] When the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is subject to status monitoring and start / stop control.

[0023] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects:

[0024] The embodiments in this specification solve the problem of not being able to deploy simultaneously due to differences in algorithms and dependent environments by determining the deployment method corresponding to a specified AI algorithm, and solve the problem of different algorithm start-up and shutdown methods after differentiated deployment by determining the running environment corresponding to a specified AI algorithm model. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0026] Figure 1 A flowchart illustrating a method for deploying and running an AI algorithm provided in one or more embodiments of this specification;

[0027] Figure 2 This is a schematic diagram of the structure of an AI algorithm deployment and operation device provided for one or more embodiments of this specification. Detailed Implementation

[0028] This specification provides an embodiment of a method, device, and medium for deploying and running an AI algorithm.

[0029] With the development of AI technology, various related technologies are gradually increasing. While promoting the development of artificial intelligence technology, this also creates obstacles to the unified management and maintenance of AI algorithms. AI algorithms developed using different graphics processing frameworks, deep learning frameworks, and machine learning frameworks cannot run in the same environment. Therefore, the embodiments in this specification can adopt container technology and environment isolation methods to enable algorithms developed using different frameworks to be easily deployed and managed on the same device, solving the problem of unification caused by the diversity of AI algorithm technology stacks from different manufacturers.

[0030] As AI technology is increasingly applied across various industries, single devices can no longer meet the needs of large application scenarios. AI algorithms need to be deployed on different server devices, making unified management of AI algorithms difficult and increasing maintenance costs. The embodiments in this specification can achieve remote control of AI algorithms through remote command transmission and establishing a long-lived link between the control and computing ends.

[0031] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0032] Figure 1 This diagram illustrates a process for deploying and running an AI algorithm according to one or more embodiments of this specification. This process can be executed by an AI algorithm deployment and running system. Certain input parameters or intermediate results in the process can be manually adjusted to help improve accuracy.

[0033] The method flow steps of the embodiments in this specification are as follows:

[0034] S102, when deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm.

[0035] S104, the specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm to form the specified AI algorithm model.

[0036] In the embodiments of this specification, AI algorithm deployment methods may include direct deployment to a server environment and containerized deployment. For different AI algorithm deployment methods, different processing methods can be used during the process of deploying the image file to the host server according to the AI ​​algorithm deployment method corresponding to the image file to form the specified AI algorithm model.

[0037] When the AI ​​algorithm is deployed directly to the server environment, it is deployed to the host server according to the preset directory structure of the specified AI algorithm to form the specified AI algorithm model;

[0038] When the AI ​​algorithm is deployed using containerization, the specified AI algorithm and its corresponding specified dependency packages are packaged into an image file using containerization technology, and the image file is deployed to the host server to form the specified AI algorithm model.

[0039] In the embodiments described in this specification, the specified dependency packages may include public dependency packages and algorithm dependency packages;

[0040] When the specified AI algorithm and its corresponding specified dependency package are packaged into an image file using containerization technology, the common dependency package can be transferred to the common runtime environment of the host server, which can save the space occupied by the image file; the specified AI algorithm and the algorithm dependency package are packaged into the image file using containerization technology.

[0041] In the embodiments described in this specification, the common operating environment includes multiple graphics processing frameworks and multiple deep learning frameworks;

[0042] Before transferring the public dependency package to the public runtime environment of the host server, in order to prevent the existence of multiple versions of a single dependency package in the public runtime environment, which could lead to errors when the AI ​​algorithm references the dependency, the differences in the public runtime environments for different AI algorithms can be compared first to form multiple sets of public runtime environments.

[0043] In the embodiments of this specification, if the common dependency package and the algorithm dependency package are incompatible, the incompatible dependency packages are arranged in a way that separates their respective algorithm dependency package directories through the common dependency environment, and the effect of isolating different algorithm dependency environments is achieved by referencing Anaconda.

[0044] S106, Determine the operating environment corresponding to the specified AI algorithm model.

[0045] S108, when the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is monitored for status and controlled for start and stop.

[0046] In the embodiments of this specification, when performing state monitoring and start / stop control on the specified AI algorithm model, the model deployment method corresponding to the specified AI algorithm model can be determined first; then, according to the model deployment method corresponding to the specified AI algorithm model, the corresponding processing method can be used to perform state monitoring and start / stop control on the specified AI algorithm model.

[0047] In the embodiments described in this specification, the model is deployed in a single-server manner. During the process of monitoring the status and controlling the start and stop of the specified AI algorithm model using the corresponding processing method:

[0048] 1) Containerized deployment: A set of container client toolkits is encapsulated. The toolkit includes container startup, stop, restart, creation, removal, single container status viewing, and server container status viewing. By controlling the container status, the purpose of controlling the start and stop of the algorithm can be achieved.

[0049] The application's toolkit includes:

[0050] Startup: First, configure the specified AI algorithm to start automatically in the container. When the container starts, the startup script of the algorithm can be automatically called. The startup command of the toolkit is used to find the image container containing the specified AI algorithm according to the number and start the container to realize the startup of the specified AI algorithm.

[0051] Stop: Locate the specified container based on its number and stop the container to halt the specified AI algorithm;

[0052] 2) Direct deployment:

[0053] The specified AI algorithm is deployed directly without container files, and the algorithm cannot be started or stopped by controlling the container.

[0054] Startup: Add a startup script to the specified AI algorithm and place it in the specified directory. Find the algorithm to be started by its number, switch to the corresponding directory, execute the startup script, and start the algorithm.

[0055] Stop: Record the name of the specified AI algorithm, use the name of the specified AI algorithm to execute operating system commands on the server to find the running process, obtain the process number, and then kill the process using operating system commands to stop the algorithm.

[0056] In the embodiments described in this specification, the model is deployed in a multi-cloud manner, and the corresponding processing method is used to monitor the status and control the start and stop of the specified AI algorithm model:

[0057] 1) Containerized deployment:

[0058] By introducing a container client toolkit, a long-lived socket connection is established, creating a TCP network connection between the management end and the algorithm deployment end that will not be closed, enabling continuous and multiple transmissions of HTTP data.

[0059] After the long connection is created, the management side can call the command tools of the container client toolkit to execute container start and stop commands. After the algorithm deployment side receives the commands, it executes the algorithm to start and stop on the server.

[0060] 2) Direct deployment:

[0061] The specified AI algorithm is directly deployed. Script files for starting and stopping the algorithm are added to the algorithm deployment server. During remote control, the management end sends an HTTP request. After the algorithm deployment server interface receives the HTTP request command, the runtime command is used to execute the local start and stop scripts, thus starting and stopping the algorithm.

[0062] The biggest advantages of the AI ​​algorithm operation support in the embodiments of this specification include two points: for AI algorithms developed using different graphics processing frameworks, deep learning frameworks, machine learning frameworks, etc., it can avoid the differences in dependent environments and achieve unified deployment and start / stop control; it can realize start / stop control of AI algorithms deployed on different server devices.

[0063] It should be noted that the embodiments of this specification disclose a method for supporting and controlling the operation of AI algorithms, which enables the unified deployment, simultaneous operation, and start-stop control of AI algorithms with different architectures and dependencies in a server environment. This method can solve the problem of not being able to deploy simultaneously due to differences in algorithms and dependent environments, as well as the problem of different start-stop methods for algorithms after differentiated deployment.

[0064] The methods provided in the embodiments of this specification mainly include three core technical means: providing a combination of a common runtime environment directory and an independent dependency environment directory for each algorithm for the running environment of AI algorithms.

[0065] The AI ​​algorithm can be deployed in two ways: containerized deployment and direct deployment in server environment; the AI ​​algorithm can be started and stopped in both single-server and multi-cloud deployment modes.

[0066] I. This specification provides various AI algorithm deployment methods in its embodiments.

[0067] 1. Containerized deployment of AI algorithms

[0068] When deploying AI algorithms, containerization technology is used to package the AI ​​algorithm and its dependencies into a single portable image. By using container technology, AI algorithms can be ported to any server environment and run without relying on a dedicated server environment.

[0069] Using container technology to package AI algorithms into images and achieve virtualization solves the environment dependency problem. However, it also creates a problem of very large image files after packaging all content; for example, video analysis AI algorithms typically package to around 10GB. This specification's embodiments address this issue by transferring some common dependency packages to the host machine's public runtime environment. By migrating large, shared dependency packages from the container image file and replacing them with dependencies from the public runtime environment, significant space is saved in the AI ​​algorithm image file. While a typical video analysis AI algorithm and its dependency packages packaged directly are around 10GB, migrating OpenCV, CUDA, and PyTorch from the container and then repackaging them results in an image file approximately 1GB in size.

[0070] 2. Supports multiple algorithm deployment methods coexisting.

[0071] To better adapt to AI algorithms with different architectures and address the challenge of deploying algorithms developed using different frameworks on the same device, the embodiments in this specification support multiple AI algorithm deployment methods. This approach also enhances the usability of the AI ​​algorithm management system, making it easier to deploy and maintain different AI algorithms. The AI ​​algorithm deployment methods supported by the embodiments in this specification include the following two:

[0072] 1) Direct deployment on server environment

[0073] This deployment method is suitable for AI algorithms that are better adapted to the public operating environment of servers, are easy to start and stop, and have a more standardized structure. AI algorithms deployed using this method do not require the process of packaging an image; they are directly deployed to the server according to the algorithm directory structure, and the algorithm is started and stopped using specified start and stop script files and commands.

[0074] 2) Containerized deployment

[0075] This deployment method is suitable for AI algorithms that are not fully compatible with the public operating environment of servers, have complex dependencies, are relatively cumbersome to deploy and start up, or are not suitable for direct deployment. These algorithms are packaged into images and deployed and managed through containers, which makes it easier to migrate algorithms, simplify management, and resolve issues related to differences in algorithm dependencies.

[0076] II. The embodiments in this specification provide the operating environment required for the AI ​​algorithm.

[0077] 1. Provide a public operating environment for AI algorithms

[0078] All AI algorithm models rely on a runtime environment. Different AI algorithms, due to variations in their application architecture or analytical objectives, will have significantly different runtime environment requirements. These differences primarily lie in the types and versions of dependency packages; in an incompatible environment, the AI ​​algorithm cannot function properly.

[0079] This specification provides a public runtime environment on an AI algorithm deployment server. This environment can deploy a public runtime environment directory, which includes the technical architectures upon which the following common AI algorithms rely: graphics processing frameworks such as OpenCV, Pillow, and NumPy; deep learning frameworks such as TensorFlow, Keras, PyTorch, and Darknet; and GPU-dependent CUDA. This environment can be used for all AI algorithms running on the server.

[0080] To prevent errors in AI algorithms due to multiple versions of a single dependency package existing in a public runtime environment, multiple public runtime environments can be created by comparing the differences in the public runtime environments that the AI ​​algorithms depend on. The embodiments in this specification support creating multiple public runtime environment directories on a single server. Different public runtime environment directories can be specified for different algorithms.

[0081] 2. AI Algorithm Independent Dependency Environment Directory

[0082] This specification provides a common runtime environment for all algorithms, while also supporting the specification of a separate dependency environment directory for each AI algorithm. This combination of the common runtime environment and the independent dependency environment directories for each algorithm addresses the different environments encountered by AI algorithms.

[0083] Each algorithm has its own independent dependency environment directory, and the processing method will vary depending on how the AI ​​algorithm is deployed.

[0084] For AI algorithms deployed directly on the server environment, if some dependency packages are incompatible with the content and versions of dependency packages in the public runtime environment, some dependencies can utilize the public dependency environment, while the incompatible dependency packages should be placed in their own independent dependency package directories. Furthermore, by referencing open-source technologies such as Anaconda, the isolation between different algorithm dependency environments is achieved, preventing multiple dependency environment directories from affecting the operation of the AI ​​algorithm.

[0085] When containerizing AI algorithms, the Dockerfile only needs to specify dependency packages different from the public runtime environment during image generation. These different packages are then packaged into the image file, while other dependencies can directly reference the host machine's public runtime environment. This ensures the algorithm runs correctly and usably, while also utilizing the public runtime environment. The image file size is also relatively small, facilitating migration and deployment.

[0086] III. The embodiments in this specification provide a method for remote control using AI algorithms.

[0087] After AI algorithms are deployed to a server environment, their status needs to be monitored and their start / stop controlled. With the widespread adoption of cloud-native technologies, multi-cloud deployment is more common than single-server deployment. The embodiments in this specification employ different processing methods depending on whether the AI ​​algorithm is deployed in a multi-cloud environment.

[0088] 1. Single-server deployment method

[0089] 1) Containerized deployment:

[0090] A container client toolkit has been encapsulated, including tools for starting, stopping, restarting, creating, and removing containers, viewing the status of single containers, and viewing the status of server containers. By controlling the container status, the goal of controlling the start and stop of algorithms can be achieved.

[0091] The main application toolkits include:

[0092] Startup: First, configure the AI ​​algorithm to start automatically in the container, so that the algorithm's startup script can be automatically called when the container starts. Using the toolkit's startup command, locate the image container containing the AI ​​algorithm based on its ID, and start the container to launch the AI ​​algorithm.

[0093] Stop: Locate the specified container based on its number and stop the container to halt the AI ​​algorithm.

[0094] 2) Direct deployment:

[0095] The AI ​​algorithm is deployed directly without container files, so the algorithm cannot be started or stopped by controlling the container.

[0096] Startup: Add a startup script to the AI ​​algorithm and place it in the specified directory. Find the algorithm to be started by its number, switch to the corresponding directory, execute the startup script, and start the algorithm.

[0097] Stop: Record the names of each AI algorithm. Using the algorithm name, execute the operating system command "ps -ef|grep [algorithm name]" on the server to find the running process, obtain the process ID, and then kill the process using the operating system command "kill -9 [process ID]" to stop the algorithm.

[0098] 2. Multi-cloud deployment method

[0099] Compared to single-server deployment, multi-cloud deployment requires not only starting and stopping algorithms in different ways, but also managing other cloud node servers and remotely controlling other AI deployment servers to start and stop algorithms.

[0100] It provides AI server management functionality, allowing online management of servers. By configuring server IP addresses, remote usernames, and passwords, servers can be added to a cluster for management. It also offers one-click start / stop control for AI algorithms deployed across multiple cloud nodes. After adding AI computing servers from different cloud nodes to the computing cluster, the underlying layer distinguishes between the algorithm's deployment method and the deployed server when starting or stopping an algorithm, enabling remote start / stop control without requiring manual connection to the remote server.

[0101] 1) Containerized deployment:

[0102] By introducing a container client toolkit, a long-lived socket connection is established, creating a TCP network connection between the management end and the algorithm deployment end that will not be closed, allowing for continuous and multiple transmissions of HTTP data.

[0103] After the long connection is created, the management end can call the command tools of the container client toolkit to execute container start and stop commands. After the algorithm deployment end receives the commands, it executes the algorithm to start and stop on the server.

[0104] 2) Direct deployment:

[0105] For directly deployed AI algorithms, scripts for starting and stopping the algorithm are added to the algorithm deployment server. For remote control, the management end sends an HTTP request. After the algorithm deployment server interface receives the HTTP request command, the runtime command is used to execute the local start and stop scripts, thus starting and stopping the algorithm.

[0106] Figure 2 A schematic diagram of the structure of an AI algorithm deployment and operation device provided for one or more embodiments of this specification includes:

[0107] At least one processor; and,

[0108] A memory communicatively connected to the at least one processor; wherein,

[0109] The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to:

[0110] When deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm;

[0111] The specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm, thereby forming the specified AI algorithm model;

[0112] Determine the operating environment corresponding to the specified AI algorithm model;

[0113] When the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is subject to status monitoring and start / stop control.

[0114] This specification provides one or more embodiments of a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows:

[0115] When deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm;

[0116] The specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm, thereby forming the specified AI algorithm model;

[0117] Determine the operating environment corresponding to the specified AI algorithm model;

[0118] When the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is subject to status monitoring and start / stop control.

[0119] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for devices and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0120] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0121] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for deploying and running an AI algorithm, characterized in that, The method includes: When deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm; The specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm, thereby forming the specified AI algorithm model; Determine the operating environment corresponding to the specified AI algorithm model; When the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is subject to status monitoring and start / stop control. The AI ​​algorithm deployment methods include direct deployment in the server environment and containerized deployment. The step of deploying the specified AI algorithm to the host server using the deployment method corresponding to the specified AI algorithm to form the specified AI algorithm model includes: When the AI ​​algorithm is deployed directly in the server environment, it is deployed to the host server according to the preset directory structure of the specified AI algorithm to form the specified AI algorithm model. When the AI ​​algorithm is deployed in a containerized manner, the specified AI algorithm and its corresponding specified dependency packages are packaged into an image file using containerization technology, and the image file is deployed to the host server to form the specified AI algorithm model. The specified dependency packages include public dependency packages and algorithm dependency packages; The step of packaging the specified AI algorithm and its corresponding specified dependency packages into an image file using containerization technology includes: The public dependency package is transferred to the public runtime environment of the host server; The specified AI algorithm and its dependency packages are packaged into the image file using containerization technology. The method further includes: If the common dependency package is incompatible with the algorithm dependency package, the incompatible dependency package is isolated from the algorithm by using the common dependency environment and by referencing Anaconda. The process of monitoring the state and controlling the start / stop of the specified AI algorithm model includes: Determine the model deployment method corresponding to the specified AI algorithm model; Based on the model deployment method corresponding to the specified AI algorithm model, the corresponding processing method is used to perform status monitoring and start / stop control on the specified AI algorithm model; The model is deployed using a single-server deployment method. The step of using corresponding processing methods to monitor the state and control the start / stop of the specified AI algorithm model includes: 1) Containerized deployment: A set of container client toolkits is encapsulated. The toolkit includes container startup, stop, restart, creation, removal, single container status viewing, and server container status viewing. By controlling the container status, the purpose of controlling the start and stop of the algorithm can be achieved. The application's toolkit includes: Startup: First, configure the specified AI algorithm to start automatically in the container. When the container starts, the startup script of the algorithm can be automatically called. The startup command of the toolkit is used to find the image container containing the specified AI algorithm according to the number and start the container to realize the startup of the specified AI algorithm. Stop: Locate the specified container based on its number and stop the container to halt the specified AI algorithm; 2) Direct deployment: The specified AI algorithm is deployed directly without container files, and the algorithm cannot be started or stopped by controlling the container. Startup: Add a startup script to the specified AI algorithm and place it in the specified directory. Find the algorithm to be started by its number, switch to the corresponding directory, execute the startup script, and start the algorithm. Stop: Record the name of the specified AI algorithm. Using the name of the specified AI algorithm, execute operating system commands on the server to find the running process, obtain the process number, and then kill the process using operating system commands to stop the algorithm.

2. The method according to claim 1, characterized in that, The common operating environment includes multiple graphics processing frameworks and multiple deep learning frameworks; Before transferring the public dependency package to the public runtime environment of the host server, the method further includes: By comparing the differences in the public operating environments that different AI algorithms rely on, multiple sets of public operating environments are formed.

3. The method according to claim 1, characterized in that, The model is deployed in a multi-cloud deployment mode. The step of using corresponding processing methods to monitor the state and control the start / stop of the specified AI algorithm model includes: 1) Containerized deployment: By introducing a container client toolkit, a long-lived socket connection is established, creating a TCP network connection between the management end and the algorithm deployment end that will not be closed, enabling continuous and multiple transmissions of HTTP data. After the long connection is created, the management side can call the command tools of the container client toolkit to execute container start and stop commands. After the algorithm deployment side receives the commands, it executes the algorithm to start and stop on the server. 2) Direct deployment: The specified AI algorithm is directly deployed. Script files for starting and stopping the algorithm are added to the algorithm deployment server. During remote control, the management end sends an HTTP request. After the algorithm deployment server interface receives the HTTP request command, the runtime command is used to execute the local start and stop scripts, thus starting and stopping the algorithm.

4. A device for deploying and running an AI algorithm, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: When deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm; The specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm, thereby forming the specified AI algorithm model; Determine the operating environment corresponding to the specified AI algorithm model; When the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is subject to status monitoring and start / stop control. The AI ​​algorithm deployment methods include direct deployment in the server environment and containerized deployment. The step of deploying the specified AI algorithm to the host server using the deployment method corresponding to the specified AI algorithm to form the specified AI algorithm model includes: When the AI ​​algorithm is deployed directly in the server environment, it is deployed to the host server according to the preset directory structure of the specified AI algorithm to form the specified AI algorithm model. When the AI ​​algorithm is deployed in a containerized manner, the specified AI algorithm and its corresponding specified dependency packages are packaged into an image file using containerization technology, and the image file is deployed to the host server to form the specified AI algorithm model. The specified dependency packages include public dependency packages and algorithm dependency packages; The step of packaging the specified AI algorithm and its corresponding specified dependency packages into an image file using containerization technology includes: The public dependency package is transferred to the public runtime environment of the host server; The specified AI algorithm and its dependency packages are packaged into the image file using containerization technology. Also includes: If the common dependency package is incompatible with the algorithm dependency package, the incompatible dependency package is isolated from the algorithm by using the common dependency environment and by referencing Anaconda. The process of monitoring the state and controlling the start / stop of the specified AI algorithm model includes: Determine the model deployment method corresponding to the specified AI algorithm model; Based on the model deployment method corresponding to the specified AI algorithm model, the corresponding processing method is used to perform status monitoring and start / stop control on the specified AI algorithm model; The model is deployed using a single-server deployment method. The step of using corresponding processing methods to monitor the state and control the start / stop of the specified AI algorithm model includes: 1) Containerized deployment: A set of container client toolkits is encapsulated. The toolkit includes container startup, stop, restart, creation, removal, single container status viewing, and server container status viewing. By controlling the container status, the purpose of controlling the start and stop of the algorithm can be achieved. The application's toolkit includes: Startup: First, configure the specified AI algorithm to start automatically in the container. When the container starts, the startup script of the algorithm can be automatically called. The startup command of the toolkit is used to find the image container containing the specified AI algorithm according to the number and start the container to realize the startup of the specified AI algorithm. Stop: Locate the specified container based on its number and stop the container to halt the specified AI algorithm; 2) Direct deployment: The specified AI algorithm is deployed directly without container files, and the algorithm cannot be started or stopped by controlling the container. Startup: Add a startup script to the specified AI algorithm and place it in the specified directory. Find the algorithm to be started by its number, switch to the corresponding directory, execute the startup script, and start the algorithm. Stop: Record the name of the specified AI algorithm. Using the name of the specified AI algorithm, execute operating system commands on the server to find the running process, obtain the process number, and then kill the process using operating system commands to stop the algorithm.

5. A non-volatile computer storage medium, characterized in that, The computer-executable instructions are stored thereon and are configured as follows: When deploying a specified AI algorithm, determine the deployment method corresponding to the specified AI algorithm; The specified AI algorithm is deployed to the host server using the deployment method corresponding to the specified AI algorithm, thereby forming the specified AI algorithm model; Determine the operating environment corresponding to the specified AI algorithm model; When the specified AI algorithm model is running in the operating environment, the specified AI algorithm model is subject to status monitoring and start / stop control. The AI ​​algorithm deployment methods include direct deployment in the server environment and containerized deployment. The step of deploying the specified AI algorithm to the host server using the deployment method corresponding to the specified AI algorithm to form the specified AI algorithm model includes: When the AI ​​algorithm is deployed directly in the server environment, it is deployed to the host server according to the preset directory structure of the specified AI algorithm to form the specified AI algorithm model. When the AI ​​algorithm is deployed in a containerized manner, the specified AI algorithm and its corresponding specified dependency packages are packaged into an image file using containerization technology, and the image file is deployed to the host server to form the specified AI algorithm model. The specified dependency packages include public dependency packages and algorithm dependency packages; The step of packaging the specified AI algorithm and its corresponding specified dependency packages into an image file using containerization technology includes: The public dependency package is transferred to the public runtime environment of the host server; The specified AI algorithm and its dependency packages are packaged into the image file using containerization technology. Also includes: If the common dependency package is incompatible with the algorithm dependency package, the incompatible dependency package is isolated from the algorithm by using the common dependency environment and by referencing Anaconda. The process of monitoring the state and controlling the start / stop of the specified AI algorithm model includes: Determine the model deployment method corresponding to the specified AI algorithm model; Based on the model deployment method corresponding to the specified AI algorithm model, the corresponding processing method is used to perform status monitoring and start / stop control on the specified AI algorithm model; The model is deployed using a single-server deployment method. The step of using corresponding processing methods to monitor the state and control the start / stop of the specified AI algorithm model includes: 1) Containerized deployment: A set of container client toolkits is encapsulated. The toolkit includes container startup, stop, restart, creation, removal, single container status viewing, and server container status viewing. By controlling the container status, the purpose of controlling the start and stop of the algorithm can be achieved. The application's toolkit includes: Startup: First, configure the specified AI algorithm to start automatically in the container. When the container starts, the startup script of the algorithm can be automatically called. The startup command of the toolkit is used to find the image container containing the specified AI algorithm according to the number and start the container to realize the startup of the specified AI algorithm. Stop: Locate the specified container based on its number and stop the container to halt the specified AI algorithm; 2) Direct deployment: The specified AI algorithm is deployed directly without container files, and the algorithm cannot be started or stopped by controlling the container. Startup: Add a startup script to the specified AI algorithm and place it in the specified directory. Find the algorithm to be started by its number, switch to the corresponding directory, execute the startup script, and start the algorithm. Stop: Record the name of the specified AI algorithm. Using the name of the specified AI algorithm, execute operating system commands on the server to find the running process, obtain the process number, and then kill the process using operating system commands to stop the algorithm.

Images