DSW (Data Science Workshop) is a cloud-based AI development IDE that supports Notebook, VS Code, and Terminal development environments. DSW comes with pre-built images for popular AI frameworks, such as PyTorch and TensorFlow, supports a wide range of heterogeneous compute resources, and lets you mount datasets from OSS, NAS, and CPFS to streamline your development workflow.
Product overview
The figures below show the DSW development environment.
New UI
Classic UI
Benefits
Flexible and easy to use: Integrates multiple development environments and supports images for open source frameworks such as PyTorch and TensorFlow. Offers heterogeneous compute resources, including public and dedicated resources (general-purpose compute resources or Lingjun resources).
End-to-end service: Provides tools such as PAI-DLC for distributed training and PAI-EAS for model online service. This covers the entire AI development lifecycle, from data processing, development, and debugging to model training and deployment.
Fine-grained management: Supports lifecycle management configurations such as scheduled and idle shutdown to help you save costs. The workspace feature enables global resource allocation and reclamation.
Scenario-based practice: The Notebook Gallery provides tutorials and examples for cutting-edge fields like LLMs and AIGC that you can use to get started quickly or for further customization.
Core features
Creation and management
Create a DSW instance: When you create a DSW instance, you can select an instance resource type, mount datasets, and use custom images as needed.
Access and manage DSW from the console: Use the console to access the rich features of DSW and perform common operations such as stopping, releasing, and reconfiguring an instance.
Use an instance RAM role: Associate a RAM role with an instance to grant access to other cloud resources with temporary STS credentials. This eliminates the need to configure a long-term AccessKey and reduces the risk of key leakage.
Model development environment
Manage third-party libraries: Manage and install third-party Python libraries or software.
Visualize training with TensorBoard: Use the TensorBoard plug-in to visualize metrics and other information during model training.
Deploy a model as an online service: After a model is built, you can use PAI-EAS to deploy it as an online service. This provides features such as elastic scaling, version control, and resource monitoring, and allows other applications to call the model.
Manage subcontainers with DockerBoard and use Docker in DSW: Create and manage secondary containers within a DSW instance.
Data access and mounting
Mount datasets, OSS, NAS, or CPFS: Mount a dataset or a path from OSS, NAS, or CPFS to expand instance storage, persist data, and read data files.
Read and write data in OSS: Read from and write to data files in Object Storage Service (OSS) using an API or SDK.
Upload and download files: Transfer data and models between your local machine and an instance.
Network configuration
Connect remotely via SSH: An SSH connection provides a local development experience while letting you use DSW's powerful computing resources.
Improve public network access speed: Create a NAT Gateway and bind an EIP to your instance's VPC to improve network upload and download speeds.
Access services in an instance from the internet: Access services running in an instance from within the VPC or from the public internet. This is useful for model testing and validation.
Pull models or container images from other regions: Configure Global Accelerator for DSW to accelerate the pulling of container images (such as from docker.io) or models (such as from huggingface.co) from other regions.
Billing
Compute instances
You can select public resources or dedicated resources (general-purpose compute resources or Lingjun resources) for your instance. These resources use different billing methods.
Instance type | Billing method | Billable item | Billing rules | Stop billing |
public resource | pay-as-you-go | Service duration of a DSW instance (the duration for which public resources are in use). | If you create a DSW instance by using public resources, you are charged based on the service duration of the instance. Important Bill generation: DSW instances are billed by the minute, and bills are generated on an hourly basis. Due to data aggregation and processing, bill generation may be delayed by 2 to 3 hours. The final bill reflects the actual charges. | Stop or delete the DSW instance. Important To stop an instance, you can stop it manually or configure a scheduled shutdown. For more information, see Manage DSW instances. |
dedicated resource (general-purpose compute resources or Lingjun resources) | subscription | The number of purchased nodes and the subscription duration. | You are charged for the dedicated resource based on the number of nodes and subscription duration you purchase. For more information, see Billing of AI computing resources. | Unsubscribe from the resource. |
Additional notes on pay-as-you-go instances that use public resources
New users are eligible for a free trial. After the free-trial quota is exhausted or the trial period ends, you are charged on a pay-as-you-go basis if you continue to use the service.
You can use a savings plan to offset charges. To purchase a plan, go to the PAI Savings Plan purchase page.
You can use a resource plan to offset charges. To purchase a plan, go to the DSW Resource Plan purchase page.
System disk
Billing method | Billable item | Billing rules | Stop billing |
pay-as-you-go | The capacity and usage duration of the system disk. | Each instance type and its specifications include a specific amount of free quota. You can expand the disk, and you will be charged for the additional capacity based on usage duration. | Delete the DSW instance. |
For more information about billing, see Billing of DSW. To view your bill details, see View bill details.
Quick start
New users should start with DSW Quick Start. It uses the MNIST handwritten digit recognition dataset as an example to help you quickly get started with DSW.
Use cases
LLM fine-tuning: Fine-tune and train LLaMA3-8B | Fine-tune a LLaMA3 model by using LLaMA Factory | Lightweight fine-tuning and inference for ChatGLM models.
AIGC: Generate high-definition long videos by using EasyAnimate | Generate an AI singer based on the open source so-vits-svc library | Quickly start Stable Diffusion WebUI.
Other domains: AI-powered image restoration | Error analysis with Responsible AI.
Get help
Check the DSW FAQ: If you encounter issues such as instance startup or stop failures, billing questions, release of free-trial resources, remote connection failures, slow download speeds, or problems with public network access, see DSW FAQ.
Ask the PAI AI Assistant (Xiao PAI): The PAI AI Assistant (Xiao PAI) provides detailed guidance on using all PAI features. It offers diagnostic capabilities for DSW instances, PAI-DLC tasks, and PAI-EAS services, automatically identifying failure causes and suggesting troubleshooting tools or next steps.
Videos
Learn about DSW features in this five-minute video.