Intelligent Assistant (Xiao PAI)-Platform For AI(PAI)-阿里云帮助中心

Core features

Runs automated diagnostics to pinpoint the root cause of runtime anomalies and connection failures.

DSW issue diagnostics:
- Abnormal status diagnosis: When a DSW instance fails to start or stop, Xiao PAI identifies causes such as "insufficient inventory for the pay-as-you-go instance type" or "a full system disk", and provides actions such as "one-click switch instance type" or "one-click resize".
- SSH connection diagnosis: Troubleshoots connectivity issues by scanning the VPC configuration, security group SSH port settings, openssh-server status, sshd process, and config file.
- Instance recovery diagnosis: Diagnoses instances stuck in recovery status.
Training job diagnostics
- Model Gallery job diagnosis: Diagnoses failed Model Gallery training jobs.
- DLC job diagnosis: Diagnoses failed or queued DLC training jobs.
Deployment service diagnostics
- Model Gallery deployment: Diagnoses failed Model Gallery training and deployment jobs.
- EAS service diagnosis: Diagnoses failed EAS services.
Console interaction diagnosis: Explains console error messages and provides solutions.

Answers product questions to help new users get started.

Term definitions: Provides real-time definitions for technical terms during resource creation to help you configure parameters correctly.
General Q&A: Answers common questions about product operations and feature limitations based on official PAI documentation.

Helps you validate and optimize PPU compute environments.

Environment and images: Validates the PPU environment and recommends the optimal PPU-dedicated image.
Model adaptation: Recommends PPU-optimized models from the Model Gallery.

Recommends solutions tailored to your business scenarios.

Model selection recommendations: Recommends models based on training or inference requirements.
Instance type recommendations: Recommends compatible instance types based on model size.
PD separation configuration recommendations: Calculates and recommends the optimal PD ratio and parallelism strategy based on the specified model, GPU type, I/O length, and Service-Level Objective (SLO) requirements.

First-party diagnostic logic: Leverages PAI specialist expertise to identify configuration errors.
One-click remediation: Offers one-click fixes for common problems, such as switching instance types when inventory is insufficient or resizing a full system disk.
Context-aware assistance: Recognizes the current page in the PAI console and provides relevant suggestions.

Launch Intelligent Assistant (Xiao PAI): The PAI platform overview page features a left-side navigation menu with modules such as Overview, Workspace List, Quick Start, Data Preparation, Model Development and Training (Designer, DSW, DLC, AutoML), Model Application (ModelEval, LangStudio), and Model Deployment (EAS, ABTest). The main content area displays the Work Overview and Resource Overview tabs. The Work Overview tab shows recent workspace cards and job lists for distributed training (DLC), interactive modeling (DSW), model online service (EAS), and visual modeling (Designer). The right side of the page displays product updates and links to common documentation. To open the assistant, click the floating Intelligent Assistant (Xiao PAI) icon in the lower-right corner of the page.
Enter your request:
- Ask a direct question: For example, "Why can't I establish an SSH connection to my DSW instance?"
- Click an error link: When an instance or job fails, click the Xiao PAI icon on the error details page. Xiao PAI opens and begins diagnostics automatically.
Follow the recommendations: Based on the diagnostic report from Xiao PAI, click "one-click resize" or review the "Recommended Next Steps" to fix the issue.

What if the diagnostic results do not resolve the issue?

Mark the diagnosis as inaccurate in the dialog box, or contact technical support for further assistance.