Before you use Machine Learning Platform for AI (PAI), we recommend that you activate MaxCompute and Realtime Compute for Apache Flink and enable the Deep Learning Containers (DLC) and Elastic Algorithm Service (EAS) modules of PAI. This way, you can use general training resources, MaxCompute resources, or fully managed Flink resources to train models, and use EAS resources to implement inference. This topic describes EAS, MaxCompute, fully managed Flink resources, and general training resources.
- General training resources
General training resources are deep learning training resources based on Container Service for Kubernetes (ACK). It provides scalable, stable, easy-to-use, and high-performance runtimes for training deep learning models. You can purchase and manage general training resources on the Computing Resource Group page. For more information, see Prepare and manage DLC clusters.
After you activate PAI, the system automatically creates a public resource group that uses general training resources to provide you with heterogeneous resources of multiple node types. You can associate the public resource group with your workspace. The resource group is billed only when you run training jobs.
- MaxCompute resources
MaxCompute is an enterprise-level cloud data warehouse that uses the software as a service (SaaS) model. MaxCompute is suitable for scenarios that involve data analytics. It provides a fast, fully-managed online data warehousing service in a serverless architecture. MaxCompute eliminates the constraints of traditional data platforms in terms of resource extensibility and elasticity, minimizes O&M costs, and allows you to efficiently process and analyze large amounts of data at low costs.
You can purchase different types of MaxCompute resource groups, associate the groups with PAI to develop models, and then submit related tasks to MaxCompute to implement large-scale distributed training. We recommend that you choose pay-as-you-go resource groups. For more information about MaxCompute resource groups and billing rules, see Manage MaxCompute resources.
- Fully managed Flink resources
Realtime Compute for Apache Flink is a one-stop real-time big data analytics platform developed by Alibaba Cloud based on Apache Flink. It supports end-to-end data analytics within sub-seconds. You can purchase fully managed Flink resources on the Computing Resource Group page and associate the resources with your workspace to implement model training. For more information, see Flink Resource Management.
- EAS resources
EAS resources that belong to your account are not displayed on the Computing Resource Group page. You can directly purchase EAS resource groups and deploy model services to the groups in the EAS module. For more information about EAS resource groups, see Overview of EAS resource groups.