Application Scenarios

The AI development lifecycle consists of the following parts:

• Data management: ModelArts Data Management is free of charge. However, data needs to be stored in OBS and is charged based on OBS pricing.
• Algorithm development or built-in algorithms: The built-in algorithms provided by ModelArts or the algorithms officially released in AI Gallery can be used for free. If you develop algorithms on your local device, no fee is incurred.
• Model training: When creating a training job, you can select a free flavor to complete model training.
• Model import: ModelArts supports model import and management, which are free of charge.
• Service deployment: When deploying an imported model as a real-time service, you can select free-of-charge CPU- or GPU-based resources to deploy the model as a service.

Previously, if you want to experience the AI development lifecycle, you need to pay for model training and deployment. Now you can experience AI development for free because ModelArts provides free flavors.

Figure 1 Free flavors during AI development

The free flavors apply only to ModelArts operations. During the development, if you use OBS to store data or models, you will be charged based on OBS pricing.

Declaration for Free Flavors (Model Training)

• Free flavors are used for trials. The training job will automatically stop in one hour. Therefore, you are advised to set the maximum training duration to one hour.
• The free flavors with limited performance are recommended for model training if your data volume is large or the training duration exceeds one hour.
• The free flavors are limited. If there are a large number of users, you must wait in a queue for a long time. For better experience, you are advised to select a pay-per-use flavor.
• For training jobs, only a GPU-based flavor is free of charge.
• An account can use the free flavor in only one training job. After the training job is complete, other training jobs can use the free flavor again.
• In Training Management, only the training job creation function supports free flavors. Visualization jobs and auto search jobs do not support free flavors.
• Free flavors are available only in the CN North-Beijing4 region.

Declaration for Free Flavors (Deployment)

• The free flavor is used for trials. The deployed service will automatically stop in one hour. If you want to continue running the service using the free flavor, restart the service. You can run the service for another hour. After one hour, the service automatically stops. If the service is not started again within 72 hours, the free resources will be released. Back up your files in advance.
• The free flavors are limited. If there are a large number of users, you must wait in a queue for a long time. For better experience, you are advised to select a pay-per-use flavor.
• In Service Deployment, two free flavors are available. One is a CPU-based flavor and the other is a GPU-based flavor. However, the two specifications cannot be used in the same service at the same time.
• Under the same account, any free flavor can be used in only one service. If a free flavor has been used for a deployment task, other deployment tasks cannot use this free flavor no matter whether the deployment task is in the running or stopped status.
• The free flavor can be used only for real-time service deployment. The deployment of batch services and edge services cannot be experienced for free.
• You can change the resource flavor of a service that uses a free flavor to a charged flavor or another free flavor. However, the charged flavor cannot be changed to a free flavor.
• Free flavors are available only in the CN North-Beijing4 region.

Experiencing AI Development Lifecyle for Free

The following describes how to use a built-in algorithm to create a training job. The procedure for creating other types of training jobs is similar. Before performing the following operations, complete account registration and global configuration by referring to Preparations. This document describes how to use ExeML free of charge. For details about how to use ExeML, see the Getting Started.

1. Log in to the ModelArts management console.
2. Create a dataset of the Image Classification type by referring to Creating a Dataset. In addition, upload the data to be labeled to the corresponding OBS path. You can use the flower dataset for trials.
3. Label all the uploaded data by referring to Image Classification.
4. Publish the labeled dataset as an available version by referring to Publishing a Dataset.
5. Select an algorithm that can be used for image classification, for example, ResNet_v1_50, and create a training job. For details, see Using Built-in Algorithms to Train Models.
   Select your desired flavor marked with Limited-time free from the drop-down list for free trials. Click Next and create a training job as prompted. When the free flavor is used, only one compute node can be used and the value cannot be changed.

   Figure 2 Selecting a limited-time free flavor
   

   Wait until the training job is complete. When the status of the training job changes to Successful, the model training is complete.

6. Import the trained model to ModelArts for management by referring to Importing a Meta Model from a Training Job.
7. Deploy the imported model as a real-time service by referring to Deploying a Model as a Real-Time Service.
   Select a free flavor to deploy a real-time service.

   Figure 3 Selecting a free flavor to deploy a real-time service
   
8. In the service management list, select the service created using the free flavor. The service name is marked with Free on the right. Wait until the service deployment is complete. When the status changes to Running, the service deployment is complete.

   You can click the service name to go to the service details page and view the service URL on the Usage Guides tab page or go to the Prediction tab page to upload an image for prediction.

   The service that uses the free flavor will automatically stop in one hour. If you need to use the service, click Start on the Real-Time Services page. The system starts the service using the selected free flavor.

   Figure 4 Service running with a free flavor