ModelArts Standard Usage

This chapter aims to help you learn how to use ModelArts Standard and get started with the ModelArts service quickly.

For developers who are experienced in coding, debugging, and working with AI engines, ModelArts provides online coding environments as well as an E2E AI development process that covers data preparation, model training, model management, and service deployment.

This document describes how to perform AI development on the ModelArts management console. If you use the APIs or SDKs for development, view ModelArts SDK Reference or ModelArts API Reference.

To view the examples of AI development lifecycle, see Getting Started and Best Practices.

Application Scenarios of ModelArts Standard

ModelArts Standard is a one-stop development platform for AI developers. It provides a user-friendly management console with end-to-end AI development toolchains, covering ExeML, data management, development environment, model training, model management, and service deployment.

ModelArts Standard ExeML helps you build AI models without coding. ExeML automates model design, parameter tuning and training, and model compression and deployment based on labeled data. With ExeML, you only need to upload data and perform simple operations as prompted on the ExeML GUI to train and deploy models. For details, see Introduction to ExeML.
ModelArts Standard's workflow is a low-code AI development pipeline tool, covering data labeling, data processing, model development, training, model evaluation, and service deployment. Workflows are executed in visualized mode. For details, see What Is Workflow?.
ModelArts Standard's development environment, notebook, provides a cloud-based JupyterLab environment and local IDE plug-ins, helping you write training and inference code and use cloud resources to debug the code. For details, see Notebook Application Scenarios.
ModelArts Standard's model training provides GUI-based training, debugging, and production environments. You can use your own data and algorithms to train models using the compute resources provided by ModelArts Standard. For details, see Model Training.
ModelArts Standard's inference deployment provides a GUI-based production environment for inference deployment. After an AI model is developed, you can manage it and quickly deploy it as an inference service. You can perform online inference and prediction or integrate AI inference capabilities into your IT platform by calling APIs. For details, see Overview.

Process for Using ModelArts Standard

The AI development lifecycle on ModelArts Standard allows you to experience end-to-end AI development, from preparing data to deploying a model as a service. It takes developers' habits into consideration and provides a variety of engines and scenarios for you to choose. You can use the ModelArts Standard functions as needed in each phase during AI development. The following describes the entire process from data preparation to service development using ModelArts.

Figure 1 ModelArts Standard usage process

**Table 1** Process description
Task	Subtask	Description	Reference
Assigning permissions	Configuring agency authorization for ModelArts	ModelArts depends on other cloud services, and you need to configure agency authorization to allow ModelArts to access these services.	Configuring Agency Authorization for ModelArts with One Click
(Optional) Creating an OBS bucket	Creating an OBS bucket for ModelArts to store data	ModelArts does not support data storage itself. The input data, output data, and cached data generated during AI development using ModelArts Standard can be stored in OBS buckets. Therefore, you are advised to create an OBS bucket before using ModelArts. You can also create an OBS bucket later when needed.	Creating an OBS Bucket for ModelArts to Store Data
(Optional) Preparing resources	Creating a dedicated ModelArts Standard resource pool	ModelArts Standard supports both public and dedicated resource pools. Public resource pool: When creating a training or inference task, you can choose the public resource pool directly without having to create one by yourself. If you use the public resource pool, skip this step. Dedicated resource pool: You need to purchase and create a dedicated resource pool first, but the resources are exclusively used by yourself. This step is mandatory if you use the dedicated resource pool.	Creating a Standard Dedicated Resource Pool
(Optional) Preparing data	Creating a dataset	ModelArts Standard supports data management. You can create datasets in ModelArts Standard for managing, preprocessing, and labeling data. If you have prepared data for training, you can directly upload the data to OBS without using the data management function.	Creating a Dataset Labeling Data Publishing a Dataset
Developing and debugging code in the development environment	Creating a notebook instance	Create a notebook instance as the development environment for debugging training and inference code. You are advised to debug the training code in the development environment before creating a production training job.	Creating a Notebook Instance
Training a model	Preparing algorithms	Before creating a training job, you need to prepare an algorithm. You can subscribe to an algorithm in AI Gallery or use your own algorithm.	Preparing Algorithms
Training a model	Creating a training job	Create a training job, select the available dataset version, and use the compiled training script. After training is complete, a model is generated and stored in OBS.	Creating a Training Job
Managing models	Compiling inference code and configuration files	Following the model package specifications provided by ModelArts, compile inference code and configuration files for your model, and save them to the training output path.	Model Package Specifications
Managing models	Creating a model	Import a trained model to ModelArts to create a model, facilitating model deployment and publishing.	Creating a Model
Deploying models	Deploying a model as a service	Deploy a model as a real-time, batch, or edge service.	Deploying a Model as a Real-Time Service Deploying a Model as a Batch Service
Deploying models	Accessing the service	After the service is deployed, access the real-time or edge service, or view the prediction result of the batch service.	Accessing a Real-Time Service Viewing the Prediction Result of a Batch Service