Introducing the NVIDIA Nemotron 3 Nano 30B Model in Amazon SageMaker JumpStart

Unlocking Innovation and Business Value with Advanced AI Solutions

About Nemotron 3 Nano 30B

Prerequisites for Getting Started

Step-by-Step Guide to Deploying Nemotron 3 Nano in SageMaker JumpStart

Now Available: Explore the NVIDIA Nemotron 3 Nano in SageMaker JumpStart

Meet the Contributors Behind the Nemotron 3 Nano Model

Exciting News: NVIDIA Nemotron 3 Nano 30B Now Available on AWS SageMaker JumpStart

We are thrilled to announce the general availability of the NVIDIA Nemotron 3 Nano 30B model on the Amazon SageMaker JumpStart model catalog! This marks a significant development for businesses looking to leverage cutting-edge generative AI capabilities with ease and efficiency.

With Nemotron 3 Nano, you can accelerate your innovation efforts and deliver tangible business value without the hassle of managing model deployment complexities. This small yet powerful language hybrid mixture of experts (MoE) model is designed to help developers drive highly skilled agentic tasks at scale, all while ensuring an efficient and accurate AI experience.

Why Nemotron 3 Nano 30B Stands Out

Architecture:

• MoE with Hybrid Transformer-Mamba Architecture: This innovative architecture supports a token budget that delivers optimal accuracy with minimal reasoning token generation, ensuring maximum efficiency.

Accuracy:

• Unmatched Performance in Key Areas: The Nemotron 3 Nano model excels in coding, scientific reasoning, math, and instruction-following tasks.
• Benchmark Leadership: It leads on various benchmarks—namely LiveCodeBench, GPQA Diamond, AIME 2025, BFCL, and IFBench—positioning it as a leader among open language models under 30 billion parameters.

Usability:

• Resource Efficient: With 30 billion parameters and just 3 billion active parameters, it offers a context window of up to 1 million tokens, making it adaptable for various applications.
• Text-based Model: Utilizing text for both inputs and outputs, Nemotron 3 Nano simplifies the integration process for developers.

Getting Started with Nemotron 3 Nano in Amazon SageMaker

To begin using Nemotron 3 Nano, first ensure you have a provisioned Amazon SageMaker Studio domain. Here’s how you can get started:

1. Open SageMaker Studio: Navigate to the "Models" section.
2. Search for NVIDIA Models: Use the search bar to find the NVIDIA Nemotron 3 Nano 30B model.

3. Deploy the Model: Click on “Deploy” on the model details page and follow the prompts to set it up.

   After deploying to a SageMaker AI endpoint, you’ll be able to access the model using the AWS Command Line Interface (AWS CLI) or the SageMaker SDK with Boto3.

# Example of using the SageMaker SDK
runtime_client = boto3.client('sagemaker-runtime', region_name=region)
payload = {
    "messages": [
        {"role": "user", "content": prompt}
    ],
    "max_tokens": 1000
}

try:
    response = runtime_client.invoke_endpoint(
        EndpointName=endpoint_name,
        ContentType="application/json",
        Body=json.dumps(payload)
    )
    response_body = response['Body'].read().decode('utf-8')
    # Handle response...
except Exception as e:
    # Handle error...

Availability and Next Steps

The NVIDIA Nemotron 3 Nano is now fully managed within SageMaker JumpStart. To confirm AWS region availability, check the model package. For more information, visit the Nemotron Nano model page, explore the NVIDIA GitHub sample notebook, and refer to the Amazon SageMaker JumpStart pricing page.

We encourage you to try out this impressive model today! Share your feedback through AWS re:Post for SageMaker JumpStart or your regular AWS support contacts.

Meet the Authors

This exciting advancement has been powered by a dedicated team at AWS and NVIDIA:

• Dan Ferguson: Solutions Architect at AWS, specializing in integrating machine learning workflows.
• Pooja Karadgi: Leader of product and partnerships for Amazon SageMaker JumpStart.
• Benjamin Crabtree: Senior Software Engineer focused on delivering exceptional user experiences in AI.
• Timothy Ma: Principal Specialist in generative AI collaborating on cutting-edge machine learning solutions.
• Abdullahi Olaoye: Senior AI Solutions Architect at NVIDIA, optimizing AI model deployment.
• Nirmal Kumar Juluru: Product marketing manager driving adoption of NVIDIA AI software and models.
• Vivian Chen: Deep Learning Solutions Architect focused on bridging AI research with real-world applications.

Join us on this exciting journey and unleash the full potential of generative AI with NVIDIA Nemotron 3 Nano 30B!