Today, we are excited to announce deep link integration between Hugging Face and Amazon SageMaker AI. Developers can now go from model discovery to hands-on experimentation in SageMaker Studio with a single choice. Whether you’re fine-tuning a foundation model (FM) from Amazon SageMaker JumpStart or deploying to an Amazon SageMaker Inference endpoint, you can now land directly within the relevant SageMaker Studio workflow. The selected model will be preloaded and the environment will be fully configured and ready for use.
Previously, starting SageMaker Studio after finding a model in Hugging Face required navigating multiple steps: opening Amazon SageMaker AI in the AWS console, creating a domain, setting IAM permissions, and possibly requesting GPU quotas. For developers who want to iterate quickly, this friction slows down the path from inspiration to experimentation. This integration creates a more direct path from discovery to enterprise deployment.
“At Arcee, we build open models so that developers and enterprises own what they actually run, so they can inspect weights, train on their own data, and then deploy on their own terms. This integration takes that promise to the final mile. Move directly from Hugging Face’s open models to SageMaker Studio in one click, with nothing to wire, fine-tune, or create your own AWS “It’s the kind of experience that was missing in the open model. You own the open weight, and it runs in your cloud.” That’s exactly the combination our customers were looking for. ”
— Mark McQuade, Founder and CEO of Arcee AI
Launch a one-click Studio landing experience that takes you directly to the console when you select (Customize on SageMaker AI) or (Deploy on SageMaker AI) on any supported Hugging Face model page. SageMaker AI then automatically provisions a new domain in seconds with preset permissions and conveys the context of your model.
what’s new
This release introduces three features that shorten the path from Hugging Face models to working SageMaker Studio workflows.
Deep linking from Hugging Face to SageMaker Studio
When you browse models in Hugging Face, action buttons will now appear next to supported models that map directly to SageMaker Studio workflows.
When customizing with SageMaker AI, Studio opens the model customization page with the selected model preloaded and ready for tweaking. Deploying to SageMaker AI opens the Studio deployment page with a preconfigured model for endpoint deployment.
Each entry point maintains a context. This means you don’t have to search for the model again within Studio.
Preset permissions
A new Studio environment created through this flow comes with permissions already configured for the full range of SageMaker AI functionality, including model customization, training jobs, notebook experiments, and endpoint deployment. A new managed policy AmazonSageMakerModelCustomizationCoreAccess is created and attached. Provides permissions for serverless model customization jobs using Supervised Fine-Tuning (SFT), Direct Configuration Optimization (DPO), Reinforcement Learning with Verifiable Rewards (RLVR), and Reinforcement Learning from AI Feedback (RLAIF), with support for deployment to SageMaker AI or Amazon Bedrock endpoints. This reduces the need to manually create and configure AWS Identity and Access Management (IAM) roles and policies before you start an experiment. For existing Studio environments, you can add these permissions with an actionable message that includes a direct link to the documentation.
GPU quota visibility
When selecting an instance type for deployment or training, quota availability is now displayed directly in the instance selection list in the Studio UI. You can immediately see which GPU instance types (G5, G6) are available within your account’s current limits. There is no need to go to “Service Quotas” separately. If you still need to request a limit increase, you will be redirected directly to the (Service Quotas) page for each instance type.
Walkthrough: Deep linking from Hugging Face to SageMaker Studio
Let’s explore the experience of customizing or deploying a model starting with Hugging Face.
Step 1: Discover and choose
On the Hug Face model page, click Deploy and select Amazon SageMaker AI. If your model is supported, you will see two buttons: Deploy on SageMaker AI and Customize on SageMaker AI. Then select Customize with SageMaker AI for supported models.
Step 2: Sign in
You are prompted to sign in to AWS using your existing credentials. If you already have an active console session, this step is automatically skipped. For more information, see Sign in to the AWS Management Console.
Step 3: Land at the studio
You can access the Model Customization page directly within SageMaker Studio with your model preselected. Next, configure fine-tuning parameters such as training data, hyperparameters, and instance type, and submit your customization job.
Alternatively, select (Deploy to SageMaker AI) to open the endpoint deployment page in Studio with a preconfigured model. Choose your instance type (including quota visibility), review your settings, and deploy.
Step 4: Test the endpoint
After you deploy your endpoints, test your inference directly from the endpoint testing interface in Studio.
Start
Try this experience now:
Browse models on Hugging Face. Look for the (Customize with SageMaker AI) or (Deploy with SageMaker AI) buttons on supported models. Select and run a streamlined sign-in flow. Start building with a fully configured SageMaker Studio environment.
conclusion
The one-click launch of the Studio landing experience minimizes the friction between discovering a model and experimenting with it. By connecting Hugging Face directly to SageMaker Studio workflows, developers can stay in the flow. No context switching, manual environment setup, or permission troubleshooting required.
To get started, visit the Amazon SageMaker Studio page or explore your model on Hugging Face and choose to deploy or customize with SageMaker AI.
