AI model reasoning method

Abstract

The present application discloses an AI model reasoning method and reasoning platform, which relates to the field of software technology. The AI ​​model reasoning platform deploys at least one AI model through TensorRT; the method includes: receiving a data upload instruction sent by a first terminal, storing the target data resource in the upload instruction to a Redis database; in response to detecting a data identification code corresponding to the target data resource from the Redis database, extracting the target data resource from the Redis database, and determining the target AI model; in response to the target AI model satisfying the model reasoning condition, performing model reasoning on the target data resource, and sending the model reasoning result to the first terminal. The present invention can adapt to a variety of framework models through the platform, input the resources uploaded by the terminal into the AI ​​reasoning model for reasoning output, and can realize online sharing and updating of the model, simplify the process of starting and inputting the terminal by editing the script, and improve the efficiency of model reasoning.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of software technology, and in particular to an AI model reasoning method applied to an AI model reasoning platform. Background Art

[0002] In recent years, with the rapid development of artificial intelligence, AI models have become the favorite of developers and users. AI models can be used to quickly process data resources. Some developer platforms provide model resources and materials for editing and use.

[0003] In related technologies, each platform is limited by technical and cost constraints and cannot achieve visual operation and sharing. Users and developers can only use the model after downloading it through the platform, or for models that can be inferred online, they need to start, input and infer the process through script input command lines in a specific compilation environment, which affects the user experience and reasoning efficiency. Summary of the invention

[0004] This application provides an AI model reasoning method. The technical solution is as follows:

[0005] On the one hand, an AI model reasoning method is provided, the method is used for an AI model reasoning platform, the AI ​​model reasoning platform deploys at least one AI model through TensorRT; the method includes:

[0006] Receive a data upload instruction sent by the first terminal, and store the target data resource in the data upload instruction in a Redis database; the target data resource includes at least one of a picture and a video, and the Redis database contains data resources sent by all terminals that have established a communication connection with the AI ​​model inference platform;

[0007] In response to detecting a data identification code corresponding to the target data resource from the Redis database, extracting the target data resource from the Redis database and determining a target AI model; the target data resource is a data resource uploaded by the first terminal and cached in the Redis database;

[0008] In response to the target AI model satisfying the model reasoning condition, model reasoning is performed on the target data resource, and the model reasoning result is sent to the first terminal.

[0009] On the other hand, an AI model inference method is provided, the method is used for a first terminal, and a communication connection is established with an AI model inference platform, and the AI ​​model inference platform deploys at least one AI model through TensorRT; the method includes:

[0010] Displaying the main operation interface, wherein the main operation interface includes an AI model list;

[0011] In response to receiving a data upload instruction for a target data resource, uploading the target data resource to the AI ​​model reasoning platform so that the AI ​​model reasoning platform stores the target data resource in a Redis database; the target data resource includes at least one of a picture and a video, and the Redis database contains data resources sent by all terminals that have established a communication connection with the AI ​​model reasoning platform;

[0012] In response to receiving the model inference result fed back by the AI ​​model inference platform, an output operation control is displayed in the main operation interface, and the result operation control contains the file resources of the target AI model inference output.

[0013] The beneficial effects brought by the above technical solution include at least: providing users with simple operations by providing a visual interface display, and users can select the target AI model that needs to be inferred from the AI ​​model list by clicking on the main operation interface; for the AI ​​model inference platform, the target data resources received from the terminal can be cached in the Redis database for subsequent extraction and processing, and for the shared inference platform, the model inference can be automatically selected according to the data identification code when the specific AI model inference conditions are met, and the model inference results can be sent to the terminal for interface display without the need for users to execute through script writing operations. The visual interface operation avoids frequent command input, simplifies the process of terminal startup and input through script editing, and improves the efficiency of model inference. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is an operation flow chart of the AI ​​model reasoning method provided in the embodiment of the present application applied to the AI ​​model reasoning platform;

[0015] Figure 2 This is a schematic diagram of the main operation interface of the AI ​​model reasoning platform provided in the embodiment of the present application;

[0016] Figure 3 It is a schematic diagram of an interface for displaying the inference output result of the first terminal provided in an embodiment of the present application;

[0017] Figure 4 It is an operation flow chart of applying the AI ​​model reasoning method provided in another embodiment of the application to the AI ​​model reasoning platform;

[0018] Figure 5 It is an operation flow chart of applying the AI ​​model reasoning method provided in another embodiment of the application to the AI ​​model reasoning platform;

[0019] Figure 6 is a schematic diagram of an operation interface for updating a display model of a second terminal provided in an embodiment of the present application;

[0020] Figure 7 This is a schematic diagram of an interface in which a first terminal displays first prompt information provided by an embodiment of the present application;

[0021] Figure 8 It is a schematic diagram of a model creation interface for creating an AI model provided in an embodiment of the present application;

[0022] Fig. 9 This is a schematic diagram of the interface after the model provided in the embodiment of the present application is successfully created. DETAILED DESCRIPTION

[0023] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0024] The term "multiple" as used herein refers to two or more than two. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the related objects are in an "or" relationship.

[0025] In the related technology, for the AI ​​inference model provided by the cloud, script editing and manual uploading are usually used for cloud model inference. For example, the existing script editing and cloud connection establishment are carried out through the python graphical user interface. It is necessary to enter the script information and operation instructions in the graphical interface, and then upload the resources for model inference after the model is determined. The graphical interface operation is not user-friendly, and the script-based editing method is inefficient. It also takes a certain amount of time to start the model, which seriously affects the user's operating efficiency.

[0026] Figure 1 This is an operation flow chart of the AI ​​model reasoning method provided in the embodiment of the present application applied to the AI ​​model reasoning platform, including the following steps:

[0027] Step 101: Displaying a main operation interface on a first terminal, wherein the main operation interface includes an AI model list.

[0028] In one possible implementation, the user establishes a communication connection with the AI ​​model reasoning platform by installing an application of the AI ​​model reasoning platform on a computer device or logging in to a website. The AI ​​reasoning platform consists of multiple servers, which deploy one or more AI reasoning models through TensorRT. The AI ​​reasoning model uses various neural network algorithms to analyze and process the input resource data to achieve specific functions or purposes. For the client, the user can display a visual interface through a web page or program interface to facilitate user operation.

[0029] Indicatively, Figure 2 As shown, the user displays the main operation interface of the AI ​​model reasoning platform through the first terminal, and the interface displays an AI model display control 201. The user can click the display control 201 to display a drop-down AI model list 210. The AI ​​model list 210 displays all online AI reasoning models. Among them, different AI reasoning models are used to implement different reasoning functions.

[0030] Step 102, in response to receiving a data upload instruction for the target data resource, upload the target data resource to the AI ​​model reasoning platform, so that the AI ​​model reasoning platform stores the target data resource in the Redis database.

[0031] When the user selects the model in the main operation interface and uploads the data resources that need to be processed, the computer device uploads it to the cloud server through the communication connection established with the cloud. When the cloud server receives the target data resources, it caches them in the Redis database. The Redis database is a database used for data caching and saving output files.

[0032] It should be noted that the first terminal can be any one of a computer device, a personal PC, a mobile terminal and a workstation equipped with a corresponding application or logging in to a web page to establish a connection, and the uploaded target data resource can be at least one of a picture, a video or an audio.

[0033] Step 103: Receive a data upload instruction sent by the first terminal, and store the target data resource in the data upload instruction into a Redis database.

[0034] When the AI ​​model inference platform receives the data upload instruction sent by the first terminal, it extracts the target data resources, including pictures, videos or audio, and stores them in the Redis database for subsequent model inference. The Redis database contains the data resources sent by all terminals that have established communication connections with the AI ​​model inference platform.

[0035] Step 104, in response to detecting the data identification code corresponding to the target data resource from the Redis database, extract the target data resource from the Redis database and determine the target AI model.

[0036] Since the AI ​​model inference platform may receive massive amounts of data in the same time period, the resources stored in different terminals in the database need to be set with unique data identification codes to facilitate the subsequent extraction of target data resources and determination of terminals based on the data identification codes.

[0037] The Redis database reads the data identification code stored therein according to a preset time interval. When the data identification code is detected, it indicates that there is unprocessed data in the Redis database, and the corresponding target data resource and the selected target AI model need to be extracted according to the data identification code. For example, the first terminal uploads a picture resource to the first AI model, and uses the first AI model to perform image recognition to complete the portrait cutout. This step requires finding the corresponding picture and the inference model that needs to be executed. Among them, the target data resource is the data resource uploaded by the first terminal and cached in the Redis database.

[0038] Step 105, in response to the target AI model satisfying the model reasoning condition, model reasoning is performed on the target data resource, and the model reasoning result is sent to the first terminal.

[0039] Since the AI ​​model inference platform is a shared operating platform, it can process all data resources uploaded to the cloud. When the target AI model in the platform is in normal working condition, and / or the uploaded data meets the inference requirements, the model inference conditions are met. At this time, the target data resources can be inferred. After the inference is completed, the model inference results are sent to the first terminal.

[0040] Step 106, in response to receiving the model inference result fed back by the AI ​​model inference platform, displaying the result operation control in the main operation interface, the result operation control containing the file resources of the target AI model inference output.

[0041] like Figure 3 As shown, after the first terminal receives the model inference result fed back by the AI ​​model inference platform, it indicates that the inference is finished, and the result operation control 310 is displayed on the main operation interface. The result operation control 310 contains the file resources of the target AI model inference output. For example, pictures and videos after calibration or image processing are output.

[0042] In summary, the AI ​​model inference method and inference model provided in the embodiments of the present application provide users with simple operation by providing a visual interface display. Users can select the target AI model that needs to be inferred from the AI ​​model list by clicking on the main operation interface; for the AI ​​model inference platform, the target data resources received from the terminal can be cached in the Redis database for subsequent extraction and processing, and for the shared inference platform, the model inference can be automatically selected according to the data identification code when the specific AI model inference conditions are met, and the model inference results can be sent to the terminal for interface display without the need for the user to execute through script writing operations. Frequent command input is avoided through visual interface operation, the process of terminal startup and input through script editing is simplified, and the efficiency of model inference is improved.

[0043] Figure 4 This is an operation flow chart of applying the AI ​​model reasoning method provided by another embodiment of the present application to the AI ​​model reasoning platform, including the following steps:

[0044] Step 401: Displaying a main operation interface on the first terminal, the main operation interface including an AI model list.

[0045] Step 402: Receive a selection operation for a target AI model in the AI ​​model list and display a data upload control.

[0046] Step 403, in response to receiving a click operation on the data upload control, the target data resource in the data input box is uploaded to the AI ​​model inference platform.

[0047] like Figure 2 As shown, check AI model 1 in the AI ​​model list, and then display the data upload control 220 in the main operation page. The data upload control 220 includes a data input box 221 and a confirmation control 222. The data input box 221 is used to input target data resources, such as pictures, audio and video. After entering the content to be uploaded, click to upload the target data resources to the AI ​​model inference platform.

[0048] Step 404: in response to receiving the target data resource sent by the first terminal, obtaining a UUID corresponding to the target resource, and generating a data identification code based on the UUID and the account information of the first terminal.

[0049] The above content mentioned that the instructions and data resources received by the AI ​​model inference platform need to be cached in the Redis database. In order to facilitate subsequent extraction and judgment, it is necessary to obtain the UUID of the target data. The UUID is generated based on the provided API interface and time, which can ensure uniqueness. At the same time, it is combined with the account information to generate a data identification code.

[0050] Step 405: cache the target data resource and the corresponding generated data identification code into the Redis database.

[0051] Step 406, in response to detecting the data identification code from the Redis database, determining the target AI model corresponding to the data identification code, and obtaining model status information of the target AI model.

[0052] The detection process set in this step is set according to the preset time. The cloud platform continuously detects whether there is a data identification code in the Redis database. When there is a data identification code, it indicates that there are unreasoned data resources, and then the data identification code is parsed to determine the corresponding terminal and the target data resources that need to be acquired. In addition, in order to further improve operational efficiency, the data upload instruction uploaded by the terminal also includes the model identification of the target AI model. The model identification is used to determine the target AI model to be executed from the AI ​​model library, and the model status information of the target model is obtained to facilitate subsequent judgment.

[0053] Step 407, when the target AI model is in the unstarted state, start the target AI model and extract the target data resource from the Redis database based on the data identification code of the first terminal.

[0054] Since there are many AI models in the platform, models that have not been executed for a long time are in a dormant or unstarted state. After the platform determines the target AI model, it needs to automatically start the corresponding target AI model through a script in the background. While starting the model, the target data resources are extracted from the Redis database based on the data identification code of the first terminal to facilitate subsequent preprocessing operations. For the target AI model in the startup state, the target data resources are directly extracted.

[0055] Step 408: pre-process the target data resource based on the format requirements of the target AI model.

[0056] It should be noted that in order to reduce processor pressure and model inference speed, the target data resources are preprocessed before inputting the target AI model or during the startup process. The preprocessing includes resolution adjustment, size cropping, audio and video editing, and format conversion of image resources, audio resources, and video resources to ensure that the data resources finally input into the target AI model meet the standards, while also improving the model inference efficiency and speed.

[0057] Step 409, obtain the model status information of the target AI model again. When the model status information indicates that the model is in the running state, accelerate the reasoning process of the target AI model through the TensorRT acceleration engine and generate the reasoning result.

[0058] Get the model status information of the target AI model again. When the target AI model has been started or started successfully, the model status information indicates that it is in normal operation. At this time, the TensorRT acceleration engine can be used to accelerate the reasoning process of the target AI model. The TensorRT acceleration engine is compatible with neural networks trained on popular frameworks, optimizes neural network calculations, generates a lightweight runtime engine, and it will maximize throughput, latency, and performance on these GPU platforms and generate reasoning results.

[0059] When the model state information of the target AI model obtained again indicates that it is in another state, such as in an updating state, the model reasoning process needs to be interrupted, and the first prompt information needs to be issued according to the account information of the first terminal. For details, refer to step 506.

[0060] Step 410: Send the model forward reasoning output result to the first terminal, or generate a file compression package based on the reasoning output file, and send a download link of the compressed package to the first terminal.

[0061] In one possible implementation, the first terminal uploads video content, and needs to frame the portraits in the video through the target AI model. After inputting it into the corresponding target AI model in the AI ​​model inference platform, forward reasoning is performed, and the portraits in the output video are automatically framed and displayed. For uploaded image resources, such as identifying various animals in the image, the corresponding target AI model is selected for online recognition and display. For the inference output results, the image or audio and video resources can be processed and generated in the original format and then sent down, or the generated corresponding file resources can be packaged into a compressed package and a download link, providing humanized operation options.

[0062] Step 411, in response to receiving the model inference result fed back by the AI ​​inference platform, displaying the result operation control in the main operation interface, the result operation control containing the file resources of the target AI model inference output.

[0063] like Figure 3 As shown, in the case where the reasoning is successful and then sent to the first terminal, the first terminal receives the feedback information and then displays the result operation control 310 in the main operation interface. The result operation control 310 may display a preview box 311 and / or a download link box 312. The preview box 311 is used to display the output file after the model reasoning, such as images, audio and video resources. For batch file resources or resources with converted file formats, a download link box 312 for downloading a compressed package is generated, and the user can save it to the computer device by clicking the Save option.

[0064] In an embodiment of the present application, when the first terminal is operated on a computer device, there is no need to start and run the AI ​​model on the AI ​​model inference platform through script writing instructions. Instead, the target data resources and target AI models to be uploaded are automatically selected through visual operations on the computer device. During the inference process, the AI ​​model inference platform can be compatible with neural network models in multiple framework formats, and deploy them to the platform for inference acceleration, model status detection before inference and automatic execution in the cloud. The startup process can pre-process the target resources to reduce the processing pressure of the target AI model and improve the inference efficiency. In addition, the inference output results can be displayed through the interface to facilitate user operation.

[0065] The above embodiment is an operation flow chart for the case where no other terminal is updating the model. When other terminals operate or modify the target AI model during the process, in order to avoid errors and data validity, the operation authority of the target AI model should be restricted. The specific operation steps are as follows: Figure 5 As shown, it is an operation flow chart of applying the AI ​​model reasoning method provided by another embodiment of the present application to the AI ​​model reasoning platform, including the following steps:

[0066] Step 501: Displaying a main operation interface on the first terminal, the main operation interface including an AI model list.

[0067] Step 502, in response to receiving a data upload instruction for a target data resource, upload the target data resource to the AI ​​model inference platform.

[0068] Step 503, in response to detecting the data identification code corresponding to the target data resource from the Redis data path, extract the target data resource from the Redis library and determine the target AI model.

[0069] Step 504: The second terminal sends a model update instruction to the AI ​​model inference platform.

[0070] As mentioned above, there may be a massive amount of data resources and corresponding data identification codes in the Redis database. When the first terminal is uploading data resources or extracting target resources on the interface, the second terminal selects the target AI model and updates it. Since the AI ​​model inference platform sets a priority, the model update instruction has a higher priority than the model startup and model inference instructions. This is because the updated model inference accuracy and inference effect are better than the model before the update. Therefore, after receiving the model update instruction, the AI ​​model is directly updated. The second terminal is similar to the first terminal, and establishes a communication connection with the AI ​​model inference platform through an account and a visual operation interface, and sends an update instruction.

[0071] In a possible implementation manner, the operation interface diagram of the second terminal is as follows: Figure 6 As shown, the user selects the update option of the target AI model in the AI ​​model list 610 in the main operation interface through the second terminal, and the update control 620 is displayed on the interface. The input box in the update control 620 can be used to upload update data. After confirming the upload operation, a prompt message such as "You will update the parameters of the selected AI inference model. The update process can no longer perform model inference. Please do not exit!" is displayed in the main operation interface.

[0072] Step 505, in response to receiving the model update instruction sent by the second terminal, the model status information of the target AI model is updated, and the operation authority of the other terminal to the target AI model is closed.

[0073] When the platform receives a model update instruction, it will close the operating permissions of other terminals to the target AI model to avoid update errors and interrupt the reasoning process of the target AI model.

[0074] Step 506: Send first prompt information to the first terminal.

[0075] During the update process, the model status information obtained again changes, and other operations can no longer be performed on the target AI model. Therefore, it is necessary to send a first prompt message to the first terminal that selects the target AI model to remind the user that the model status information has changed. The first prompt message displayed on the first terminal is as follows: Figure 7 shown.

[0076] Step 507: Update the model based on the update file uploaded by the second terminal, and restore the operating permissions of other terminals to the target AI model after the model update is completed.

[0077] After the update is completed, the model status information of the target AI model is modified to a normal working state, and the operation permissions of other terminals to the target AI model are restored. For the interrupted reasoning process, continue to take over and execute the reasoning process according to the updated target AI model.

[0078] Step 508: When the model status information indicates that the model is in the running state, the reasoning process of the target AI model is accelerated by the TensorRT acceleration engine, and the reasoning result is generated.

[0079] Step 509: Send the model forward reasoning output result to the first terminal, or generate a file compression package based on the reasoning output file, and send a download link of the compressed package to the first terminal.

[0080] Step 510, in response to receiving the model inference result fed back by the AI ​​inference platform, displaying the result operation control in the main operation interface, the result operation control containing the file resources of the target AI model inference output.

[0081] Step 511: The second terminal sends a model creation instruction to the AI ​​model inference platform.

[0082] The AI ​​model inference platform also provides users with model creation functions, such as Figure 2 As shown, the main operation interface displays a new model creation control 202, and the user can click to send a model creation instruction through the second terminal. It should be noted that any account with platform assigned permissions can create a new model online, such as the accounts of platform operation and maintenance personnel and technical personnel, and the same operation can be performed when the account of the first terminal meets the requirements. Figure 8 It is a schematic diagram of the model creation interface for creating an AI model. The model creation interface 800 displays a data upload control 801 and a data input box 802; the data input box 802 is used to input model files and configuration files, and the data upload control 801 is used to upload the input model files and configuration files to the AI ​​model inference platform, so that the AI ​​model inference platform can deploy the model and update the AI ​​model list 810.

[0083] Step 512: in response to receiving the model creation instruction sent by the second terminal, obtaining the model file and the configuration file included in the model creation instruction.

[0084] Step 513, obtain the configuration information in the configuration file, parse the model file and reconstruct the network based on the configuration information, and deploy the reconstructed AI reasoning model to the AI ​​model reasoning platform.

[0085] The uploaded neural network model file is usually in the onnx format. TensorRT parses the configuration file and model file to eliminate useless output layers in the network to reduce calculations. Secondly, the vertical integration of the network structure merges the three layers of conv, BN, and Relu of the current mainstream neural network into one layer. Once again, the network is horizontally combined to merge the layers that take the same tensor as input and perform the same operation. Finally, the reconstructed output AI inference model is deployed to the AI ​​model inference platform. After the platform is successfully created, it needs to send update instructions to all terminals that have established communication connections to ensure the real-time sharing of the platform. Such as Fig. 9 Shown is a schematic diagram of the interface after the model is successfully created.

[0086] The method and platform provided in the embodiments of the present application fully consider the model update scenario in a multi-person sharing platform. Instead of terminating the reasoning process of the target AI model to be updated by setting priorities, the model is updated and optimized to achieve more accurate and better model performance, and the operating permissions of other terminals are closed. The terminated reasoning process is restored after the update is completed. In addition, the platform also provides the function of creating a new model online, which is also operated through a visual interface. There is no need to call script editing. You only need to upload the trained neural network model and configuration file to the cloud, and parse and deploy them through the cloud. Compared with the traditional way of implementing AI model reasoning through script editing, the efficiency of model reasoning is greatly improved in the present invention.

[0087] The preferred embodiments of the present invention are described above; it should be understood that the present invention is not limited to the above-mentioned specific embodiments, and the devices and structures not described in detail should be understood to be implemented in a common manner in the art; any technician familiar with the art can make many possible changes and modifications without departing from the technical solution of the present invention, or modify them into equivalent embodiments with equivalent changes, which does not affect the essential content of the present invention; therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of protection of the technical solution of the present invention.

Claims

1. An AI model reasoning method, characterized in that: The method is used for an AI model reasoning platform, and the AI ​​model reasoning platform deploys at least one AI model through TensorRT; the method includes: Receive a data upload instruction sent by the first terminal, and store the target data resource in the data upload instruction in a Redis database; the target data resource includes at least one of a picture and a video, and the Redis database contains data resources sent by all terminals that have established a communication connection with the AI ​​model inference platform; In response to detecting a data identification code corresponding to the target data resource from the Redis database, extracting the target data resource from the Redis database, and determining a target AI model; the target data resource is a data resource uploaded by the first terminal and cached in the Redis database; wherein, when the target AI model is not started, first start the target AI model and pre-process the target data resource; In response to the target AI model satisfying the model inference condition, model inference is performed on the target data resource, and the model inference result is sent to the first terminal; specifically, the model status information of the target AI model is acquired again; when the model status information indicates that it is in a model update state, a first prompt information is sent to the first terminal; wherein the first prompt information is used to indicate that the model status information changes when other terminals other than the first terminal perform model update, and the operation permission for the target AI model is closed; Update the model based on the update file uploaded by the second terminal, and restore the operation permissions of other terminals to the target AI model after the model update is completed; When the model status information indicates that it is in the running state, based on the preprocessed target data resources, the TensorRT acceleration engine is used to accelerate the reasoning process of the target AI model and generate a model inference result; Send the forward reasoning output result of the model to the first terminal; or generate a file compression package based on the reasoning output file, and send a download link of the file compression package to the first terminal.

2. The method according to claim 1, characterized in that The receiving the target data resource sent by the first terminal and storing it in the Redis database includes: In response to receiving the target data resource sent by the first terminal, acquiring a UUID universal unique identifier corresponding to the target data resource, and generating the data identification code based on the UUID and the account information of the first terminal; The target data resource and the corresponding generated data identification code are cached in the Redis database.

3. The method according to claim 2, characterized in that In response to detecting a data identification code corresponding to the target data resource from the Redis database, extracting the target data resource from the Redis database, and determining a target AI model, including: In response to detecting the data identification code from the Redis database, determining the target AI model corresponding to the data identification code, and acquiring model status information of the target AI model; the model status information is used to characterize the working status of the AI ​​reasoning model; When the target AI model is in an unstarted state, start the target AI model, and extract the target data resource from the Redis database based on the data identification code of the first terminal; Based on the format requirements of the target AI model, the target data resources are preprocessed; the preprocessing includes at least one of resolution adjustment, size cropping and format conversion of the image resources and video resources.

4. The method according to claim 1, characterized in that: The method further comprises: In response to receiving a model update instruction sent by the second terminal, updating the model status information of the target AI model, and closing the operation authority of other terminals to the target AI model; wherein the priority of the model update instruction is higher than the model inference instruction; The model is updated based on the update file uploaded by the second terminal, and after the model update is completed, the operation permissions of other terminals to the target AI model are restored.

5. The method according to claim 1, characterized in that The method further comprises: In response to receiving the model creation instruction sent by the second terminal, acquiring the model file and the configuration file included in the model creation instruction; Obtain the configuration information in the configuration file, parse and reconstruct the model file based on the configuration information, and deploy the reconstructed AI reasoning model to the AI ​​model reasoning platform.

6. An AI model reasoning method, characterized in that: The method is used for the first terminal described in any one of claims 1 to 5, and establishes a communication connection with an AI model reasoning platform, and the AI ​​model reasoning platform deploys at least one AI model through TensorRT; the method includes: Displaying the main operation interface, wherein the main operation interface includes an AI model list; In response to receiving a data upload instruction for a target data resource, uploading the target data resource to the AI ​​model reasoning platform so that the AI ​​model reasoning platform stores the target data resource in a Redis database; the target data resource includes at least one of a picture and a video, and the Redis database contains data resources sent by all terminals that have established a communication connection with the AI ​​model reasoning platform; In response to receiving a feedback instruction issued by the AI ​​model inference platform, displaying a first prompt information in the main operation interface, wherein the first prompt information is used to indicate that the model status information is changed when the model is updated by other terminals other than the first terminal; In response to receiving the model inference result fed back by the AI ​​model inference platform, an output result operation control is displayed in the main operation interface, and the result operation control contains the file resources of the target AI model inference output.

7. The method according to claim 6, characterized in that In response to receiving a data upload instruction for a target data resource, uploading the target data resource to the AI ​​model inference platform includes: Receive a selection operation of the target AI model in the AI ​​model list, and display a data upload control, wherein the data upload control includes a data input box for uploading data resources; In response to receiving a click operation on the data upload control, the target data resource in the data input box is uploaded to the AI ​​model reasoning platform.

8. The method according to claim 6, characterized in that The main operation interface also displays a model creation control; in response to receiving a trigger operation on the model creation control, a model creation interface is displayed, and the model creation interface displays a data upload control and a data input box; the data input box is used to input model files and configuration files, and the data upload control is used to upload the input model files and configuration files to the AI ​​model inference platform, so that the AI ​​model inference platform can deploy the model and update the AI ​​model list.

Images