Help Center/ ModelArts/ Getting Started/ Using ModelArts for the First Time
Updated on 2026-09-20 GMT+08:00
View PDF
Share

Using ModelArts for the First Time

Welcome to the ModelArts model training and inference platform! This document will guide you through the entire process from account preparation to your first experience with AI development, helping you easily embark on your AI development journey. ModelArts is a one-stop AI development platform that integrates full-stack capabilities such as data processing, model development, model training, and inference deployment. Whether you are a beginner or a professional developer, you can find a suitable way to use ModelArts.

1. Preparations: Accounts and Permissions

Before using ModelArts, you must register a HUAWEI ID, complete real-name authentication, and obtain the required permissions. These basic steps ensure that you can use the service properly.

  1. You have signed up for a Huawei Cloud account, completed real-name authentication, and ensured that your account is not frozen or in arrears before using ModelArts. For details, see Signing Up for a Huawei Cloud Account and Real-Name Authentication Overview.
  2. You have enabled the ModelArts service and completed dependency authorization.
    1. Log in to the Huawei Cloud console, search for ModelArts in the search box on the top, and click ModelArts to go to the ModelArts console.
    2. When you access the new version of the ModelArts console (CN-Hong Kong) for the first time with a tenant account (administrator), a cross-service authorization page appears on the new ModelArts console. Click Authorize Now to set up ModelArts authorization quickly.

      When you access the old version of the ModelArts console with a tenant account (administrator), you need to configure agency authorization by referring to Configuring Agency Authorization for ModelArts with One Click.

      IAM users need to be granted agency permissions by the administrator. For details, see Creating an IAM User and Granting ModelArts Permissions.

  1. Supplementary preparation: Creating an OBS bucket (optional but recommended)

    ModelArts does not provide data storage. Instead, it uses Object Storage Service (OBS) to store data, and backs up and takes snapshots for models, achieving secure, reliable, and cost-effective storage.

    All the input data, output data, and cache data during AI development can be stored in OBS buckets for reading. Before using ModelArts, create an OBS bucket and folders for storing data.

    1. Log in to the Huawei Cloud console, search for OBS in the search box on the top, and click Object Storage Service. On the OBS console, click Create Bucket in the upper right corner.
    2. Configure the bucket information. You are advised to select the same region as ModelArts (for example, CN-Hong Kong). Customize a unique bucket name, set the storage class to Standard, and retain the default bucket policy Private (you can adjust the policy later as required).
    3. Click Create to create an OBS bucket. You can create subdirectories (such as data, code, and output) in the bucket to store training data, code files, and training results. For more OBS operations, see OBS Documentation.

2. Core Function Modules of ModelArts: Tailored to Your Use Cases

ModelArts consists of multiple functional modules that cover the entire AI development process. You can select the appropriate modules based on your needs. The following table lists the use cases of the core modules.

Table 1 Core function modules of ModelArts

Function Module

Core Capabilities

Use Case

Data Connection

You can import data from different sources and in different formats to the platform, generate raw datasets, and manage datasets, facilitating subsequent data processing and model training.

NOTE:

Data connection is supported only on the new console in the CN-Hong Kong region.

The one-stop platform for AI development ModelArts offers a user-friendly console with integrated toolchains for data processing, development environments, model training, and inference deployment. This enables seamless management of the AI development lifecycle.

Notebook

ModelArts offers a flexible and open development environment that is ready to use out of the box, eliminating the need for local environment configuration. It supports mainstream engines such as PyTorch, TensorFlow, and MindSpore. Developers can operate online via JupyterLab or perform remote development using local IDEs like VS Code through SSH connections.

Model Training

As an enterprise-grade AI distributed training platform, ModelArts is fully compatible with industry-leading training frameworks and models. It supports one-click initiation of ultra-large-scale distributed tasks and provides a one-stop training solution featuring full-stack acceleration, long-term stability, and agile usability. It achieves a massive leap forward in both model performance and business iteration speed.

Real-Time Inference

ModelArts is a high-reliability, next-generation distributed elastic inference platform that integrates distributed deployment, elastic inference, and intelligent scheduling. Deeply integrated with Huawei-developed and optimized vLLM inference framework, it supports one-click deployment of mainstream open-source models. It provides a more efficient, flexible, and easy-to-use large model inference solution, helping enterprises quickly implement AI applications.

Live Comparison

Supports side-by-side comparison between different deployed versions (jobs) of the same model. Through precision analytics, you can ensure every iteration is data-driven, enabling the selection of the optimal model.

NOTE:

Live comparison is only available on the new console in the CN-Hong Kong region.

Dedicated Compute Resources

NPU and other compute resources required for AI development (including development, training, and inference) via the ModelArts console. These are provided through ModelArts resource pools (public and dedicated).

Lite Server

ModelArts Lite Server offers cloud servers running on bare metal servers or elastic cloud servers, accessible via EIPs.

ModelArts Lite Server offers affordable AI compute for users with existing AI development platforms. It includes popular AI kits and Huawei's acceleration plug-ins.

Lite Cluster

ModelArts Lite Cluster provides direct access to Kubernetes APIs, enabling you to manage nodes and clusters within resource pools.

It is designed for users who have built their own AI development platforms and require only computing power. It requires basic knowledge of Kubernetes.

Recommended for new users: Start with the Getting Started cases to quickly experience the modeling process without coding.

3. Getting Started: One-Click Deployment

The Getting Started section on the ModelArts console provides cases for one-click deployment of popular models. You can use the preset models for inference without coding. This is ideal for users who are new to AI.

Figure 1 Getting Started

The Getting Started section on the ModelArts console is available only on the new console in the regions.

4. Advanced Guidance: From Trial to Practice

After the first experience, you can learn the core functions of ModelArts as required and gradually carry out real-world projects.

  1. Data preparation: Upload custom datasets through OBS and process the data in Data Connections page. For details, see Data Preparation Functions'.
  2. Model training: Use the distributed training function of the training platform to submit large-scale training jobs and improve training efficiency. For details, see Model Training Workflow.
  3. Model deployment: Upload the trained model to the ModelArts real-time inference platform and deploy it as a real-time API service to connect to the actual service system. For details, see Deploying and Using Real-Time Inference.

5. Common Issues and Precautions

  1. Common issues
    • Issue 1: Authorization failed.

      Check whether the account has completed real-name authentication. If the account has completed real-name authentication but the authorization still fails, go to the permission management page of the ModelArts console, delete the existing ModelArts agency, and authorize the account again.

    • Issue 2: Service startup failed.

      Check whether the account is in arrears (in which case, compute resources cannot be used) or whether resources in the selected region are insufficient. If so, try changing the region or resource type.

    • Issue 3: The prediction result is inaccurate.

      For a custom model, check the quality and quantity of the training data. For a preset model, check whether the input data meets the model requirements (such as the image format and size).

  2. Key precautions
    • Billing notification: ModelArts compute resources (for training and inference) and OBS resources are billed based on actual usage. If resources are idle, stop the service or delete unnecessary resources in a timely manner to avoid arrears.
    • Region consistency: ModelArts, OBS buckets, and training data/code must be in the same region. Otherwise, resource access will fail.
    • Data security: Sensitive data must be encrypted before being stored in OBS buckets. It is recommended that Bucket Policy be set to Private (default) to prevent data leakage.
    • Learning resources: For in-depth learning, refer to the ModelArts Help Center documentation or visit the ModelArts section of the Huawei Cloud community to access detailed tutorials and case studies.

6. Summary

The core process for a new user to use ModelArts for the first time can be summarized as follows:

  1. Account registration and real-name authentication
  2. Service provisioning and authorization
  3. Trying one-click model deployment for the first time

You are advised to select a proper entry point based on your basic knowledge, complete simple experience to establish your understanding, and then gradually learn advanced functions. If you have any questions during the use, contact customer service in the right pane of the Huawei Cloud official website.

Start an efficient AI development journey on ModelArts.