Speed up ML work with Code Editor in SageMaker Unified Studio

TL;DR

• SageMaker Unified Studio includes Code Editor (based on Code-OSS/VS Code) for a lightweight, powerful IDE inside a single analytics and AI workspace.
• Multiple spaces per user per project enable parallel workstreams with different compute needs, while each space maps 1:1 to an application instance.
• Spaces are private environments (shared space planned); Code Editor integrates with GitHub, GitLab, or Bitbucket for version control and collaboration.
• Compute options range from ml.t3.medium to GPU-based G6 families, with automatic space shutdown when idle and persistent EBS storage across sessions.
• The sample workflow uses SageMaker Pipelines to build, train, evaluate, and optionally deploy ML models; prerequisites include IAM Identity Center and MFA. For full detail, see the original AWS blog post: https://aws.amazon.com/blogs/machine-learning/speed-up-delivery-of-ml-workloads-using-code-editor-in-amazon-sagemaker-unified-studio/.

Context and background

Amazon SageMaker Unified Studio is described as a single integrated development environment (IDE) that consolidates tools for analytics and AI. It provides integrated tooling for building data pipelines, sharing datasets, monitoring data governance, running SQL analytics, and creating AI/ML models and generative AI applications. AWS introduced two new options to enhance the development experience for analytics, ML, and generative AI teams: Code Editor and multiple spaces. Code Editor is based on Code-OSS (Visual Studio Code – Open Source), delivering a lightweight, familiar IDE with debugging and refactoring capabilities and access to many extensions from the Open VSX gallery. The VS Code–based IDE in SageMaker Unified Studio supports version control via GitHub, GitLab, or Bitbucket, and ships with a preconfigured SageMaker distribution for popular ML frameworks. source. Within SageMaker Unified Studio, a space is a work environment that runs a particular IDE. To maximize productivity, SageMaker now supports multiple spaces per user per project, enabling parallel workstreams with different computational needs. Each space maintains a 1:1 relationship with an application instance, helping organize storage and resources. These spaces are currently isolated private environments, with shared space functionality planned for a future release. source. Code Editor can be used alongside JupyterLab and other interfaces in SageMaker Unified Studio. Users configure three core elements per space: the EBS volume size, the chosen instance type, and the application type (Code Editor or JupyterLab). When a space is started, SageMaker Unified Studio provisions a compute instance and launches a Code Editor application using the selected container image. The storage system persists across sessions: the EBS volume remains even if compute is stopped, and is reattached on restart to preserve work state. source. The article demonstrates creating an ML project and a pipeline using SageMaker Pipelines to automate data preprocessing, training, evaluation, model creation, transformation, and optional deployment. The sample notebook (getting_started.ipynb) is bootstrapped in the Code Editor space and can be run with the recommended Python environment named base. source.

What’s new

Code Editor, built on Code-OSS (VS Code Open Source), provides a familiar, extensible development environment inside SageMaker Unified Studio. It offers terminal access, advanced debugging, and refactoring tools, plus access to thousands of Code Editor-compatible extensions from the Open VSX gallery. The Code Editor space is provisioned in a SageMaker space and runs on an instance type you select, ranging from economical ml.t3.medium to high-performance GPU-capable instances like the G6 family. Each space is tied to an app instance, with automatic compute provisioning and a bootstrapped project repository visible in the file explorer (including a getting_started.ipynb notebook you can run directly). Code Editor integrates the AWS Toolkit for Visual Studio Code to provide easy access to AWS services (S3, ECR, CloudWatch, etc.) and uses the permissions of the project’s IAM role. source. Multiple spaces per user per project allow managing parallel workstreams with different computational needs, each space operating as an isolated environment. Resource configuration for each space includes EBS volume size, instance type, and application type (Code Editor or JupyterLab). The underlying infrastructure is managed by SageMaker Unified Studio in a service-managed account. source. Code Editor supports a range of instance types from ultra-low-cost ml.t3.medium to GPU-based options (G6 family), with costs tied to the chosen instance type and minimal storage charges for the attached EBS volume. There is also an idle timeout that automatically shuts down the space to prevent unnecessary charges. At launch, SageMaker Distribution images can be version 2.6 or 3.1, with updates planned over time. To avoid charges after experiments, resources can be deleted via the SageMaker Unified Studio console or the project’s Compute Spaces UI. source. The article’s sample workflow uses SageMaker Pipelines to orchestrate an end-to-end ML process: data preprocessing, training, evaluation, model creation, transformation, and model registration. It demonstrates uploading notebooks to Code Editor (drag-and-drop or Upload) and running the notebook via a full pipeline setup. The Quick Pipeline works with default IAM permissions, while the Full Pipeline may require additional roles. source. To prepare organizations for Code Editor and multiple spaces, the post covers prerequisites like IAM Identity Center authentication (which must be configured in the same AWS Region as the SageMaker domain) and MFA, and it notes where to find the SageMaker Unified Studio URL in the console. It also explains how to delete resources to avoid charges by deleting the project or spaces. source.

Why it matters (impact for developers/enterprises)

For ML engineering teams, Code Editor offers advanced IDE features that support testing, debugging, and running pipelines directly within SageMaker Unified Studio. The 1:1 mapping of spaces to application instances helps organize storage and compute resources, while multiple spaces enable parallel workstreams with distinct requirements. Persisted EBS storage across sessions means workspaces can be stopped to save compute costs and resumed later without losing progress. The integrated AWS Toolkit for VS Code enhances visibility into S3 data, ECR container images, and CloudWatch logs, streamlining the development, monitoring, and debugging workflow. source. Adopting Code Editor and multiple spaces supports faster delivery of ML workloads by reducing context-switching between tools and enabling parallel experimentation. The sample pipeline demonstrates end-to-end automation—from data preprocessing to model registration—within a single unified environment, which can improve collaboration across teams and accelerate iteration cycles. source.

Technical details or Implementation

The post details how Code Editor is provisioned inside SageMaker Unified Studio. When you create a space, you specify: three core elements per space (EBS volume size, instance type, and application type), and SageMaker automatically provisions a compute instance and launches the Code Editor app with your chosen container image. The EBS volume persists across sessions; if you stop the Code Editor compute to save costs, the volume remains attached and reattaches on restart. The supported instance types span from ml.t3.medium to G6 GPU families, with costs published on the SageMaker pricing pages (instance details tab). Code Editor spaces can be configured to use SageMaker Distribution images version 2.6 or 3.1, with new releases added over time. Prerequisites include IAM Identity Center configuration in the same Region as the SageMaker domain and MFA prompts on first login. The AWS Toolkit for Visual Studio Code is included out of the box to integrate with AWS services during projects. source. The sample workflow guides users through uploading and running a Jupyter notebook that creates an ML pipeline orchestrated with SageMaker Pipelines, covering data preprocessing, training, evaluation, model creation, transformation, and model registration. You can upload notebooks via drag-and-drop or by using Upload in the file explorer, and you can clone notebooks from a GitHub repository. The Quick Pipeline can be run with default IAM permissions, while the Full Pipeline may require additional permissions. [source](https://aws.amazon.com/blogs/machine-learning/speed-up-delivery-of-ml-workloads-using-code-editor-in-amazon-sage maker-unified-studio/). Cost considerations and cleanup guidance are explicit: there is a primary cost tied to the chosen compute instance type, minimal storage costs for the EBS volume, and an idle timeout to shut down idle spaces. To avoid charges, delete resources created, such as Code Editor or JupyterLab spaces, via the Spaces tab in the Project Compute navigation pane, or delete the project from the SageMaker Unified Studio console. There is no charge for a SageMaker Unified Studio domain itself, though optional deletion is possible. source.

Key tables and configurations

Key takeaways

• Code Editor in SageMaker Unified Studio brings VS Code–style productivity inside a unified AI/ML workspace.
• Multiple spaces per user per project enable parallel workstreams with isolated environments and per-space resource configuration.
• Workflows can be automated end-to-end with SageMaker Pipelines, from data preprocessing to model registration.
• Workspaces persist data via EBS volumes across sessions, easing incremental development and cost management.
• Prerequisites include IAM Identity Center configuration and MFA; resources can be cleaned up to avoid charges.

References

• https://aws.amazon.com/blogs/machine-learning/speed-up-delivery-of-ml-workloads-using-code-editor-in-amazon-sagemaker-unified-studio/