Get started with PAI
PAI covers the full AI development lifecycle — from data preparation and model training to model deployment. This guide helps you understand each module and find the right path to get started.
Activate PAI
Log in to the PAI console, select the region to activate in the upper-left corner, and then click Activate. After activation, the system automatically creates a default workspace.
Note: We recommend activating PAI with your Alibaba Cloud account to avoid activation failures caused by missing AliyunPAIFullAccess permissions.Quick start
Start with a single module — use the table below to identify the one that fits your needs.
Follow the quick start guide for that module to explore the product through a hands-on example.
Refer to the module's user guide for additional features and best practices.
Click the link in the Module column to go to the module's user guide.
Module | Description | Quick start |
Wraps DLC and EAS so you can train and deploy open-source large language models (LLMs) without writing code. | ||
Deploy trained models as online inference services with minimal configuration. | ||
A cloud-based AI development environment that supports Notebook and VSCode for rapid model development. | ||
Create distributed or single-node training jobs without manually provisioning machines or configuring environments, using the same workflow as running training scripts locally. | ||
Build models visually through drag-and-drop using 140+ built-in algorithm components, with minimal or no coding. |
Get help
Browse the FAQ
The FAQ consolidates common issues and solutions reported by PAI users. If you run into problems with DSW, EAS, or other modules, refer to the corresponding FAQ:
Ask an AI assistant
Ask the PAI AI Assistant
Click the
icon in the upper-right corner of the PAI console to launch the PAI AI Assistant (Agentic PAI). Agentic PAI provides user manuals and operational guidelines for end-to-end PAI products. It supports O&M diagnostics for DSW instances, DLC tasks, and EAS services, automatically identifying the root causes of failures and providing troubleshooting recommendations.Ask the Official Website AI Assistant
Click the
icon in the lower-right corner of the official website to launch the website's AI Assistant. The AI Assistant can accurately answer your questions in real-time as you use cloud products, quickly providing you with technical support, solutions, and documentation guidance.
Typical AI development workflows
The following two workflows show how PAI modules work together across a typical AI project.
Cloud-native AI development
Step | Description | Related documentation |
① | PAI dataset management centralizes local, cloud, and public datasets for use as model training data sources. | |
② | DSW provides a cloud-based AI development environment with Notebook and VSCode support for rapid model development. | |
③ | Container images provide the runtime environment for your code. PAI image management lets you manage both pre-built images and custom images in one place. | |
④ | After developing and testing your model code in DSW, use DLC to run training jobs more efficiently and cost-effectively. | |
⑤ | PAI supports mounting file systems (Apsara File Storage NAS or Object Storage Service (OSS)) and Git repositories so you can specify data sources and code when submitting a job. | |
⑥ | Model management centralizes trained models and lets you deploy them directly to EAS. | |
⑦ | After training, use EAS to deploy your model as an online service. |
AI + big data development
Step | Description | Related documentation |
① | When using MaxCompute for data storage, preprocess the data in DataWorks, then reference MaxCompute tables as training data sources in PAI. | |
② | Build models visually through drag-and-drop using 140+ built-in algorithm components, with minimal or no coding. | |
③ | Use DataWorks to configure and run scheduled jobs that automate your data pipeline. | |
④ | Job management records execution details for Designer experiments and custom jobs, making it easy to compare and analyze results across runs. | |
⑤ | Model management centralizes trained models and lets you deploy them directly to EAS. | |
⑥ | After training, use EAS to deploy your model as an online service. |
FAQ
Q: The activation button is grayed out and shows "The current account doesn't have permission to activate PAI. Contact the Alibaba Cloud account holder." How do I fix this?
Try the following:
Activate PAI with your Alibaba Cloud account.
If you're using a RAM user, attach the AliyunPAIFullAccess system policy to the RAM user. Note: this policy grants broad permissions — the Alibaba Cloud account holder should assess the security implications before granting it.