Artificial intelligence full life cycle practical training platform and method

Abstract

The application provides an artificial intelligence full life cycle training platform and method, and belongs to the technical field of artificial intelligence, and comprises a data management and labeling module, an online IDE module, a Git repository management module, a templated no-code training module, a templated no-code deployment module, a resource sharing module and a project collaboration module. The application provides a templated no-code training and deployment module, allows students with weak programming foundation to complete the full process of AI model development through visual operation, simultaneously provides an online IDE with embedded common frameworks and code templates for students with advanced ability, realizes the teaching goal of parallel learning and rapid operation of students at different levels, and independently creates a resource sharing mechanism, supports modular sharing and one-key reuse of datasets, labeling tasks, training templates, models and even entire projects. This not only facilitates the management and promotion of excellent teaching achievements by teachers, but also promotes the cooperation and secondary innovation among students, and forms a sustainable teaching resource ecology.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to an artificial intelligence full lifecycle training platform and method. Background Technology

[0002] To gain a competitive edge in the field of artificial intelligence, breakthroughs must be achieved in fundamental theories, methods, and tools. From the perspective of AI empowering higher education, general education courses in AI are fundamental to supporting the digital transformation of universities. Using this as a starting point, the comprehensive digitalization of courses, majors, teaching, and management can be demonstrated. As a means of cultivating professional talent, the goal of general AI courses is to cultivate well-rounded individuals with both professional AI knowledge and engineering practice capabilities, ensuring that course outcomes can be practically applied and continuously iterated. Traditional general AI education has been highly effective in cultivating professional competence, but it has shortcomings and deficiencies in engineering practice capabilities, as detailed below.

[0003] First, the engineering practice has a high barrier to entry. Traditional AI practical teaching not only requires students to master basic knowledge such as advanced mathematics and algorithm theory, but also heavily relies on the ability to write native code, demanding high levels of skills in Python programming and framework API calls. This causes many beginners with weak programming foundations to get stuck in the "code implementation" stage, making learning difficult and frustrating, which has become one of the core challenges in AI practical teaching.

[0004] Secondly, the complete AI R&D process is fragmented and time-consuming. Professional AI development involves multiple stages, including data preparation, model training, deployment, and verification, typically requiring the use of various platforms and tools, resulting in a long overall development cycle. This complex process is ill-suited to the time constraints and frequent real-time human-computer interaction required in classroom teaching, posing a significant challenge for beginners.

[0005] Therefore, the existing AI practical training teaching model has shortcomings in cultivating students' engineering application capabilities. Summary of the Invention

[0006] To address the shortcomings of existing AI training models in cultivating students' engineering application capabilities, this invention provides an AI full lifecycle training platform that can lower the practical threshold, integrate R&D processes, and improve teaching efficiency.

[0007] To achieve the above objectives, the present invention provides the following technical solution: An artificial intelligence full lifecycle training platform, comprising: The project collaboration module is used to create independent project instances based on the topic name and description entered by the teacher; The Git repository management module is used to create Git repositories that are bound to project instances and basic task templates; The data management and annotation module is used to construct a dataset based on the original sample data uploaded by students and to store the uploaded dataset in a versioned manner. The dataset includes original files and annotation files. It is also used to annotate the original files in the dataset online according to the visual annotation task template selected or created by the students. When multiple students are received to annotate the same target in the same area with the same semantics, the module calculates the maximum overlapping area to achieve automatic conflict merging. The online IDE module is used to select a dataset from the data management and annotation module based on the Git repository code associated with the current project directly pulled by the student. The dataset is then mounted into the online IDE environment according to the specified file organization format and supports simultaneous access by multiple users. The module also allows students to modify and debug the algorithm logic based on the basic task template, and then submit the code to the corresponding Git repository with one click upon completion. The templated no-code training module is used to allow students to select the dataset, predefined model structure and training hyperparameters through visual drag-and-drop or form filling, start the training task with one click, and automatically register the weight file of the trained model. The templated no-code deployment module is used to receive students' selections of a specified version of a model from a list of trained models, and to package and publish the selected model as an online microservice with one click.

[0008] Preferably, the project collaboration module is also used to decompose the life cycle of the model into standardized stages, in which teachers define different tasks and assign them to one or more students, and the students assigned tasks obtain the permissions to use the tools and data assets needed to complete the corresponding tasks.

[0009] Preferably, it also includes a preprocessing module, which is used to obtain the specified version of data from the data management and annotation module according to the dataset ID selected by the student when the training task starts, and automatically perform preprocessing; when the training task starts, it automatically schedules the available computing resources according to the task requirements.

[0010] Preferably, the templated no-code deployment module is also used to display the test model's performance through a visual interactive UI.

[0011] Preferably, the data management and annotation module achieves isolated storage of data source files and annotation data through an independent canvas, and provides preset annotation task templates, allowing for real-time preview of annotation results.

[0012] Preferably, the data management and annotation module provides a visual annotation interface, supports the creation and application of various task templates such as classification, detection, and segmentation, and stores the annotation results in a structured manner in association with the original data.

[0013] Preferably, it also includes a resource sharing module, which is used to manage model weight files and to publish student datasets, selected projects, custom training templates or models to the platform's public space; other teachers and students can copy the published resources to their own workspaces for learning or secondary development with one click.

[0014] Preferably, it also includes a permission management module, which is used to assign different functional module usage permissions to users with different identities.

[0015] Preferably, it also includes a code writing module, a debugging module, a commit module, and a branch management module; the code writing module is used to write code based on the basic task template, the debugging module is used to locate and correct code errors, the commit module is used to synchronize the code to the Git repository and save the code version, and the branch management module is used to create, merge, and isolate different development tasks.

[0016] This invention also provides a training method for an artificial intelligence full lifecycle training platform, characterized by comprising the following steps: After logging into the platform, teachers first create an independent project instance for the topic project in the project collaboration module, and then create a Git repository and basic task template bound to the project instance in the Git repository management module. Within the project, students or teaching assistants upload raw data samples through the data management and annotation module to build datasets and set the labels required for the target scenario. The system stores the uploaded data in a versioned manner. Teachers create data annotation tasks through the data management and annotation module and assign them to multiple students, while also managing the annotation of the uploaded datasets. In the online IDE module, students can directly pull the Git repository code associated with the current project. After selecting the dataset in the data management and annotation module, the system uses a POSIX interface based on the TCP / IP protocol to mount the dataset from the remote service into the online IDE environment according to the specified file organization format, and supports simultaneous access by multiple users. Students can modify and debug the algorithm logic based on the template, and then commit the code to the corresponding Git repository with one click. In the templated no-code training module, students can select the dataset, predefined model structure, and training hyperparameters by visual drag-and-drop or form filling, then start the training task with one click and observe the visual parameter indicators generated by the model training. Students can select a specific version of a model from the list of trained models in the templated no-code deployment module, package it, and publish it as an online microservice with one click.

[0017] The AI ​​full lifecycle training platform provided by this invention has the following beneficial effects: This invention provides templated, no-code training and deployment modules, allowing students with weak programming skills to complete the entire AI model development process through visual operations. Simultaneously, it offers an online IDE with embedded commonly used frameworks and code templates for students with advanced abilities, achieving the teaching goal of parallel learning and rapid mastery for students at different levels. A unique resource-sharing mechanism supports modular sharing and one-click reuse of datasets, annotation tasks, training templates, models, and even entire projects. This not only facilitates teachers' management and promotion of excellent teaching outcomes but also promotes collaboration and secondary innovation among students, forming a sustainable teaching resource ecosystem. Attached Figure Description

[0018] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a functional module diagram of the artificial intelligence full lifecycle training platform provided in Embodiment 1 of the present invention; Figure 2 The flowchart below shows the training method for the artificial intelligence full life cycle training platform provided in Embodiment 2 of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0021] Example 1 This invention provides an artificial intelligence full lifecycle training platform, specifically a technical solution for a full lifecycle artificial intelligence training platform based on machine learning operations and maintenance (MLOps). This solution adopts a modular and microservice architecture, integrating the entire process of artificial intelligence practical teaching into a unified platform. It aims to reduce the engineering threshold and improve training efficiency through a series of technical means.

[0022] The core technology of this invention lies in constructing an integrated system that includes functions such as data management, online development, no-code training, automated deployment, and resource collaboration. The overall system architecture consists of a series of functional modules, which are interconnected and communicate through a unified service bus or microservice framework, jointly supporting the entire teaching practice process from data to model deployment. Specifically, as follows... Figure 1 As shown, the overall architecture of this invention consists of multiple functional modules, including a permission management module 101, a Git repository management module 102, a project collaboration module 103, a preprocessing module 104, a data management and annotation module 105, an online IDE module 106, a templated no-code training module 107, a templated no-code deployment module 108, and a resource sharing module 109. Each module runs as an independent microservice and communicates through a unified service bus, ensuring high cohesion, low coupling, and scalability of the system. Simultaneously, the system globally employs role-based access control (RBAC) technology, enabling fine-grained permission management for different roles such as teachers, teaching assistants, and students.

[0023] Specifically, the access control module 101 is used to assign different access permissions for different functional modules to users with different identities, with teachers being the primary users.

[0024] The project collaboration module 103 is used to create independent project instances based on the topic name and description entered by the teacher; the project instance includes the topic name and description. The project collaboration module 103 breaks down the life cycle of the artificial intelligence model into standardized stages. In each stage, the teacher defines different tasks and assigns them to one or more students. The students assigned tasks will have access to the tools and data assets needed to complete the corresponding tasks.

[0025] Git repository management module 102 is used to create Git repositories bound to project instances and a set of basic task templates. This module is not directly exposed to ordinary users of the platform, but it will support version management of user-created data assets and code assets.

[0026] To achieve standardized management of AI data, this invention provides a data management and annotation module 105. The data management technology supports unified access to multi-source heterogeneous data such as CSV files, images, and audio, and employs a versioned storage mechanism to ensure the traceability and reproducibility of any version of the data. The data annotation technology provides a visual annotation interface, supports the creation and application of various task templates such as classification, detection, and segmentation, and provides structured association storage between annotation results and original data, offering data version management technology covering the entire lifecycle.

[0027] Specifically, the data management and annotation module 105 is used to construct a dataset based on the original sample data uploaded by students and to store the uploaded dataset in a versioned manner. The dataset includes original files and annotation files. It is also used to annotate the original files in the dataset online according to the visual annotation task template selected or created by the students. When multiple students annotate the same target area with the same semantic meaning, the system calculates the maximum overlapping area to achieve automatic conflict merging. The original data samples are in formats such as images, text, audio, and CSV files. For multimodal data such as images and text, both original file uploads and standardized dataset file uploads are supported. The system will automatically parse the formats of the original files and annotation files in the dataset.

[0028] The data management and annotation module 105 achieves isolated storage of data source files and annotation data through an independent canvas, and provides pre-built annotation task templates, allowing for real-time preview of annotation results. Simultaneously, the data management and annotation module 105 offers a visual annotation interface, supporting the creation and application of various task templates such as classification, detection, and segmentation, and structurally associates and stores annotation results with the original data. The data management and annotation module 105 supports the data types required for common AI scenarios and maintains an open architecture that allows for continuous integration of the latest application scenarios.

[0029] The online IDE module 106 is used to select a dataset from the data management and annotation module 105 based on the Git repository code directly pulled by the student from the current project. The dataset is then mounted into the online IDE environment according to a specified file organization format, and multiple users can access it simultaneously. Students can modify and debug the algorithm logic based on the template, and then commit the code to the corresponding Git repository with one click. The online IDE module 106 supports simultaneous access by multiple users.

[0030] To address the challenge of insufficient programming skills among learners, one of the core technologies of this invention is template-based no-code training. This technology abstracts and encapsulates the complex model training process into predefined training pipeline templates, which encapsulate standard steps such as data loading, preprocessing, model definition, and training optimization. Users only need to configure key hyperparameters through visual drag-and-drop or form filling, and the system can automatically parse the configuration and schedule backend computing resources to execute training tasks, thus transforming the complex coding process into a simple configuration process.

[0031] Specifically, the templated no-code training module 107 allows students to select the dataset, predefined model structure, and training hyperparameters via visual drag-and-drop or form filling, and start the training task with one click. The completed model weight file is automatically registered. It also provides a visual interactive UI for testing model performance and standardized interfaces for providing AI capability support to external systems.

[0032] To bridge the "last mile" from model training to application, this invention employs templated no-code deployment technology, providing a templated no-code deployment module 108. This module allows students to select a specific version of a model from a list of trained models and package it into a standardized microservice (supporting RESTful or gRPC protocols) that can be invoked online with a single click. This process automatically integrates key MLOps aspects such as model containerization, service deployment, version management, and online A / B testing, enabling students to quickly deploy and validate models without requiring specialized operational knowledge. The templated no-code deployment module 108 also displays the test model's performance through a visual interactive UI.

[0033] To facilitate the accumulation and dissemination of teaching outcomes, this invention includes a resource sharing module 109. All core assets within the platform, including datasets, annotation tasks, model weights, training templates, and even entire projects, can be shared and reused as independent structured resources with access control. The "one-click copy to personal space" technology greatly promotes the rapid dissemination and secondary innovation of knowledge.

[0034] Specifically, the resource sharing module 109 is used to manage model weight files and to publish student datasets, selected outstanding projects, custom training templates or models to the platform's public space; other teachers and students can copy the published resources to their own workspaces with one click for learning or secondary development.

[0035] The preprocessing module 104 is used to obtain the specified version of data from the data management and annotation module 105 according to the dataset ID selected by the student when the training task starts, and automatically perform preprocessing; when the training task starts, it automatically schedules the available computing resources according to the task requirements.

[0036] This invention provides a deep integration technology between an online IDE and a version control system. To eliminate the complexity of configuring a local development environment, it offers an online IDE module 106 that integrates mainstream AI frameworks (such as TensorFlow and PyTorch). More importantly, this IDE is deeply integrated with the built-in Git repository management module 102. This embodiment also provides a code writing module, a debugging module, a commit module, and a branch management module. The code writing module is used for writing code based on basic task templates; the debugging module is used to locate and correct code errors; the commit module is used to synchronize code to the Git repository and save code versions; and the branch management module is used to create, merge, and isolate different development tasks. Developers can directly complete a series of version control operations such as code writing, debugging, committing, and branch management within the browser environment, achieving seamless integration of development and version management.

[0037] Example 2 Based on the aforementioned AI full lifecycle training platform, this invention also proposes an AI full lifecycle training method, such as... Figure 2 As shown, the specific steps include: Step 1: Project Creation and Initialization After logging into the platform, teacher users first select the "Create Practice Topic" function in the project collaboration module 103, and enter the topic name and description. At this time, the system will automatically generate an independent project instance for the topic project, and create a Git repository bound to the project instance and a set of basic task templates in the Git repository management module 102.

[0038] Step 2: Data Preparation and Labeling Within the project, students or teaching assistants upload raw data samples (such as images, CSV files, etc.) through the data management and annotation module 105 to build datasets and set the labels required for the target scenario. The system stores the uploaded data in a versioned manner.

[0039] Teachers create data annotation tasks and assign them to multiple students, while also managing the annotation of the uploaded datasets.

[0040] Subsequently, in the data management and annotation module 105, users can select or create a visual annotation task template (such as classification, detection, etc.) to annotate the original files in the dataset online. The data management and annotation module 105 uses an independent canvas to isolate and store the data source files and annotation data, and provides pre-set annotation task templates, such as image classification, object detection (bounding boxes, polygons), semantic segmentation, text classification, named entity recognition, etc., covering mainstream AI task types. Annotation results can be previewed in real time. The platform supports multi-user collaborative annotation tasks, and teachers can monitor the annotation progress in real time. When multiple students annotate the same target area with the same semantic meaning, the system calculates the maximum overlapping area to achieve automatic conflict merging capability.

[0041] Step 3: Code Development and Debugging In the online IDE module 106, students can directly pull the Git repository code associated with their current project. This IDE is a browser-based online coding environment that comes pre-installed with commonly used AI frameworks (such as TensorFlow and PyTorch) and code templates for customized AI algorithms tailored to students' majors. The online IDE module 106 achieves seamless integration between online editing and the Git repository. After students finish editing their code and save it in the UI, the system automatically creates a Git version. The online IDE module 106 also achieves seamless integration with the data management and annotation module 105. When students launch the IDE and select a dataset from the data management and annotation module 105, the system will mount the dataset from the remote service to the online IDE environment according to the specified file organization format via a POSIX interface based on the TCP / IP protocol, supporting simultaneous access by multiple users. Students can modify and debug the algorithm logic based on the templates, and then commit the code to the corresponding Git repository with a single click.

[0042] Step 4: No-code model training For students with weak programming skills or those needing rapid validation, users can access the templated no-code training module 107. Through visual drag-and-drop or form filling, users can select the dataset, predefined model structure, and training hyperparameters, then start the training task with one click and observe the visualized parameter metrics generated by the model training. The system will automatically allocate computing resources and provide real-time feedback on training logs and performance indicator charts. The trained model weight file will be automatically registered and managed through the resource sharing module 109. The resource sharing module 109 employs automated machine learning and model training orchestration technologies. The platform maintains a predefined model structure library, as well as commonly used machine learning and deep learning algorithms. These models and algorithms are encapsulated and can be updated and instantiated via remote interfaces. The training module is tightly integrated with the data management and annotation module 105. When a training task starts, the system retrieves the specified version of data from the data management and annotation module 105 based on the user-selected dataset ID and automatically performs necessary preprocessing. When a training task starts, the platform automatically allocates available computing resources based on task requirements (e.g., number of GPUs, memory size). Typically, training tasks run in isolated containers. These containers come pre-installed with all the necessary AI frameworks (such as TensorFlow and PyTorch) and libraries, ensuring consistency and reproducibility of the training environment. The execution of training tasks is managed by an orchestrator. It is responsible for: initializing the model and loading the dataset; executing training loops, including forward propagation, loss calculation, backpropagation, and parameter updates; periodically saving model checkpoints and logs; collecting training metrics (such as loss value, accuracy, and validation set performance) in real time and pushing them to the front end for visualization; and handling training interruptions and error recovery.

[0043] Step 5: Deployment and Validation of the No-Code Model In the templated no-code deployment module 108, users can select a specific version of a model from the list of pre-trained models, package it with one click, and publish it as an online microservice (such as a RESTful API). The purpose of model packaging is to encapsulate the trained machine learning model and all its necessary dependencies (such as code, configuration, and runtime environment) into an independently runnable, easily deployed, and managed, isolated service unit. After successful deployment, students can verify the model's prediction performance on real data by calling the service interface through the platform's provided test page or in the online IDE module 106. The platform automatically generates a user-friendly test page for the deployed model. Students can directly enter test data on this page (e.g., upload an image for an image classification model; enter text for a text classification model) and then click the "Predict" button. The page will display the model's prediction results in real time, such as the predicted class label and confidence score. For image tasks, it will display the result image with bounding boxes or segmentation masks.

[0044] Step Six: Sharing and Iterating Results Teachers can publish outstanding student projects, customized training templates, or models to the platform's public space via the resource sharing module 109. Other teachers and students can easily copy these resources to their own workspaces for learning or further development, thereby enabling rapid reuse of teaching resources and iterative innovation of knowledge. The platform can also provide a basis for continuous optimization of teaching templates and platform functions through statistical analysis of user data.

[0045] Compared with existing technologies, the MLOps-based full lifecycle AI training platform provided by this invention has the following significant innovations and beneficial effects: 1. Significantly lowers the barrier to AI engineering practice: This invention provides templated, no-code training and deployment modules, allowing students with weak programming skills to complete the entire AI model development process through visual operations. Simultaneously, it provides an online IDE with embedded commonly used frameworks and code templates for students with advanced abilities, achieving the teaching goal of parallel learning and rapid mastery for students at different levels.

[0046] 2. Significantly shortens the AI ​​application development cycle: This invention highly integrates previously fragmented processes such as data management, code development, model training, and service deployment onto a unified platform. Through automated processes such as one-click training start and one-click deployment, it greatly reduces the time required for environment configuration, platform switching, and manual debugging, effectively meeting the needs of real-time interaction and rapid verification in classroom teaching.

[0047] 3. Achieving the Accumulation and Efficient Reuse of Teaching Resources: This invention features a unique resource-sharing mechanism that supports modular sharing and one-click reuse of datasets, annotation tasks, training templates, models, and even entire projects. This not only facilitates teachers' management and promotion of excellent teaching outcomes but also fosters collaboration and secondary innovation among students, forming a sustainable teaching resource ecosystem.

[0048] 4. A continuous iterative closed loop for teaching, learning, and research has been established: The platform supports collaborative teaching models between teachers and students by integrating project collaboration modules. More importantly, the platform can dynamically iterate and optimize functional modules and built-in templates based on user data and teaching feedback, thereby driving continuous improvement of teaching content and methods and forming a virtuous cycle of development.

[0049] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. An artificial intelligence full lifecycle training platform, characterized in that, include: The project collaboration module is used to create independent project instances based on the topic name and description entered by the teacher; The Git repository management module is used to create Git repositories that are bound to project instances and basic task templates; The data management and annotation module is used to construct datasets based on the raw sample data uploaded by students and to store the uploaded datasets in a versioned manner; the datasets include raw files and annotation files; It is used to annotate the original files in the dataset online based on the visual annotation task templates selected or created by students; when multiple students annotate the same target in the same area with the same semantics, it calculates the maximum overlapping area to achieve automatic conflict merging. The online IDE module is used to select a dataset from the data management and annotation module based on the Git repository code associated with the current project directly pulled by the student, mount the dataset into the online IDE environment according to the specified file organization format, and support simultaneous access by multiple users; Students can modify and debug the algorithm logic based on the basic task template, and then submit the code to the corresponding Git repository with one click after completion. The templated no-code training module is used to allow students to select the dataset, predefined model structure and training hyperparameters through visual drag-and-drop or form filling, start the training task with one click, and automatically register the weight file of the trained model. The templated no-code deployment module is used to receive students' selections of a specified version of a model from a list of trained models, and to package and publish the selected model as an online microservice with one click.

2. The artificial intelligence full lifecycle training platform according to claim 1, characterized in that, The project collaboration module is also used to decompose the model's lifecycle into standardized stages. In each stage, teachers define different tasks and assign them to one or more students. The students assigned tasks obtain access to the tools and data assets needed to complete the corresponding tasks.

3. The artificial intelligence full lifecycle training platform according to claim 1, characterized in that, It also includes a preprocessing module, which retrieves the specified version of data from the data management and annotation module based on the dataset ID selected by the student when the training task starts, and automatically performs preprocessing; when the training task starts, it automatically schedules available computing resources according to the task requirements.

4. The artificial intelligence full lifecycle training platform according to claim 1, characterized in that, The templated no-code deployment module is also used to display the test model's performance through a visual interactive UI.

5. The artificial intelligence full lifecycle training platform according to claim 1, characterized in that, The data management and annotation module achieves isolated storage of data source files and annotation data through an independent canvas, and provides preset annotation task templates, allowing annotation results to be previewed in real time.

6. The artificial intelligence full lifecycle training platform according to claim 4, characterized in that, The data management and annotation module provides a visual annotation interface, supports the creation and application of various task templates such as classification, detection, and segmentation, and stores the annotation results in a structured manner in association with the original data.

7. The artificial intelligence full lifecycle training platform according to claim 1, characterized in that, It also includes a resource sharing module, which is used to manage model weight files and publish student datasets, selected projects, custom training templates or models to the platform's public space; other teachers and students can copy the published resources to their own workspaces with one click for learning or secondary development.

8. The artificial intelligence full lifecycle training platform according to claim 1, characterized in that, It also includes a permission management module, which is used to assign different access permissions for different functional modules to users with different identities.

9. The artificial intelligence full lifecycle training platform according to claim 1, characterized in that, It also includes a code writing module, a debugging module, a commit module, and a branch management module; the code writing module is used to write code based on the basic task template, the debugging module is used to locate and correct code errors, the commit module is used to synchronize code to the Git repository and save code versions, and the branch management module is used to create, merge, and isolate different development tasks.

10. A training method for an artificial intelligence full lifecycle training platform as described in claim 1, characterized in that, Includes the following steps: After logging into the platform, teachers first create an independent project instance for the topic project in the project collaboration module, and then create a Git repository and basic task template bound to the project instance in the Git repository management module. Within the project, students or teaching assistants upload raw data samples through the data management and annotation module to build datasets and set the labels required for the target scenario. The system stores the uploaded data in a versioned manner. Teachers create data annotation tasks through the data management and annotation module and assign them to multiple students, while also managing the annotation of the uploaded datasets. In the online IDE module, students can directly pull the Git repository code associated with the current project. After selecting the dataset in the data management and annotation module, the system uses a POSIX interface based on the TCP / IP protocol to mount the dataset from the remote service into the online IDE environment according to the specified file organization format, and supports simultaneous access by multiple users. Students modify and debug the algorithm logic based on the template, and then submit the code to the corresponding Git repository with one click. In the templated no-code training module, students can select the dataset, predefined model structure, and training hyperparameters by visual drag-and-drop or form filling, then start the training task with one click and observe the visual parameter indicators generated by the model training. Students can select a specific version of a model from the list of trained models in the templated no-code deployment module, package it, and publish it as an online microservice with one click.

Images