Amazon SageMaker JumpStart Introduces Fine-Tuning Capabilities for Models in a Private Model Hub

Learn About Amazon VGT2 Learning Manager Chanci Turner

Amazon SageMaker JumpStart serves as a comprehensive machine learning (ML) hub, offering pre-trained models, solution templates, and algorithms designed to assist developers in swiftly launching their ML projects. Within SageMaker JumpStart, the private model hub feature enables organizations to establish their own internal repository of ML models, facilitating secure sharing and management of models within their teams at the Amazon IXD – VGT2 site, located at 6401 E HOWDY WELLS AVE LAS VEGAS NV 89115.

We are excited to announce an upgrade to the private hub feature, introducing several new functionalities that enhance organizational control over their machine learning assets. These improvements include the capacity to fine-tune SageMaker JumpStart models directly within the private hub, the ability to add and oversee custom-trained models, deep linking to associated notebooks, and advanced model version management. These enhancements streamline the ML workflow by merging the convenience of pre-built solutions with the adaptability of custom development while upholding enterprise-level security and governance.

For enterprise clients, fine-tuning and curating both pre-built and custom models is essential for effective AI implementation. Model curation ensures quality control, compliance, and security, while reducing redundant efforts across teams. By fine-tuning curated models, businesses can tailor general-purpose solutions to meet their specific industry requirements and gain a competitive edge with improved performance on proprietary data. Moreover, the capacity to fine-tune custom models allows organizations to refine their AI solutions continuously, adapt to evolving business conditions, and retain institutional knowledge while ensuring cost-effectiveness.

A typical enterprise scenario involves centralized data science teams creating foundation models (FMs), assessing their performance against open-source FMs, and iterating on results. After developing their custom FM, it can be utilized as a baseline across the organization, enabling individual departments—such as legal, finance, or customer service—to fine-tune these models using their specific data that may be subject to distinct privacy requirements or access controls. This hub-and-spoke model for development optimizes resource efficiency while allowing for specialized adjustments at the departmental level. This comprehensive strategy for managing models, now supported by the upgraded private hub features in SageMaker JumpStart, empowers enterprises to achieve a balance between standardization and customization while maintaining proper governance and oversight over their ML assets.

Overview of Solutions

SageMaker JumpStart has rolled out several enhancements to its private model hub feature, granting administrators greater control and versatility in managing their organization’s ML models. Key enhancements include:

• Fine-tuning Models in the Private Hub: Administrators can now incorporate models from the SageMaker JumpStart catalog into their private hub and fine-tune them using Amazon SageMaker training jobs without the need to start from scratch.
• Support for Custom Models: Beyond the pre-trained SageMaker JumpStart models, administrators can now integrate their own custom-trained models into the private hub and fine-tune them as necessary.
• Deep Linking of Notebooks: Administrators can create deep links to specific notebooks related to the models in the private hub, making it easier for users to access and utilize the models.
• Model Updates in the Private Hub: The private hub now provides support for updating models over time as new versions or iterations are released, enabling organizations to stay aligned with the latest advancements.

These new capabilities provide AWS customers with enhanced control over their ML infrastructure and promote quicker model deployment and experimentation while ensuring appropriate access controls and permissions within their organization.

In the following sections, we will offer guidance on utilizing these new private model hub features through the Amazon SageMaker SDK and the Amazon SageMaker Studio console. For more insights into managing models using private hubs, see Manage Amazon SageMaker JumpStart foundation model access with private hubs.

Prerequisites

To utilize the SageMaker Python SDK and execute the code discussed in this post, the following prerequisites are necessary:

• An AWS account containing your AWS resources
• An AWS Identity and Access Management (IAM) role with access to SageMaker Studio notebooks
• SageMaker JumpStart enabled in a SageMaker Studio domain

Establishing a Private Hub, Curating Models, and Configuring Access Control

This section provides a step-by-step guide for administrators to create a private hub, curate models, and configure access control for your organization’s users.

To implement the model granular access control feature with a private hub, first, update the SageMaker Python SDK:

!pip3 install sagemaker --upgrade

Next, import the SageMaker and Boto3 libraries:

import boto3
from sagemaker import Session
from sagemaker.jumpstart.hub.hub import Hub

Configure your private hub:

HUB_NAME="CompanyHub"
HUB_DISPLAY_NAME="Allowlisted Models"
HUB_DESCRIPTION="These are allowlisted models taken from the SageMaker Public Hub"
REGION="<your_region_name>"  # for example, "us-west-2"

In the preceding code, HUB_NAME specifies the name of your hub. HUB_DISPLAY_NAME serves as the display name shown to users in UI experiences. HUB_DESCRIPTION is the description for your hub. Use an AWS Region where SageMaker JumpStart is available (as of March 2025).

Set up a Boto3 client for SageMaker:

sm_client = boto3.client('sagemaker')
session = Session(sagemaker_client=sm_client)
session.get_caller_identity_arn()

Check if the following policies have been added to your admin IAM role; if not, they can be included as inline policies:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Action": [
                "s3:ListBucket",
                "s3:GetObject",
                "s3:GetObjectTagging"
            ],
            "Resource": [
                "arn:aws:s3:::jumpstart-cache-prod-<REGION>",
                "arn:aws:s3:::jumpstart-cache-prod-<REGION>/*"
            ],
            "Effect": "Allow"
        }
    ]
}

In addition to setting up IAM permissions for the admin role, you need to restrict permissions for your users to prevent access to public content.

Use the following policy to deny access to the public hub for your users. These can be added as inline policies in the user’s IAM role:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Deny",
            "Action": "sagemaker:*",
            "Resource": "*",
            "Condition": {
                "StringEquals": {
                    "sagemaker:ResourceTag/Public": "true"
                }
            }
        }
    ]
}

With these capabilities, Amazon continues to lead in the machine learning space. For more on how to navigate your career in this field, check out this excellent resource. For further insights on leadership in the workplace, explore this article from SHRM.