Implementation method of ai model, electronic device, and storage medium

Abstract

The application relates to the technical field of artificial intelligence, and specifically discloses an AI model implementation method, an electronic device and a storage medium. The method comprises the following steps: acquiring demand information published by a user on a client of an AI transaction platform; determining a corresponding AI model algorithm according to the demand information, encapsulating the AI model algorithm into an image file, and loading the image file into the AI transaction platform; performing hyperparameter analysis on the image file, and sending the analyzed hyperparameters to the client for display, so that the user can confirm the hyperparameters; acquiring model training data, and replacing the data path of the model training data with a specified path of the AI transaction platform, so that the model training data can be mapped to the client for display; receiving and responding to a determination instruction of the user, loading the image file, and performing automatic machine learning training according to the model training data to obtain an AI model, and publishing the AI model, thereby realizing the convenience and intelligence of AI model training, and improving the user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to an AI model implementation method, an electronic device and a storage medium. BACKGROUND

[0002] With the rapid development of science and technology, the world has gradually entered the era of artificial intelligence (AI). As an application technology, how to quickly transform AI models into products to realize commercial landing is an urgent problem to be solved in the field of artificial intelligence. The application of AI models includes data processing, model design, model training, model publishing, etc. Nowadays, the implementation method of AI models is usually handed over to AI model engineers for debugging and training, and finally AI models are obtained.

[0003] Due to the difficulty of AI model training, the needs of different types of users for models are not the same. For example, some engineers need AI models with relatively high precision, while some ordinary users do not have such high precision requirements for AI models. However, professional engineers still need to repeatedly debug, which is a waste of resources. SUMMARY

[0004] The present application relates to the technical field of artificial intelligence, and particularly relates to an AI model implementation method, an electronic device and a storage medium, which aims to solve the problem that the current AI model training is difficult, and professional engineers need to repeatedly debug in the training process, which is time-consuming and laborious, resulting in fewer AI model providers and failing to meet the demand and supply.

[0005] To achieve the above-mentioned purpose, the present application provides an AI model implementation method, which comprises:

[0006] Obtaining demand information published by a user on a client of an AI transaction platform, wherein the demand information comprises description information of AI models required by the user;

[0007] Determining a corresponding AI model algorithm according to the demand information, encapsulating the AI model algorithm into an image file, and loading the image file into the AI transaction platform;

[0008] Performing hyperparameter analysis on the image file, and sending the analyzed hyperparameters to the client for display, so that the user can confirm the hyperparameters;

[0009] Obtaining model training data, and replacing the data path of the model training data with a specified path of the AI transaction platform to map to the client for display;

[0010] Receiving and responding to the user's determined instruction, loading the mirror file and automatically machine learning training according to the model training data, obtaining an AI model, and publishing the AI model.

[0011] In addition, to achieve the above object, the present application also provides an electronic device, comprising a memory and a processor; the memory is used for storing a computer program; the processor is used for executing the computer program and realizing the AI model implementation method provided by any one of the embodiments of the present application when the computer program is executed.

[0012] In addition, to achieve the above object, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program makes the processor realize the AI model implementation method provided by any one of the embodiments of the present application when the processor executes the computer program.

[0013] The AI model implementation method, the electronic device and the storage medium disclosed by the embodiments of the present application can determine the corresponding AI model algorithm according to the demand information of the AI model provided by the user, perform automatic machine learning training, and obtain the AI model by building the service end and the client end of the AI transaction platform, so as to realize the provision of the AI model by fewer providers and cannot meet the demand, and improve the user experience. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0015] Figure 1 is an application scenario diagram of an AI transaction platform provided by the embodiments of the present application;

[0016] Figure 2 is a schematic flow chart of an AI model implementation method provided by the embodiments of the present application;

[0017] Figure 3 is a schematic flow chart of obtaining demand information published by a user on a client end of an AI transaction platform provided by the embodiments of the present application;

[0018] Figure 4 is a schematic flow chart of determining whether a user provides an open source AI model algorithm provided by the embodiments of the present application;

[0019] Fig. 5(a) is a schematic diagram of configuring hyperparameters provided by the embodiments of the present application;

[0020] Fig. 5(b) is a schematic diagram of a client task type provided by an embodiment of the present application;

[0021] Fig. 5(c) is a schematic diagram of a configuration database provided by an embodiment of the present application;

[0022] Figure 6 is a schematic diagram of K8S scheduling provided by an embodiment of the present application;

[0023] Figure 7 is a schematic diagram of a defined model parameter provided by an embodiment of the present application;

[0024] Figure 8 is a schematic block diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0026] The flowcharts shown in the drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor do they necessarily have to be executed in the order described. For example, some operations / steps can be further divided, combined or partially merged, so the actual execution order may be changed according to the actual situation. In addition, although the functional modules are divided in the device schematic diagram, in some cases, the module division may be different from that in the device schematic diagram.

[0027] The term "and / or" used in the present application and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0028] At present, the application of AI (Artificial Intelligence) models includes data processing, model design, model training, model publishing, etc. In recent years, the demand for AI models has increased day by day. Due to the difficulty of AI model training, professional engineers need to repeatedly debug during the training process, which is time-consuming and labor-intensive, resulting in fewer providers of AI models, which cannot meet the demand and supply. At the same time, different types of users have different needs for models. For example, some engineers need AI models with high precision, while some ordinary users do not require high precision for AI models. However, professional engineers still need to repeatedly debug, which is a waste of resources.

[0029] To this end, the application provides an AI model implementation method, an electronic device, and a storage medium to solve the above problems.

[0030] Some embodiments of the application will be described in detail below with reference to the accompanying drawings. The following examples and features in the examples can be combined with each other without conflict.

[0031] The AI model implementation method provided by the embodiments of the application is implemented based on an AI transaction platform. Therefore, to facilitate understanding, the AI transaction platform is introduced before the AI model implementation method is introduced.

[0032] The AI transaction platform is a platform system implemented based on a third-party code hosting platform, including a server and a client. The third-party code hosting platform is, for example, a github, gitlab, or other open source code library. There are many AI model algorithms developed by algorithm experts and open sourced on the third-party open source community corresponding to the third-party code hosting platform, which facilitates model training, but still requires complex environment configuration, code debugging, and other work. The demand side cannot easily obtain the required AI model. The AI transaction platform is implemented based on the third-party code hosting platform and improved, so that users can implement without tedious configuration. At the same time, the AI transaction platform also implements low threshold of AI model training based on the third-party code hosting platform to solve the problem of supply and demand mismatch.

[0033] As shown in Figure 1 The client of the AI transaction platform can be installed in a terminal device, and the server can be installed in a server. The terminal device can include, for example, a mobile phone, a tablet computer, a notebook computer, a palm computer, a personal digital assistant (PDA), and a fixed terminal such as a digital TV, a desktop computer, etc. The server can be, for example, a single server or a server cluster.

[0034] The client can obtain demand information published by a user, the demand information including description information of the AI model required by the user, and send the demand information to the server, so that the server determines the corresponding AI model algorithm and training data according to the demand information, performs model training, and sends the trained model to the client for use by the user.

[0035] Specifically, the client can be an application program (APP), when a user opens the APP, the APP displays a demand information interface, so that the user fills in the demand information on the demand information interface, and obtains the demand information filled in by the user on the demand information interface, sends the demand information to the server, and the server executes the AI model implementation method provided in the embodiment of the application according to the demand information to perform model training, and sends the trained model to the user for use.

[0036] The AI model implementation method provided in the embodiment of the application will be described in detail below based on the application scenario in Figure 1 .

[0037] Please refer to Figure 2 , Figure 2 is a schematic flow chart of an AI model implementation method provided in the embodiment of the application. The AI model implementation method can be applied to the server of an AI transaction platform, realizes the intelligentization of AI model training, improves the convenience of users obtaining the required AI model, reduces the cost of manual training of AI model, and improves the user experience.

[0038] As shown in Figure 2 , the AI model implementation method comprises steps S101 to S105.

[0039] S101, obtaining demand information published by a user on a client of an AI transaction platform, the demand information comprising description information of an AI model required by the user.

[0040] In the embodiment of the application, the description information of the AI model comprises the function, type and applicable range of the AI model, that is, the description of what the user wants to do with the AI model, which is used to determine the corresponding AI model algorithm.

[0041] Specifically, the client can display a demand information interface, so that the user fills in the demand information on the demand information interface, and obtains the demand information filled in by the user on the demand information interface.

[0042] In some embodiments, the voice of the user can also be obtained from the client, such as a voice key provided on the client to prompt the user to publish the demand information through voice, and the demand information of the user is obtained by recognizing the voice of the user.

[0043] In some embodiments, as shown in Figure 3 , that is, the step of obtaining the demand information published by the user on the client of the AI transaction platform, specifically comprising the following steps:

[0044] S1011, obtaining the model demand of the user through the client of the AI transaction platform.

[0045] In the embodiments of the present application, the model requirement of the user is used to determine the description information of the corresponding AI model. The AI model requirement can be a functional requirement or a scenario requirement.

[0046] For example, the user can enter the AI transaction platform on the client, such as a notebook computer, and send an AI model requirement for screening specific pictures to the server.

[0047] For example, the user can enter the AI transaction platform on the client, such as a notebook computer, and send an AI model requirement for screening specific pictures to the server.

[0048] S1012, determining a target AI model and description information of the target AI model from the historical AI models according to the model requirement.

[0049] In the embodiments of the present application, the AI model closest to the model requirement is searched from the historical AI models according to the model requirement, and the closest AI model is set as the target AI model.

[0050] The historical AI models are stored on a third-party code hosting platform by encapsulating into image files, and the historical AI models include high-star models collected on the Internet or AI models stored after training through the AI transaction platform. The server can also search high-star models on the Internet and encapsulate them into corresponding image files. The high-star models are the most used AI models on the Internet, the AI models with higher praise rates, or the AI models with stronger practicality. Searching high-star models on the Internet can improve the efficiency and accuracy of determining the target AI model, greatly saving the time cost.

[0051] For example, the server searches the AI model closest to the description information of the AI model provided by the user from the historical AI models in the third-party code hosting platform as the target AI model according to the description information of the AI model provided by the user, and outputs the description information of the target AI model.

[0052] In some embodiments, the server can store the AI model as a historical AI model by encapsulating it into a corresponding image file and storing it on a third-party code hosting platform after publishing the AI model. Encapsulating the AI model into a corresponding image file and storing it on a third-party code hosting platform can facilitate the search for a target AI model next time, improve the efficiency of determining a target AI model, and greatly save the time cost.

[0053] S1013, displaying the description information through the client for the user to confirm.

[0054] In the embodiments of the present application, the description information of the target AI model is sent to the client, and the client displays the received description information of the target AI model for the user to confirm whether the AI model is the model required by the user.

[0055] In some embodiments, the server can also send the description information of one or more target AI models to the client and sort them according to the functional similarity and the size of the applicable range, so that the client displays the description information of these multiple target AI models to the user for the user to select by himself. By sorting multiple target AI models, the most suitable model can be presented to the user more clearly.

[0056] For example, the client displays three target AI models sorted by priority, where the first target AI model has a functional similarity of 95% and an applicable range of the construction industry, the second target AI model has a functional similarity of 90% and an applicable range of the industry, and the third target AI model has a functional similarity of 80% and an applicable range of all industries. At this time, the user can select according to his actual needs.

[0057] It can be understood that the higher the priority of the AI model, the more suitable the target AI model selected by the server for the user's model requirements.

[0058] In some embodiments, if the user is not satisfied with the target model result, the model requirements can also be modified on the client to continue searching for the target AI model. For example, if the user is not satisfied with the implementation function of the target model, the model requirements can be refined and searched again.

[0059] S1014, in response to receiving the user's confirmation instruction of the description information, taking the description information as the requirement information.

[0060] For example, the user's confirmation, modification or exit instruction of the description information of the target AI model is detected. If the user's confirmation instruction of the description information of the target AI model is received, the description information of the target AI model is taken as the requirement information.

[0061] Of course, if the user's modification instruction of the description information of the target AI model is received, the model requirement information modification page is displayed on the client, and if the user's exit instruction of the description information of the target AI model is received, the current page is closed.

[0062] S102, determining the corresponding AI model algorithm according to the requirement information, encapsulating the AI model algorithm into an image file, and loading the image file into the AI transaction platform.

[0063] After determining the AI model algorithm according to the demand information, the AI model algorithm is packaged into an image file for hyperparameter parsing. The AI model algorithm includes an open source AI model algorithm collected by the AI transaction platform or an open source AI model algorithm collected by a user.

[0064] The image file is an executable software package that can be run independently and directly loaded into the AI transaction platform. The software package has low requirements for the runtime environment and is basically not affected by the runtime environment, so that the image file can be stably loaded and run.

[0065] For example, the image file can be a container image (Image), which carries binary data encapsulating the code.

[0066] In some embodiments, if the image file is stored in a third-party code hosting platform, a tag can be added after the image file name to identify different versions in the same image sequence through the tag.

[0067] The same image sequence carries different versions of the image file, and the different versions are an initial image file and an image file after user parameter modification.

[0068] In some embodiments, in order to quickly determine whether a user provides an open source AI model algorithm, the step of determining whether the user provides the open source AI model algorithm includes the following steps: Figure 4

[0069] S1021, detecting whether a user provides an open source AI model algorithm;

[0070] S1022, if the user provides an open source AI model algorithm, obtaining a URL corresponding to the open source AI model algorithm collected by the user, and transmitting the URL to a preset interface of the AI transaction platform;

[0071] S1023, if the user does not provide an open source AI model algorithm, using an open source AI model algorithm collected by the AI transaction platform.

[0072] Specifically, whether a user provides an open source AI model algorithm is detected. If the user provides an open source AI model algorithm, a URL corresponding to the open source AI model algorithm collected by the user is obtained, and the URL is transmitted to a preset interface of the AI transaction platform. If the user does not provide an open source AI model algorithm, an open source AI model algorithm collected by the AI transaction platform is used.

[0073] ​Specifically, the detection of whether the user provides an open-source AI model algorithm can be determined according to the data address of the AI model algorithm. If it is detected that the URL corresponding to the open-source AI model algorithm is a private address, it is determined that the AI model algorithm provided by the user is open source. If it is detected that the URL corresponding to the open-source AI model algorithm is an open-source address, it is determined that the AI model algorithm provided by the user is not open source.

[0074] If it is detected that the user provides an open-source AI model algorithm, the URL corresponding to the open-source AI model algorithm collected by the user is obtained, and the URL (Uniform Resource Locator) corresponding to the open-source AI model algorithm is transmitted to the preset interface of the AI transaction platform, so that the AI transaction platform parses the code of the open-source AI model algorithm collected by the user according to the URL (Uniform Resource Locator) and maps to the client display. Wherein, the preset interface is used to obtain the code of the open-source AI model algorithm. By obtaining the URL corresponding to the open-source AI model algorithm collected by the user, the corresponding AI model algorithm can be determined according to the user's demand.

[0075] If it is detected that the user does not provide an open-source AI model algorithm, the AI transaction platform searches for the corresponding open-source AI model algorithm in the third-party code hosting platform. The AI transaction platform collects mainstream open-source AI model algorithms to meet the algorithm requirements of most AI models.

[0076] For example, the server transmits the URL (Uniform Resource Locator) provided by the user to the AI transaction platform, and the AI transaction platform parses the URL to obtain the code of the AI model algorithm and maps the code to the client display for user confirmation.

[0077] In some embodiments, if the user needs to modify the code, the user can modify the code through the client and send a confirmation instruction to the server. If the user does not need to modify the code, the user can send a confirmation instruction to the server after confirmation.

[0078] S103, the hyperparameters of the image file are parsed, and the parsed hyperparameters are sent to the client for display, so that the user confirms the hyperparameters.

[0079] The hyperparameters of the image file are parsed, exposed as strings, and the strings corresponding to the parsed hyperparameters are automatically presented on the client using dynamic front-end technology for user modification.

[0080] In some embodiments, if the mirror file is searched out by a third-party code hosting platform such as github, most of the algorithm projects in the mirror file are written in a relatively standard manner. The mirror file is parsed by a separate hyperparameter parsing function, and the parsed configuration parameters are sent to the client for display and configuration. By sending the parsed configuration parameters to the client for display and configuration, users can view the configuration parameters in real time, so as to modify and change them in a timely manner.

[0081] In some embodiments, if the mirror file is provided by the user himself, the algorithm project in the mirror file may not be written in a standard manner. Therefore, the hyperparameter parsing function in the hyperparameter parsing tool module, such as the parser function, is used to parse the hyperparameters into strings and expose them.

[0082] For example, by using a file parsing tool, the hyperparameters in the algorithm open source project are automatically parsed, and the parsed hyperparameters are automatically presented on the client side using dynamic front-end technology.

[0083] In some embodiments, different parsing functions correspond to different parsing formats, but there are a limited number of parsing methods, which can be uniformly encapsulated as a unified interface.

[0084] For example, the parsing function of the json format configuration file is json.loads, and the parsing function of the argparse format configuration file is parse_args. They have different parsing formats, but their parsing methods are based on python parsing, so they can be encapsulated as a unified interface.

[0085] In some embodiments, when the client requests the parameter configuration page of the server, an HTML (Hypertext Markup Language) page is generated and sent to the client. Users can configure and modify the hyperparameters on the HTML (Hypertext Markup Language) page, and after the modification is confirmed, the server is sent again and the mirror file is updated.

[0086] In some embodiments, the hyperparameters displayed on the client allow users to modify and change them; and / or the code displayed on the client allows users to modify it. Users can more accurately train the required model by modifying and changing the hyperparameters on the client.

[0087] For example, users can configure the hyperparameters on the HTML page generated by the server. The configuration of the hyperparameters is presented as selecting a button on the client side such as a web page, sending a confirmation instruction to the server, and updating the mirror file.

[0088] As shown in FIG. 5(a), the user can select an AI training data set on the HTML page generated by the server, wherein the date refers to a data set that has been manually labeled. The user can configure the AI training data set on the HTML page generated by the server, wherein the configuration parameters are presented in the form of a category option on the client side, such as a web page, to filter and select a part of the data in the data set, so as to train a model with more suitable configuration parameters. After the configuration is completed, a confirmation instruction is sent to the server, and the image file is updated.

[0089] As shown in FIG. 5(b), which is a schematic diagram of the configuration parameters. The configuration parameters can include a model backbone, a head network, a neck network, a learning rate, a weight decay, and the like.

[0090] In S104, the model training data is obtained, and the data path of the model training data is replaced with a specified path of the AI transaction platform so as to be mapped to the client for display.

[0091] The model training data is obtained through a third-party code hosting platform, the model training data is uniformly replaced with a specified data path, and the model training data is mapped to the client for display. The model training data includes open source data or self-owned data. The self-owned data is training data provided by the user. The specified path of the AI transaction platform is the data path of the open source data.

[0092] In some embodiments, in order to quickly confirm whether the model training data is standard, the step of determining the type of the model training data includes the following contents:

[0093] The data type of the model training data is detected. If the model training data is self-owned data, the data format of the self-owned data model training data is replaced with the data format of the open source data. If the model training data is open source data and the data format is a predefined open source data format, the data format does not need to be converted.

[0094] The data type of the model training data can be determined according to the data address of the model training data. If the data address of the model training data is an open source address, the model training data is determined to be open source data. If the data address of the model training data is a private URL, the model training data is determined to be self-owned data.

[0095] If the model training data is self-owned data, the data format of the self-owned data is converted to a predefined data format of open source data, that is, the data path of the self-owned data model training data is also replaced with a predefined data path of the open source data, and an instruction for determining the model training data is sent.

[0096] In some embodiments, the transaction platform only needs to predefine the open source data format of the plurality of tasks, and the user can convert the self-owned data format into the open source data format according to the open source data format, so as to help the user to load the self-owned data into the open source algorithm training with the help of automation.

[0097] As shown in FIG. 5(c), for example, when the requirement information of the AI model is a detection task, the model training data format of the user can be agreed to use the COCO format or the VOC format to avoid replacing the data format, so as to more quickly determine the model training data.

[0098] For example, when the requirement information of the AI model is a classification task, the model training data format of the user can be agreed to use the VOC format to avoid replacing the data format.

[0099] For example, if the detection model training data is open source data and the data format is a pre-defined open source data format, the open source data does not need to replace the data path, and the data path of the open source data is the specified path of the AI transaction platform, and an instruction for determining the model training data is sent.

[0100] For example, if the detection model training data is open source data but the data format is not a pre-defined open source data format, the data format of the open source data still needs to be converted into a pre-defined open source data format, that is, the data path of the open source data model training data is also replaced by the data path of the pre-defined open source data, and an instruction for determining the model training data is sent.

[0101] S105, receiving and responding to the user's determination instruction, loading the image file and automatically machine learning training according to the model training data to obtain an AI model, and publishing the AI model.

[0102] Among them, the user's determination instruction includes the user's instruction for determining the hyperparameters and the user's instruction for determining the model training data, when receiving the determination instruction and responding to the determination instruction, loading the image file and automatically machine learning training (AutoML) according to the model training data to obtain an AI model, and publishing the AI model on the transaction platform. Thus, the automatic training of the AI model training can be realized, the efficiency of the AI model training is improved, and the time cost is greatly saved.

[0103] The AI model is published on the transaction platform, and specifically, an access interface of the AI model can be generated, and the access interface is sent to the user through the client, so that the user uses the AI model through the access interface.

[0104] In some embodiments, the AI ​​trading platform includes a container orchestration and scheduling tool, which includes any one of Kubernetes (K8S), Swarm, or Mesos.

[0105] For example, taking the K8S (Kubernetes) tool as an example, the image file is loaded and automatic machine learning training is performed based on the model training data. Specifically, based on the K8S tool, the image file is loaded and corresponding computing resources are scheduled. Using the image file and the computing resources, automatic machine learning training is performed based on the model training data to obtain an AI model, and the AI ​​model is published on the AI ​​trading platform.

[0106] The scheduling method using Kubernetes, such as Figure 6 As shown, that is Figure 6 The scheduling method of the K8S platform includes the following steps:

[0107] S1051, the K8S scheduler obtains cluster information from the application programming interface server;

[0108] S1052: Add unscheduled algorithm pods to the scheduling queue and select deployment nodes in sequence;

[0109] S1053, the K8S scheduler sends the deployment node to the aforementioned application interface server, which modifies the deployment node of the algorithm pod;

[0110] S1054, the Kubelet daemon on the deployment node creates the pod image.

[0111] Specifically, such as Figure 6 As shown, Node3 and Node2 correspond to the daemon processes before scheduling, as illustrated in diagrams A and B. After scheduling, previously unscheduled pods can appear in daemon processes such as those in Node1. Figure 6 The C block diagram is shown in the figure.

[0112] In some embodiments, such as Figure 7 As shown, automated machine learning training also allows users to define model parameters, including minimum model width, maximum model width, maximum model depth, minimum model depth, operator type, chip type, whether to compress or jumper topology, etc., and send them to the client for display and update its image file.

[0113] Specifically, the model parameter interface can be displayed on the client side, such as... Figure 7As shown, the model parameter interface at least includes a minimum model width, a maximum model width, a maximum model depth, a minimum model depth, an operator type, and a chip category, so that the user selects or determines the corresponding model parameters, and then trains according to the model parameters determined by the user.

[0114] In some embodiments, the user can modify the parameters on the client, and after the modification is completed, send a confirmation modification instruction to the server. The server defines the parameters according to the modification instruction, and then starts the AI model automatic training.

[0115] In the embodiments of the present application, the server receives the transaction information input by the user through the client, and provides the user with the access interface corresponding to the AI model according to the transaction information. In this way, the convenience of model transaction can be improved, and the user experience is improved.

[0116] For example, the user inputs an order of the AI model of the classification function on the client, and the server trains the AI model according to the demand and provides the user with the access interface corresponding to the AI model.

[0117] The method of the present application can be used in many general or special computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0118] For example, the above method can be implemented in the form of a computer program, which can run on an electronic device as shown. Figure 8

[0119] Please refer to Figure 8 , Figure 8 is a schematic diagram of an electronic device 200 provided by an embodiment of the present application. The electronic device can be a server or a terminal.

[0120] As shown in Figure 8 , the electronic device 200 includes a processor 202 and a memory 201 connected by a system bus, wherein the memory can include a non-volatile storage medium and an internal memory.

[0121] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions which, when executed, can cause the processor to execute any implementation method of an AI model.

[0122] The processor is used to provide computing and control capabilities to support the operation of the entire electronic device.

[0123] ​The internal memory provides an environment for running a computer program in a non-volatile storage medium, which, when executed by the processor, enables the processor to perform an implementation method of any AI model.

[0124] Those skilled in the art can understand that the structure of the electronic device is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0125] It should be understood that the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0126] In some embodiments, the processor is configured to run a computer program stored in the memory to perform the following steps:

[0127] Obtain demand information published by a user on a client of an AI transaction platform, the demand information including description information of an AI model required by the user; determine a corresponding AI model algorithm according to the demand information, encapsulate the AI model algorithm into an image file, and load the image file into the AI transaction platform;

[0128] Perform hyperparameter analysis on the image file, and send the analyzed hyperparameters to the client for display, so that the user confirms the hyperparameters; obtain model training data, and replace the data path of the model training data with a specified path of the AI transaction platform to map to the client for display; receive and respond to a determination instruction of the user, load the image file and perform automatic machine learning training according to the model training data to obtain an AI model, and publish the AI model.

[0129] In some embodiments, the URL corresponding to the open-source AI model algorithm collected by the user is obtained; the URL is input into a preset interface of the AI transaction platform, so that the AI transaction platform parses the code of the open-source AI model algorithm collected by the user according to the URL, and maps to the client display.

[0130] In some embodiments, the hyperparameters of the client display allow the user to modify and change; and / or, the code of the client display allows the user to modify.

[0131] In some embodiments, the processor, when realizing the acquisition of the demand information published by the user on the client of the AI transaction platform, is specifically used for:

[0132] The model demand of the user is obtained through the client of the AI transaction platform; the target AI model and the description information of the target AI model are determined from the historical AI model according to the model demand; the description information is displayed on the client for the user to confirm; when the confirmation instruction of the user on the description information of the target AI model is received, the description information of the target AI model is taken as the demand information.

[0133] In some embodiments, the processor, when realizing the determination of whether the user provides the open-source AI model algorithm, is specifically used for:

[0134] If it is detected that the user provides the open-source AI model algorithm, the URL (uniform resource locator) is input into the preset interface of the AI transaction platform, so that the AI transaction platform parses the code of the open-source AI model algorithm collected by the user according to the URL (uniform resource locator), and maps to the client display. If it is detected that the user does not provide the open-source AI model algorithm, the AI transaction platform searches for the corresponding open-source AI model algorithm in the third-party code hosting platform. The AI transaction platform collects mainstream open-source AI model algorithms to meet the algorithm demand of most AI models.

[0135] In some embodiments, the processor, when realizing the determination of the model training data type, is specifically used for:

[0136] It can be determined according to the data address of the AI model algorithm whether the user provides the open-source AI model algorithm. If it is detected that the URL corresponding to the open-source AI model algorithm is not recorded on the third-party code hosting platform, it is determined that the user provides the open-source AI model algorithm. If it is detected that the URL corresponding to the open-source AI model algorithm is recorded on the third-party code hosting platform, it is determined that the user does not provide the open-source AI model algorithm.

[0137] In some embodiments, based on a container orchestration scheduling tool, the image file is loaded and corresponding computing resources are scheduled; and automatic machine learning training is performed according to the model training data by using the image file and the computing resources.

[0138] In some embodiments, the user inputs transaction information through the client, and the AI model corresponding access interface is provided to the user according to the transaction information.

[0139] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program, and the computer program includes program instructions, and the program instructions are executed to implement any of the AI model implementation methods provided by the embodiments of the present application.

[0140] The computer readable storage medium can be an internal storage unit of the electronic device, for example, a hard disk or a memory of the electronic device. The computer readable storage medium can also be an external storage device of the electronic device, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0141] Further, the computer readable storage medium can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, etc., and the data storage area can store data created according to the use of the blockchain node, etc.

[0142] The blockchain language model referred to in the present application is a new application mode of computer technology, including storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain, in essence, is a decentralized database, which is a series of data blocks associated using cryptography. Each data block contains information about a batch of network transactions, which is used to verify the validity of the information (anti-fake) and generate the next block. Blockchain can include blockchain underlying platform, platform product service layer, and application service layer, etc.

[0143] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for implementing an AI model, characterized in that, include: Obtain the demand information published by users on the client of the AI ​​trading platform. The demand information includes the user's description information of the required AI model, and the description information is the description information of the target AI model. Based on the required information, the corresponding AI model algorithm is determined, the AI ​​model algorithm is packaged into an image file, and the image file is loaded into the AI ​​trading platform; The image file is parsed for hyperparameters, and the parsed hyperparameters are sent to the client for display so that the user can confirm the hyperparameters. Obtain model training data and replace the data path of the model training data with the specified path of the AI ​​trading platform so that it can be mapped to the client for display; Receive and respond to the user's confirmation instruction, load the image file and perform automatic machine learning training based on the model training data and its corresponding open-source algorithm to obtain an AI model, and publish the AI ​​model; The target AI model is determined through the following methods: The user's model requirements are obtained through the client of the AI ​​trading platform; Based on the model requirements, search for the AI ​​model that is closest to the model requirements from historical AI models, and set the closest AI model as the target AI model; The method further includes: Detect whether the user provides open-source AI model algorithms; If a user provides an open-source AI model algorithm, obtain the URL corresponding to the open-source AI model algorithm collected by the user, and pass the URL to the preset interface of the AI ​​trading platform; If the user does not provide an open-source AI model algorithm, the open-source AI model algorithm collected by the AI ​​trading platform will be used. The model training data includes open-source data or proprietary data, wherein the proprietary data is training data provided by the user. The step of replacing the data path of the model training data with a specified path of the AI ​​trading platform for mapping to the client for display includes: The data type of the model training data is determined based on the data address of the model training data; If the data type of the model training data is proprietary data, the data format of the proprietary data is converted into the data format of open source data so that it can be mapped to the client for display.

2. The method according to claim 1, characterized in that, The AI ​​model algorithm includes open-source AI model algorithms collected by the AI ​​trading platform or open-source AI model algorithms collected by the user; If the AI ​​model algorithm is an open-source AI model algorithm collected by the user, the method further includes: Obtain the URLs corresponding to the open-source AI model algorithms collected by the user; The URL is passed to the preset interface of the AI ​​trading platform so that the AI ​​trading platform can parse the code of the open-source AI model algorithm collected by the user according to the URL and map it to the client for display.

3. The method according to claim 2, characterized in that, The client displays hyperparameters that users can modify; and / or, the client displays code that users can modify.

4. The method according to claim 1, characterized in that, The method further includes: The client of the AI ​​trading platform displays predefined data formats for open-source data from multiple different tasks, allowing users to convert their own data formats according to these open-source data formats.

5. The method according to claim 1, characterized in that, The AI ​​trading platform includes a container orchestration and scheduling tool. The process of loading the image file and performing automatic machine learning training based on the model training data includes: Based on the container orchestration and scheduling tool, the image file is loaded and the corresponding computing resources are scheduled; Using the image file and the computing resources, automatic machine learning training is performed based on the model training data.

6. The method according to claim 1, characterized in that, The release of the AI ​​model includes: The system receives transaction information input by the user through the client and provides the user with the access interface corresponding to the AI ​​model based on the transaction information.

7. The method according to claim 1, characterized in that, The process of obtaining user demand information posted on the AI ​​trading platform's client includes: The user's model requirements are obtained through the client of the AI ​​trading platform; Based on the model requirements, the target AI model and its descriptive information are determined from historical AI models. The description information is displayed on the client so that the user can confirm it. In response to receiving the user's confirmation instruction regarding the description information, the description information is used as the demand information.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement: The method for implementing the AI ​​model as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the method for implementing the AI ​​model as described in any one of claims 1-7.

Images