WO2022078280A1

WO2022078280A1 - Artificial intelligence (ai) training method, system, and device

Info

Publication number: WO2022078280A1
Application number: PCT/CN2021/123021
Authority: WO
Inventors: 陈普
Original assignee: 华为云计算技术有限公司
Priority date: 2020-10-14
Filing date: 2021-10-11
Publication date: 2022-04-21
Also published as: CN114358302A

Abstract

Provided is an artificial intelligence (AI) training method, applicable to a guidance system (106). After a user of a cloud training system (120) triggers training code to run on a local device (100), said method comprises: according to the training code run on the local device (100), obtaining information about a training task, the training code being used for training an AI model, the AI model being developed and obtained by the user on the basis of an AI framework (104) installed on the local device (100); uploading training task information to a cloud training system (120), and notifying the cloud training system (120) to execute the training task corresponding to the training task information. The method allows a developer performing AI model development and training on the local device (100) to be no longer constrained by the computing power of the local device (100).

Description

A method, system and device for artificial intelligence AI training

This application requires a Chinese patent application with the application number 202011096163.6 and the application name "AI model training method and device" to be submitted to the China Intellectual Property Office on October 14, 2020, and China Intellectual Property Office on December 30, 2020 Bureau, the application number is 202011626123.8, and the application title is "A method, system and device for artificial intelligence AI training". The priority of the Chinese patent application, the entire content of which is incorporated in this application by reference.

technical field

The present application relates to the technical field of artificial intelligence (AI), and in particular, to a method, system and device for AI training.

Background technique

With the development of AI technology, AI is applied to all walks of life. Typical AI technologies include deep learning technology, machine learning technology, etc. The basic idea is to design an AI model and train the AI model based on training data, so that the trained AI model has certain functions, such as object detection. , target recognition function, etc. An AI model is an algorithm implemented using AI technology, such as a deep learning model based on deep learning technology.

In order to facilitate the development and training of AI models, various AI frameworks have emerged in the industry. Developers are usually accustomed to installing AI frameworks on local devices and developing and training AI models based on the AI framework. Because the training of AI models needs to rely on the support of strong computing power. The AI model developed by the developer on the local device based on the AI framework is usually limited by the computing power of the local device.

SUMMARY OF THE INVENTION

This application provides an artificial intelligence AI training method. The method uploads the training task information to the cloud training system when the training code in the local device is running, and uses the resources of the cloud to realize the training of the AI model, so that developers can The development of AI models executed on the local device is no longer limited by the computing power of the local device.

In the first aspect, the present application provides an AI training method, which is applied to a guidance system. After a user of the cloud training system triggers the running of the training code on the local device, the guidance system performs the following steps: Training code to obtain training task information, where the training code is used to train an AI model, and the AI model is developed and obtained by the user based on the AI framework installed on the local device; upload the training task information to the cloud training system, and notify The cloud training system executes the training task corresponding to the training task information.

Through the above method, when the user of the cloud training system needs to train the locally built AI model, he can use the editor to write the corresponding training code on the local device. After the user runs the training code locally, the guidance system in this application is It can perform the operations of acquiring training task information and uploading training task information to the cloud training system. This method enables to use the resources on the cloud to train the AI model while maintaining the user's habit of developing and training models locally, which solves the problem of insufficient resources for training on local devices and brings convenience to users. .

In a possible implementation of the first aspect, the aforementioned training task information includes the information obtained from the training code by using the obtaining component in the AI framework, and the information obtained from the local device according to the information in the training code. The acquisition component in the AI framework can obtain information by reading the training code and intercepting the calling API in the training code during the running of the training code. Obtaining the information in the training code through the acquisition component can make the guidance system faster to acquire to the training task information.

In a possible implementation of the first aspect, the aforementioned training task information includes one or more of the following data: training parameters in the training code, the AI model to be trained, and the training code used for training the AI model The training program logic, the training environment information of the local device, and the cloud training access information used to connect with the cloud training system. The uploaded training task information enables the cloud training system to smoothly prepare the cloud training environment and execute the training task in the cloud environment.

In a possible implementation of the first aspect, the training environment information of the local device includes: version information of the AI framework, and version information of the programming language of the training code.

Since the AI model and training code are developed by the user based on the AI framework and programming language installed in the local device, upload the version information of the AI framework and the programming language of the training code in the local training environment information to the cloud. The training system enables the cloud training system to prepare in advance the version of the AI framework and programming language that matches the AI model and training code to be trained when preparing the cloud environment.

In a possible implementation of the first aspect, the method for guiding the execution of the system further includes: receiving a training data acquisition request sent by a cloud training system during the execution of the training task; acquiring the training data and sending the training data data to the cloud training system.

The above method makes it unnecessary for the guidance system to upload training data or upload all training data to the cloud training system before the cloud training system starts to execute the training task, avoids too long waiting for transmission before the cloud training task starts, and improves the user experience.

In a possible implementation of the first aspect, before notifying the cloud training system to execute the training task corresponding to the training task information, the method further includes: receiving an environment preparation success response returned by the cloud training system.

In a possible implementation of the first aspect, the method further includes: receiving a trained AI model returned by the cloud training system. After receiving the trained AI model, the guidance system can store the trained AI model in the local device, and can also provide prompts to the user, such as prompting the user where the trained AI model is stored on the local device, so that the user can The trained AI model can be obtained more conveniently, which facilitates the user's subsequent application of the trained AI model.

In a possible implementation of the first aspect, the guidance system may be obtained from a cloud training system and installed in the local device. For example, the cloud training system can provide an interface for downloading the bootstrap system.

In a second aspect, the present application also provides an AI training method, which is applied to a cloud training system, including: acquiring training task information sent by the guidance system after a user triggers running a training code on a local device, where the training task information includes local Training environment information of the device; perform preparation of a cloud training environment according to the training environment information; perform training tasks corresponding to the training task information based on the cloud training environment.

In a possible implementation of the second aspect, the training environment information of the local device includes: version information of the AI framework on which the AI model to be trained depends and the version of the programming language used by the training code used to train the AI model information.

In a possible implementation of the second aspect, performing the preparation of the cloud training environment according to the training environment information includes: setting the cloud training environment to execute training tasks according to the version information of the AI framework and the version information of the programming language of the training code AI frameworks and programming languages used.

In a possible implementation of the second aspect, the training task information further includes: training parameters in the training code, an AI model, and training program logic in the training code for training the AI model; based on the cloud training environment Performing the training task corresponding to the training task information includes: performing the training of the AI model in the prepared cloud training environment according to the training parameters and the training program logic.

In a possible implementation of the second aspect, the training task information further includes cloud training access information, and before performing the preparation of the cloud training environment according to the training environment information, the method further includes: according to the cloud training access information The input information performs authentication and/or charging query on the training task corresponding to the training task information.

For the beneficial effects of the features in the second aspect and possible implementation manners of the second aspect, reference may be made to the beneficial effects of the corresponding features in the foregoing first aspect, which will not be repeated here.

In a third aspect, the present application further provides a guidance system, comprising: an acquisition module for, after a user of the cloud training system triggers the running of the training code on the local device, according to the training code running on the local device, Obtain training task information, wherein the training code is used to train an AI model, and the AI model is developed and obtained by the user based on the AI framework installed on the local device; a sending module is used to send the training task The information is uploaded to the cloud training system, and the cloud training system is notified to execute the training task corresponding to the training task information.

In a possible implementation of the third aspect, the training task information includes information obtained from the training code by using an obtaining component in the AI framework, and information obtained from the local device according to the information in the training code information.

In a possible implementation of the third aspect, the training task information includes one or more of the following data: training parameters in the training code, the AI model, and data used in the training code for all The training program logic for training the AI model, the training environment information of the local device, and the cloud training access information for connecting with the cloud training system.

In a possible implementation of the third aspect, the training environment information of the local device includes: version information of the AI framework, and version information of the programming language of the training code.

In a possible implementation of the third aspect, the system further includes a receiving unit, the receiving unit is configured to receive a training data acquisition request sent by the cloud training system during the execution of the training task; the The obtaining unit is further configured to obtain the training data; the sending unit is further configured to send the training data to the cloud training system.

In a possible implementation of the third aspect, the system further includes a receiving unit, configured to receive, before the sending unit notifies the cloud training system to execute the training task corresponding to the training task information The environment returned by the cloud training system is ready to respond successfully.

In a possible implementation of the third aspect, the system further includes a receiving unit, where the receiving unit is configured to receive the trained AI model returned by the cloud training system.

In a possible implementation of the third aspect, the guidance system is obtained from the cloud training system and installed in the local device.

In a fourth aspect, the present application further provides a cloud training system, including: an environment preparation unit for acquiring training task information sent by the guidance system after a user triggers running a training code on a local device, where the training task information includes the local device the training environment information; perform the preparation of the cloud training environment according to the training environment information; a training task execution unit is configured to execute the training task corresponding to the training task information based on the cloud training environment.

In a possible implementation of the fourth aspect, the training environment information of the local device includes: version information of the AI framework on which the AI model to be trained depends, and version information of the programming language used to train the training code of the AI model .

In a possible implementation of the fourth aspect, the environment preparation unit is specifically configured to perform training tasks in the cloud training environment according to version information of the AI framework and version information of the programming language of the training code AI frameworks and programming languages used.

In a possible implementation of the fourth aspect, the training task information further includes: training parameters in the training code, an AI model, and training program logic in the training code for training the AI model; the training task The execution unit is specifically configured to execute the training of the AI model in the prepared cloud training environment according to the training parameters and the training program logic.

In a possible implementation of the fourth aspect, the training task information further includes cloud training access information, and the environment preparation unit is further configured to: correspond to the training task information according to the cloud training access information authentication and/or billing queries for training tasks.

In a fifth aspect, the present application also provides a computing device, comprising a processor and a memory, the memory storing computer instructions, the processor executing the computer instructions, so that the computing device performs the aforementioned first aspect or the first Aspects may implement the methods described in, or perform the methods described in the preceding second aspect or possible implementations of the second aspect.

In a sixth aspect, the present application further provides a computer-readable storage medium, where the computer-readable storage medium stores computer program codes, and when the computer program codes are executed by a computing device, the computing device executes the aforementioned first aspect or a method in a possible implementation of the first aspect, or perform the aforementioned second aspect or a method in a possible implementation of the second aspect. The computer-readable storage medium includes, but is not limited to, volatile memory, such as random access memory, non-volatile memory, such as flash memory, hard disk (English: hard disk drive, abbreviation: HDD), solid state disk (English: solid state drive, abbreviation: SSD).

In a seventh aspect, the present application further provides a computer program product, the computer program product comprising computer program code, when the computer program code is executed by a computing device, the computing device may perform the aforementioned first aspect or the first aspect. Implement the method provided in, or perform the method provided in the aforementioned second aspect or possible implementations of the second aspect. The computer program product may be a software installation package, and when the method provided in the foregoing first aspect or possible implementation of the first aspect needs to be used, or the method provided in the foregoing second aspect or possible implementation of the second aspect needs to be used, The computer program product can be downloaded and executed on a computing device.

In an eighth aspect, the present application further provides an artificial intelligence AI system, including the guidance system described in the foregoing third aspect and possible implementation of the third aspect, and the foregoing fourth aspect and possible implementation of the fourth aspect. Cloud training system.

Description of drawings

1 is a schematic diagram of a system architecture provided by an embodiment of the present application;

FIG. 2 is a schematic structural diagram of a guidance system 106 and a cloud training system 120 provided by an embodiment of the present application;

3 is a schematic flowchart of an AI training method provided by an embodiment of the present application;

4 is a schematic flowchart of an authentication and charging query performed by a cloud training system according to training task information according to an embodiment of the present application;

5 is a schematic flowchart of a cloud training system performing a training task according to an embodiment of the present application;

FIG. 6 is a schematic structural diagram of a computing device 300 according to an embodiment of the present application;

FIG. 7 is a schematic structural diagram of a computing device 400 according to an embodiment of the present application.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

In order to describe the technical solutions of the embodiments of the present application more clearly, the technical terms involved in the present application are explained below:

AI training: It is a process of using training data to update the parameters in the AI model, so that the AI model can learn the features and laws of the training data, so that the AI model can realize specific application functions.

In the AI training process, a large amount of computing work needs to be performed, which requires more computing resources (ie: computing power) and storage resources.

Training code: It is the code that drives the training of the AI model. The training code defines the training parameters and training program logic in the AI training process. It can also include the name or address information of the AI model to be trained, the name of the training data set or address information, etc. The training code also includes cloud training access information for connecting with the cloud training system in the cloud.

AI framework: It is a library (or tool) used for the development, training and testing of AI models. The AI framework includes a variety of packaged code components. Each code component is a set of pre-built and optimized functions. This application The code components in the AI framework are also called functional components. Developers can use the AI framework to develop and train AI models that meet application goals more quickly, without having to understand the underlying algorithm implementation for AI model construction and training in detail. At present, there are many AI frameworks in the industry. Using one or more AI frameworks to develop and train AI models brings great convenience to developers. Typical AI frameworks include deep learning frameworks. There are various types of deep learning frameworks in the industry. Different deep learning frameworks can provide different functions, but they all aim to provide developers with deep learning models. A library (or tool) for testing.

AI technology is developing rapidly and is gradually applied to more and more complex scenarios. At present, the structure of AI models is becoming more and more complex, and many complex scenarios have higher and higher performance requirements for AI models. Therefore, it is often necessary to use a large amount of training data to train AI models with complex structures. However, the AI development mode that developers are used to is: install the AI framework on the local device, and then use the local editor, for example: use the local application integrated development environment (integrated development environment, IDE), call the AI framework to build the AI model, and write the training code . Then use local resources to perform training to obtain the trained AI model. Although some AI frameworks currently support distributed training of AI models, developers still often encounter the problem that locally developed AI models cannot have sufficient training resources locally.

It should be understood that the local or local device in this application refers to the device used by the developer to develop the AI model (for example, the server or virtual machine used by the developer), and/or belongs to the same device as the device used by the developer to develop the AI model The owner's other devices or clusters of devices. Usually, each local device is in a relatively close physical environment, such as a computer room. It is worth noting that, in the case of using the client in the local terminal to use a remote virtual machine (for example, a desktop cloud) for AI development, the local terminal and the remote virtual machine can also be called a local device together.

In the cloud computing model, cloud service providers use a large number of basic resources (such as computing resources, storage resources, communication resources) to build a cloud environment, and cloud service providers can provide cloud tenants with various basic resources, platforms or application capabilities. This enables cloud tenants to carry out their own business.

Based on the above background, the present application provides an AI training method, which can use the basic resources on the cloud to train the locally constructed AI model for the developer while the developer maintains the habit of local AI development. This method can solve the problem of The problem of insufficient local training resources, and it is more convenient and quick to realize the training of AI models for developers. It should be understood that the developer described in this application may also be referred to as a user, which means a person who uses the AI training method of this application. Since, before using the AI training method of the present application, the user needs to register on the cloud platform and purchase the cloud training service, therefore, the user (or developer) in the present application is also the user of the cloud training system.

Before describing the specific solution of the present application, the system architecture of the application of the present application is described first. FIG. 1 is a schematic diagram of an exemplary system architecture provided by the present application. As shown in FIG. 1 , the system architecture of the present application includes a local device 100 and a cloud training system 120 in a cloud environment. The local device 100 may be a server, a virtual machine, or a server cluster or a virtual machine cluster owned by the developer, and the developer uses the local device 100 for AI development. The cloud training system 120 in the cloud environment may be a background system corresponding to a cloud training service provided by a cloud service provider, and the cloud training system 120 may use the basic resources in the cloud environment to perform cloud training on AI models developed locally by developers .

An editor 102 and an AI framework 104 are installed in the local device 100 , and a guidance system 106 is also installed in the local device 100 . The AI framework 104 is used as a library for AI development, and is used to provide functional components for developers in the process of building and training AI models. The editor 102 is used by the developer to perform editing of the training code, and the editor 102 may also be used to compile the training code. The developer can call the components in the AI framework 104 through the editor 102 in the local device 100 to construct and train the AI model. To train the built AI model, developers usually develop training code in a local editor, write the training parameters required for training, and the training program logic used during training. The training program logic in the training code instructs how to use the training data used during training and train the AI model to be trained according to the set training parameters. When the training code edited by the developer is run, the training of the AI model is started.

In order to enable the AI model built by the developer to be trained using the resources on the cloud, a guidance system 106 is also installed on the above-mentioned local device. The guidance system 106 is configured to acquire training task information when the training code is running, and upload the training task information to the cloud training system 120 in the cloud environment, so that the cloud training system 120 executes the training task corresponding to the training task information. The training task information may include training parameters required to perform training on the AI model, training program logic, training environment information, cloud training access information, and the AI model to be trained. It should be understood that the guidance system 106 described above may be a stand-alone software program installed in the local device 100 or a tool coupled with the AI framework 104 . FIG. 1 of the present application takes the guidance system 106 as an independent software program as an example.

The cloud training system 120 in the cloud environment is configured to receive the training task information sent by the guidance system 106, and use the training task information to prepare the cloud training environment, where the set cloud training environment matches the AI model and the training program logic. The cloud training system 120 is further configured to receive the training notification sent by the guidance system 106, and train the developer's AI model in the set cloud training environment. After the training task is completed, the cloud training system may also return a successful training response and/or return the trained AI model to the guidance system 106 .

As can be seen from the above, the present application mainly utilizes the guidance system 106 in the local device 100 to interact with the cloud training system 120 in the cloud environment, so as to realize the assignment of training tasks corresponding to the training codes developed in the local device 100 to the cloud training system 120 for execution.

The structure and functional division of the guidance system 106 and the cloud training system 120 are described below with reference to FIG. 2 .

Since the structure of the AI model developed by the developer and the training code developed by the developer depend on the functional components in the AI framework, the AI framework 104 needs to be called when the training code runs. In an embodiment of the present application, an acquisition component 1042 is embedded in the AI framework 104 used for the construction and training of the AI model, and the acquisition component 1042 may be a tool, a plug-in or a patch in the AI framework 104.

The obtaining component 1042 is configured to communicate with the guidance system 106 and transmit the obtained initial information of the training task to the guidance system 106 . Specifically, the obtaining component 1042 is configured to obtain the initial information of the training task from the training code when the training code is running. In the embodiment of the present application, the information acquired by the acquisition component 1042 in the AI framework is collectively referred to as the initial information of the training task. The initial information of the training task may include two types of information, one is the model that needs to be used when executing the training task Or the identification information or address information of the data, such as: the address information of the AI model to be trained, the address information of the training data, etc.; the other is the information that can be used directly when performing training tasks, such as: training parameters, training program logic, etc. . The acquisition component 1042 acquires the initial information of the training task and sends it to the guidance system 106 .

The guidance system 106 may include a receiving unit 1061 , an obtaining unit 1062 and a sending unit 1063 . It should be understood that the unit division of the guidance system 106 described above is only an example, and does not constitute a limitation on the guidance system 106 of the present application.

The receiving unit 1062 is configured to receive the initial information of the training task sent by the obtaining component 1042 . The receiving unit 1062 may also receive instructions or information sent by the cloud training system 120 .

Since the initial information of the training task includes identification information or address information of some models or data that need to be used when executing the training task, the obtaining unit 1064 in the guidance system 106 can be configured to obtain information from the local device according to the initial information of the training task received by the receiving unit 1062 100 to obtain training task information. For example, after the receiving unit 1062 receives the address information of the AI model obtained by the obtaining component 1042 of the AI framework, the obtaining unit 1064 obtains the AI model to be trained according to the address information of the AI model. It should be understood that, for the initial information of this type of training tasks that can be directly used when performing training tasks, the initial information of these training tasks obtained by the obtaining component 1042 and the content of the corresponding training task information sent by the guidance system 106 to the cloud training system 120 same. In other words, after the receiving unit 1062 in the guidance system 106 receives this type of information from the acquisition component 1042 , without further acquisition by the acquisition unit 1064 , this type of information can be used as the training information to be sent by the guidance system to the cloud training system 120 . task information. In the embodiment of the present application, the data and information sent by the guidance system 106 to the cloud training system 120 for performing the training task are collectively referred to as training task information.

The sending unit 1066 is configured to send the training task information to the cloud training system 120 . The sending unit 1066 is further configured to send training notifications and/or training data to the cloud training system 120 .

As shown in FIG. 2 , the cloud training system 120 includes an environment preparation unit 122 and a training task execution unit 124 .

The environment preparation unit 122 is configured to receive the training task information sent by the guidance system 1066, and prepare the cloud training environment according to the training task information, for example, setting the resources to be used when executing the training task on the cloud and the libraries that the training depends on. Optionally, the environment preparation unit 122 may also be configured to return a response message indicating that the environment preparation is complete to the guidance system 106 . Optionally, the environment preparation unit 122 may also be configured to send partial training task information and/or training task execution instructions to the training task execution unit 124 .

The training task execution unit 124 may execute the training task according to the training task execution instruction sent by the environment preparation unit 122 or the guidance system 106 . When executing the training task, the training task execution unit 124 mainly executes the training program logic in the training task information to train the AI model to be trained. The training task execution unit 124 is further configured to return a successful training response to the guidance unit and/or return the trained AI model.

Before the developer needs to use the cloud training system 120 described in this application to realize the training of the locally developed AI model by using the resources on the cloud, the developer can purchase the cloud training service through the cloud platform of the cloud service provider and configure the cloud training. The total amount of resources available to the system 120 for training, setting authentication key information, and the like. In some embodiments, the acquisition component 1042 in the guidance system 106 and the AI framework 104 may be a software program or tool developed by a cloud service provider in conjunction with a cloud service that provides cloud training. After the developer purchases and configures the cloud training service, the above-mentioned guidance system 106 and acquisition component 1042 can be installed on the local device. Thus, the developer can run the training code locally to start the training of the AI model performed by the cloud training system 120 .

2 described above is only one possible embodiment of the present application. In other embodiments, the function of the obtaining component 1042 in the AI framework 104 may also be provided by the guidance system 106 , that is, the aforementioned action of the obtaining component 1042 obtaining the initial information of the training task according to the training code may be performed by the guidance system 106 . In this case, the guidance system 106 may include four functional units, namely: an acquisition unit that performs the actions of the acquisition component 1042 , and the aforementioned receiving unit 1062 , acquiring unit 1064 , and transmitting unit 1066 . The foregoing division of the functional units of the guidance system 106 is only an example, and there may be different division manners, which will not be repeated here.

The following embodiments of the present application are described by taking the guidance system 106 not including the function of the acquisition component 1042 in the AI framework 104 as an example for description. It should be understood that the content of the following embodiments can also be adaptively applied to scenarios where the guidance system 106 includes the functions of the acquisition component 1042 in the AI framework 104 .

FIG. 3 is a schematic flowchart of an AI training method provided by an embodiment of the present application. The specific implementation of the AI training method of the present application will be described in detail below with reference to FIG. 3 . The AI training method may be performed collaboratively by the aforementioned guidance system 106 , the cloud training system 120 , and the acquisition component 1042 in the AI framework 104 .

S201: The developer develops and runs the training code.

Specifically, developers can develop the code for training the built AI model in the editor. The editor can be various IDEs in the industry. The training code can include the training parameters set by the developer, such as: learning rate, batch (batch) The processing value, batch size, etc., can also include the name of the training dataset, the name of the AI model to be trained, and the training program logic. The training program logic may include the training program logic written by the developer, or the training program logic in the AI framework invoked by the developer. The training program logic may include algorithm logic such as loss function and optimizer. It should be understood that the AI model built by the developer depends on the AI framework, and the execution of the training program logic in the training code also depends on the AI framework.

Since the training task of the AI model in this application is performed by the cloud, when developing the training code, the developer also needs to set the cloud training mode in the training code, for example, use the code to indicate that the training mode is the cloud training mode. Developers can also write cloud training access information in the training code, such as cloud access address information, cloud authentication information, account information, etc.

After the training code developed by the developer is compiled by the compiler, it can be started and run on the local device. After starting and running the training code, since the cloud training mode is set in the training code, in some embodiments, the cloud training mode in the training code can trigger the acquisition component in the AI framework to perform the acquisition operation for the initial information of the training task.

S202: The acquisition component in the AI framework acquires the initial information of the training task corresponding to the training code according to the training code, and sends the initial information of the training task to the guidance system.

The initial information of the training task that the acquisition component can acquire from the training code includes: address information of the AI model to be trained. When acquiring the address information of the AI model to be trained, the acquisition component can intercept the application program interface (API) when the model is loaded in the training code, thereby obtaining the path information of the AI model to be trained in the local device .

The initial information of the training task that the acquisition component can acquire from the training code further includes: address information of the training data. When the acquisition component acquires the address information of the training data, it can also intercept the API when the training data is loaded in the training code, and then obtain the path information of the training data in the local device.

The initial information of the training task that the acquisition component can acquire from the training code also includes: some training program logic, cloud training access information, training parameters, training environment information, and the like. The cloud training access information may include: cloud access address information, cloud authentication information, and account information. Training parameters can include: learning rate used during training, batch value, batch size, etc. The training environment information may include: version information of the AI framework, programming language version information of the training code, programming language version or some plug-in or library information of the AI framework version, resource specifications and quantities used for training, and the like. The above training environment information can be divided into two categories. One type represents the training environment information of the local device, which represents the environment information when the local device builds the AI model and develops the training code, including: version information of the AI framework, programming language version information of the training code, programming language version or AI framework Version of some plugins or library information, etc. The other type represents the training environment information set in the local device, which represents the environment information when AI model training is performed, including: resource specifications and quantities used to perform training, etc. This type of information is usually set by the user in the training code.

S203: Guide the system to upload training task information to the cloud training system.

After the guidance system receives the initial information of the training task sent by the acquisition component in the AI framework, the guidance system can obtain the training task information to be sent according to the initial information of the training task, and upload the training task information to the cloud training system. For example, read the AI model to be trained from the local device according to the address information of the AI model to be trained, and upload it to the cloud training system. Optionally, the guidance system can also actively detect and acquire some training task information, and this action can be performed by an acquisition unit in the guidance system. For example, when the acquisition component in the AI framework cannot acquire some training environment information (such as the programming language version of the training code), the guidance system can acquire the training environment information by actively detecting the training environment in the local device. Therefore, the training task information includes the information obtained from the training code by using the obtaining component in the AI framework, and the information obtained from the local device according to the information in the training code.

Before the guidance system uploads the acquired training task information to the cloud training system, it can establish a connection with the cloud training system according to the cloud training access information in the training task information, and establishing a connection may include operations such as authentication and billing query.

Specifically, the schematic flowchart of the connection between the guidance system and the cloud training system can be shown in FIG. 4 , which specifically includes the following steps S2031-S2036:

S2031: Guide the system to send an upload request to the cloud training system.

The cloud training access information included in the above upload request includes cloud access information, account information and authentication information of the local device. The cloud access information may be address information of the cloud training system, and an upload request may be sent to the cloud training system according to the address information of the cloud training system. The account information and authentication information may be the information registered and acquired by the developer when purchasing the cloud training service in the cloud platform before using the solution of the present application. For example, the account information may be the user name of the developer on the cloud platform, and the authentication information may be the key obtained from the cloud platform and corresponding to the cloud training service.

It is worth noting that, in some cases, the upload request of the bootstrap system may only include the above-mentioned cloud access information for access, authentication and fee query, account information and authentication information of the local device, and in other cases Next, the upload request may also include part or all of the foregoing training task information. If the upload request only includes cloud access information, account information and authentication information, the guidance system can upload other training task information to the cloud training system after receiving the prompt that the authentication and billing query is passed.

S2032: The cloud training system receives the upload request, and sends the account information and authentication information to the cloud authentication center.

S2033: The cloud authentication center authenticates the training task requested by the upload request according to the acquired account information and authentication information, and returns an authentication result.

The specific authentication mode may adopt any feasible authentication mode in the industry, which is not limited in this application.

After the authentication of the cloud authentication center is completed, the authentication result is returned to the cloud training system.

S2034: The cloud training system sends the account information to the cloud billing center.

S2035: The cloud billing center confirms the fee information of the account corresponding to the account information according to the account information, and returns the fee information to the cloud training system.

The present application does not limit the execution order of the above steps S2034-S2035 and steps S2032-S2033. The execution of the above steps S2034-S2035 may also be optional.

S2036: The cloud training system returns a response that the authentication and billing query is passed to the guidance system, and receives training task information.

When the authentication is passed and the account fee corresponding to the account information is greater than or equal to the preset threshold, the cloud training system returns a response to the authentication and billing query to the guidance system, and receives other training task information uploaded by the guidance system. The training task information includes the aforementioned training task information except the cloud training access information used for cloud access, authentication, and fee query, such as: training parameters, training program logic, AI model to be trained, training environment information, etc. .

It should be understood that the above steps are optional, and when the above upload request includes other training task information, the cloud training system may directly receive the uploaded training task information without returning a response.

In the case that the authentication fails or the pre-stored fee of the account corresponding to the account information is less than the preset threshold, the cloud training system can return the upload request failure response to the guidance system, and can also return the request failure reason, such as: authentication failed and/ or insufficient pre-storage fees.

After performing the above steps S2031-S2036, the guidance system can successfully send the training task information to the cloud training system.

It should be understood that the above steps S202-S203 are described based on one of the aforementioned embodiments of the present application (that is, the embodiment in which the guidance system and the acquisition component in the AI framework cooperate to acquire the training task information). In another embodiment, the above-mentioned guidance system may include the function of the acquisition component in the above-mentioned AI framework, and the above-mentioned steps S202 and S203 are both performed by the guidance system.

After the cloud training system receives the training task information, it can prepare the cloud training environment on the cloud and execute the training tasks. The following is a detailed description of step S204:

S204: The cloud training system executes a training task corresponding to the training task information according to the received training task information.

As shown in Figure 5, specifically, step S204 can be divided into the following steps:

S2041: The cloud training system prepares a cloud training environment according to the received training task information.

Before the cloud training system performs training tasks, it needs to prepare the cloud training environment so that the cloud training environment matches the local training code and the local AI model to be trained. Specifically, the cloud training system needs to prepare a cloud training environment according to the training environment information in the training task information. The training environment information may include: version information of the AI framework, version information of the programming language of the training code, programming language version or AI framework version Some plugin or library information, resource specifications used to perform training.

The cloud training system needs to ensure that the AI framework and programming language versions that cloud training depends on are ready in the cloud environment according to the version of the AI framework and the programming language of the training code. Usually, the cloud environment includes versions of various mainstream AI frameworks and programming languages. Therefore, usually the cloud training system only needs to detect and confirm when preparing the cloud training environment, without temporarily installing these versions.

The cloud training system also needs to ensure that these plug-ins or libraries have been installed in the cloud environment according to the programming language version or some plug-in or library information of the AI framework version. Usually, the cloud environment will also update and download the plug-ins of mainstream AI frameworks and programming software in time. and required libraries. If the cloud training system finds that the cloud environment does not have the plug-ins and libraries required to perform training tasks installed in the cloud environment preparation stage, it can be downloaded and installed in time.

The cloud training system also needs to prepare corresponding training resources on the cloud according to the information on the resource specification for performing training included in the training task information. For example, according to the required resource specification information, start the relevant virtual machines and containers in the cloud, and mount the corresponding hardware resources, such as graphics processing units (GPUs) or AI training chips.

After the cloud training environment is prepared, the cloud training system can perform the following steps:

S2042: The cloud training system returns an environment preparation success response to the guidance system.

S2043: The guidance system sends a training notification to the cloud training system.

It should be noted that, in other embodiments, the above steps S2042 and S2043 may not be performed. For example, the guidance system can notify the cloud training system to perform the training task when uploading the training task information, and the cloud training system can start to execute the training task after performing the aforementioned step S2041, and the aforementioned steps S2042 and S2043 are omitted.

S2044: The cloud training system executes the training task corresponding to the training task information in the cloud training environment.

The cloud training system can start the training container with the relevant resources ready to execute the training task. Specifically, when the training task is executed, the functional components in the corresponding AI framework on the cloud are called according to the logic of the training program. Input the training data into the AI model to be trained, use each component in the model to calculate the training data based on the training resources, and update the values of the parameters in the model according to some training parameters and training program logic. When the training reaches the training stop condition, the training of the model is stopped, and the trained AI model is obtained.

Due to the large amount of training data used to train the AI model (for example: tens of thousands of pictures, or tens of thousands of videos), when uploading training task information, the guidance system reads the training data set in the local device, and once Uploading to the cloud training system indiscriminately will result in high transmission delay, and the environment preparation time of the cloud training system will be longer, which will affect the user experience. In some embodiments, the cloud training system may send a training data acquisition request to the guiding device at least once in a process of performing training.

That is, optionally, the following steps may also be performed during the execution of the training task:

S2045: The cloud training system sends a training data acquisition request to the guidance system;

S2046: Guide the system to read the training data from the local device, and send the training data to the cloud training system.

In other embodiments, the training data used for training the AI model may also be pre-saved by the user in a place that can be read by the cloud training system, such as cloud storage.

It is worth noting that during the above process of executing the training task, the cloud billing center can also continuously charge according to the duration, resource specifications, resource quantity, etc. of the resources used during training.

After the above step S204, the cloud training system can successfully train the AI model and obtain the trained AI model.

S205: The cloud training system returns the trained AI model to the guidance system.

It is worth noting that the above step S205 is only a step performed in one case, and in other cases, the cloud training system may not return the trained AI model to the guidance system after performing the training task. For example, the cloud training system can return a successful training response to the guidance system, or return the address information of the trained AI model stored in the cloud environment to the guidance system. What the cloud training system returns to the guidance system after the training is completed can be determined by the developer through preset settings. Optionally, the cloud training system may also return the billing bill to the guidance system.

Through the above steps S201-S205, the developer can write and run the training code locally, that is, the AI model can be trained using the resources of the cloud environment. It avoids the problem that the resources required for local training are insufficient to support the training of AI models. The above method greatly facilitates developers, and developers do not need to change the habit of locally building AI models and developing training codes when faced with insufficient local resources. The solution of the present application also does not require the developer to perform complex configuration and adaptation, and the cloud training can be quickly realized through the cooperation of the guidance system and the cloud training system.

Embodiments of the present application further provide the guidance system 106 shown in FIG. 2 . In some embodiments, the guidance system 106 is specifically configured to perform the steps performed by the guidance system shown in the foregoing FIGS. 3 to 5 , and the functions of the guidance system 106 The functions of the units are the same as those described above for FIG. 2 , which will not be repeated here. In other embodiments, the guidance system 106 may also be specifically configured to perform the functions of the acquisition components in the guidance system and the AI framework shown in the foregoing FIGS. 3-5 .

The embodiment of the present application also provides the cloud training system 120 shown in FIG. 2 . The cloud training system 120 can specifically be used to perform the steps performed by the cloud training system shown in FIG. 3 to FIG. 5 . The functions are the same as those described above for FIG. 2 , which will not be repeated here.

This embodiment of the present application further provides a computing device 300 as shown in FIG. 6 , and the computing device 300 may be the aforementioned local device. Computing device 300 includes memory 301 , processor 302 , communication interface 303 , and bus 304 . The memory 301 , the processor 302 , and the communication interface 303 are connected to each other through the bus 304 for communication. It should be understood that the present application does not limit the number of processors and memories in the computing device 300 . Computing device 300 may also represent a cluster of devices composed of multiple servers or virtual machines.

The memory 301 may be a read only memory (Read Only Memory, ROM), a static storage device, a dynamic storage device, or a random access memory (Random Access Memory, RAM). The memory 301 may store computer instructions, and when the computer instructions stored in the memory 301 are executed by the processor 302, the processor 302 and the communication interface 303 perform part or all of the AI training performed by the guidance system described in the aforementioned Figures 3-5. method. That is, the computer instructions in the aforementioned guidance system 106 may be stored in the memory 301 . The AI module to be trained and training data can also be stored in the memory 301 .

The processor 302 may adopt a general-purpose central processing unit (Central Processing Unit, CPU), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a graphics processor (graphics processing unit, GPU) or any combination thereof. The processor 302 may include one or more chips, and the processor 302 may include an AI accelerator, such as a neural processing unit (NPU).

The communication interface 303 uses a transceiver module such as, but not limited to, a transceiver to enable communication between the computing device 300 and other devices or a communication network. For example, the response of successful training or the AI model after training can be obtained through the communication interface 303 .

Bus 304 may include pathways for communicating information between various components of computing device 300 (eg, memory 301, processor 302, communication interface 303).

The embodiment of the present application further provides a computing device 400 as shown in FIG. 7 , the computing device 400 may be a cloud server or a cloud server cluster provided by a cloud service provider, or a virtual machine or a virtual machine cluster. Computing device 400 includes memory 401 , processor 402 , communication interface 403 , and bus 404 . The possible hardware structures of the memory 401 , the processor 402 , the communication interface 403 and the bus 404 and the relationship between each part may be the same as or similar to the corresponding parts in the aforementioned computing device 300 , and will not be repeated here. The memory 401 in the computing device 400 may store the computer instructions included in the environment preparation unit 122 and the training task execution unit 124 in the aforementioned cloud training system 120. When the computer instructions stored in the memory 401 are executed by the processor 402, the processor 402 and the communication interface 403 perform part or all of the AI training method performed by the cloud training system as described in the aforementioned FIGS. 3-5 .

The descriptions of the processes corresponding to the above figures have their own emphasis, and for parts that are not described in detail in a certain process, please refer to the relevant descriptions of other processes.

In the above-mentioned embodiments, it may be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented in software, it can be implemented in whole or in part in the form of a computer program product. The computer program product implementing the above-mentioned AI training method includes one or more computer instructions, and when these computer program instructions are loaded and executed on the computer, the whole or part of the AI training method flow described in the aforementioned FIGS. 3-5 of the present application is executed. .

The computer may be a general purpose computer, special purpose computer, computer network, or other programmable device. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be downloaded from a website site, computer, server, or data center Transmission to another website site, computer, server, or data center by wire (eg, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (eg, infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes an integration of one or more available media. The available media may be magnetic media (eg, floppy disk, hard disk, magnetic tape), optical media (eg, digital versatile disc (DVD)), or semiconductor media (eg, solid state disk (SSD)) )Wait. .

Claims

An artificial intelligence AI training method, wherein the method is applied to a guidance system, and after a user of the cloud training system triggers the running of a training code on a local device, the method includes:

Acquire training task information according to the training code running on the local device, where the training code is used to train an AI model, and the AI model is performed by the user based on the AI framework installed on the local device Development and acquisition; upload the training task information to the cloud training system, and notify the cloud training system to execute the training task corresponding to the training task information.
The method according to claim 1, wherein the training task information includes information obtained from the training code by using an obtaining component in the AI framework, and obtained from the training code according to the information in the training code. Information obtained by the local device.
The method according to claim 1 or 2, wherein the training task information includes one or more of the following data: training parameters in the training code, the AI model, the training code Training program logic for training the AI model, training environment information of the local device, and cloud training access information for connecting with the cloud training system.
The method according to claim 3, wherein the training environment information of the local device comprises: version information of the AI framework, and/or version information of a programming language of the training code.
The method according to any one of claims 1-4, wherein the method further comprises:

receiving a training data acquisition request sent by the cloud training system during the execution of the training task;

Acquire the training data according to the training data acquisition request, and send the training data to the cloud training system.
The method according to any one of claims 1-5, wherein before notifying the cloud training system to execute the training task corresponding to the training task information, the method further comprises:

Receive an environment preparation success response returned by the cloud training system.
The method according to any one of claims 1-6, wherein the method further comprises: receiving a trained AI model returned by the cloud training system.
The method according to any one of claims 1-7, wherein the guidance system is obtained from the cloud training system and installed in the local device.
An AI training method, wherein the method is applied to a cloud training system, comprising:

acquiring training task information sent by the guidance system after the user triggers the running of the training code on the local device, where the training task information includes the training environment information of the local device;

performing the preparation of the cloud training environment according to the training environment information;

The training task corresponding to the training task information is executed based on the cloud training environment.
The method according to claim 9, wherein the training environment information of the local device includes: version information of the AI framework on which the AI model to be trained depends, and/or the information used for training the AI model. Version information for the programming language used by the training code.
The method according to claim 10, wherein the performing the preparation of the cloud training environment according to the training environment information comprises:

The AI framework and programming language used to execute the training task in the cloud training environment are set according to the version information of the AI framework and the version information of the programming language of the training code.
The method according to any one of claims 9-11, wherein the training task information further comprises: training parameters in the training code, an AI model, and training in the training code for training the AI model program logic;

The performing the training task corresponding to the training task information based on the cloud training environment includes: performing the training of the AI model in the prepared cloud training environment according to the training parameters and the training program logic .
The method according to any one of claims 9-12, wherein the training task information further comprises: cloud training access information, and before performing the preparation of the cloud training environment according to the training environment information, the method Also includes:

According to the cloud training access information, the authentication and/or charging query is performed on the training task corresponding to the training task information.
A guidance system, characterized in that it includes:

The acquisition module is used to acquire training task information according to the training code running on the local device after the user of the cloud training system triggers the running of the training code on the local device, wherein the training code is used to train AI a model, where the AI model is developed and obtained by the user based on the AI framework installed on the local device;

A sending module, configured to upload the training task information to the cloud training system, and notify the cloud training system to execute the training task corresponding to the training task information.
The system according to claim 14, wherein the training task information comprises information obtained from the training code by using an obtaining component in the AI framework, and obtained from the training code according to the information in the training code Information obtained by the local device.
The system according to claim 14 or 15, wherein the training task information includes one or more of the following data: training parameters in the training code, the AI model, the training code Training program logic for training the AI model, training environment information of the local device, and cloud training access information for connecting with the cloud training system.
The system according to claim 16, wherein the training environment information of the local device comprises: version information of the AI framework, and/or version information of the programming language of the training code.
The system according to any one of claims 14-17, wherein the system further comprises a receiving unit,

the receiving unit, configured to receive a training data acquisition request sent by the cloud training system during the execution of the training task;

The obtaining unit is further configured to obtain the training data according to the training data obtaining request;

The sending unit is further configured to send the training data to the cloud training system.
The system according to any one of claims 14-18, wherein the system further comprises a receiving unit,

The receiving unit is configured to receive an environment preparation success response returned by the cloud training system before the sending unit notifies the cloud training system to execute the training task corresponding to the training task information.
The system according to any one of claims 14-19, wherein the system further comprises a receiving unit,

The receiving unit is configured to receive the trained AI model returned by the cloud training system.
The system according to any one of claims 14-20, wherein the guidance system is obtained from the cloud training system and installed in the local device.
A cloud training system, comprising:

An environment preparation unit, configured to obtain training task information sent by the guidance system after the user triggers the running of the training code on the local device, where the training task information includes the training environment information of the local device; executes the preparation of the cloud training environment according to the training environment information ;

A training task execution unit, configured to execute a training task corresponding to the training task information based on the cloud training environment.
The system according to claim 22, wherein the training environment information of the local device comprises: version information of the AI framework that the AI model to be trained depends on, and/or the training information used for training the AI model Version information for the programming language of the code.
The system according to claim 23, wherein the environment preparation unit is specifically configured to set execution in the cloud training environment according to the version information of the AI framework and the version information of the programming language of the training code The AI framework and programming language used for the training task.
The system according to any one of claims 22-24, wherein the training task information further comprises: training parameters in the training code, an AI model, and training in the training code for training the AI model program logic;

The training task execution unit is specifically configured to execute the training of the AI model in the prepared cloud training environment according to the training parameters and the training program logic.
The system according to any one of claims 22-25, wherein the training task information further comprises: cloud training access information, and the environment preparation unit is further configured to: according to the cloud training access information Perform authentication and/or charging query on the training task corresponding to the training task information.
A computing device, characterized by comprising a processor and a memory, the memory storing computer instructions, the processor executing the computer instructions to cause the computing device to perform any one of the preceding claims 1-8 the method described, or perform the method described in any one of the preceding claims 9-13.
A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program code, and when the computer program code is executed by a computing device, the computing device executes any one of the preceding claims 1-8 A method as claimed in claim 1, or performing a method as claimed in any one of the preceding claims 9-13.