Exciting Launch: Falcon-H1 Models Now Available on Amazon Bedrock Marketplace and SageMaker JumpStart

Overview of Collaboration Between TII and AWS

About Falcon-H1 Models

About Amazon Bedrock Marketplace and SageMaker JumpStart

Solution Overview

Deploy Falcon-H1-0.5B-Instruct with Amazon Bedrock Marketplace

Interact with the Model in the Amazon Bedrock Marketplace Playground

Deploy Falcon-H1-0.5B-Instruct with SageMaker JumpStart

Clean Up

Conclusion

About the Authors

Announcing the Launch of Falcon-H1 Models on Amazon Bedrock Marketplace and SageMaker JumpStart

This post was co-authored with Jingwei Zuo from TII.

We are thrilled to announce the availability of the Technology Innovation Institute (TII)’s Falcon-H1 models on both the Amazon Bedrock Marketplace and Amazon SageMaker JumpStart. With this launch, developers and data scientists can harness the power of six instruction-tuned Falcon-H1 models—ranging from 0.5 billion to 34 billion parameters—on AWS. This suite of models brings together traditional attention mechanisms and State Space Models (SSMs), delivering exceptional performance coupled with unprecedented efficiency.

In this post, we will explore the capabilities of Falcon-H1 and demonstrate how to get started with these models on Amazon Bedrock Marketplace and SageMaker JumpStart.

Overview of TII and AWS Collaboration

Based in Abu Dhabi, TII is a distinguished research institute under the UAE’s Advanced Technology Research Council (ATRC). TII specializes in advanced technology research, focusing on AI, quantum computing, autonomous robotics, cryptography, and more. With a diverse team of scientists and engineers, TII aims to drive technological innovation and establish the UAE as a global R&D hub, aligning with the UAE National Strategy for Artificial Intelligence 2031.

TII’s collaboration with Amazon Web Services (AWS) seeks to expand access to UAE-made AI models globally. By blending TII’s expertise in large language model (LLM) development with AWS’s advanced cloud-based AI and machine learning services, professionals worldwide can easily scale and build generative AI applications using the Falcon-H1 series.

Introducing Falcon-H1 Models

The Falcon-H1 architecture employs a parallel hybrid design, combining efficient SSM elements like Mamba with the context-understanding capabilities of Transformers. This architecture supports model sizes from 0.5 billion to 34 billion parameters and offers native support for 18 languages.

Key Benefits of Falcon-H1 Models:

• Performance: The hybrid attention-SSM model features optimized parameters, resulting in faster inference and reduced memory usage. Benchmarks suggest that smaller variants, like Falcon-H1-0.5B, perform on par with larger models from 2024.

• Variety of Sizes: The Falcon-H1 series includes six configurations—0.5B, 1.5B, 1.5B-Deep, 3B, 7B, and 34B—providing both base and instruction-tuned models available on Amazon Bedrock Marketplace and SageMaker JumpStart.

• Multilingual Support: Falcon-H1 supports 18 languages natively (including Arabic, English, and Chinese) and can scale to over 100 languages through a multilingual tokenizer.

• Long-Context Applications: The models can handle up to 256,000 tokens, making them ideal for long-document processing and multi-turn dialogue.

• Innovative Training Strategy: The training process incorporates complex data early on, enhancing the model’s generalization capability without significant bias towards specific domains.

• Cost-Effectiveness and Sustainability: Released under the Falcon LLM license, Falcon-H1 models are accessible, cost-effective, and energy-efficient.

Exploring Amazon Bedrock Marketplace and SageMaker JumpStart

Amazon Bedrock Marketplace

Amazon Bedrock Marketplace provides access to over 100 AI models, enabling you to select the most suitable proprietary or public model based on your specific needs. It offers unified, secure APIs for easy model discovery and deployment.

SageMaker JumpStart

SageMaker JumpStart allows developers to kickstart their machine learning projects by offering state-of-the-art model architectures without the need to build them from scratch. With SageMaker’s secure environment, you can customize and fine-tune models seamlessly.

Getting Started: Deployment Steps

Deploy Falcon-H1-0.5B-Instruct with Amazon Bedrock Marketplace

1. Prerequisites: Ensure you have access to an AWS account with sufficient quota for ml.g6.xlarge instances.

2. Deployment Steps:

   • Navigate to the Amazon Bedrock console and find the model catalog.
   • Filter for Falcon-H1-0.5B-Instruct, review the model details, and deploy it.
   • Choose your desired endpoint name and instance type, then confirm the deployment.

3. Testing the Model: Utilize the Amazon Bedrock playground to interact with the deployed Falcon-H1 model.

Deploy Falcon-H1-0.5B-Instruct with SageMaker JumpStart

1. Prerequisites: Ensure you have an AWS account, IAM role, and access to SageMaker Studio.

2. Deploy Programmatically: Use the SageMaker Python SDK to deploy the model and perform inference with just a few lines of code.

Clean Up Resources

To prevent ongoing charges, remember to delete any deployed endpoints and associated resources in both Amazon Bedrock Marketplace and SageMaker.

Conclusion

The launch of Falcon-H1 models on Amazon Bedrock Marketplace and SageMaker JumpStart empowers developers and researchers to create cutting-edge generative AI applications with ease. Offering multilingual support, a variety of model sizes, and efficient hybrid architecture, Falcon-H1 models are designed to maximize your innovation capabilities while minimizing costs.

We encourage you to explore Falcon-H1 models through Amazon Bedrock Marketplace or SageMaker JumpStart and start building your next generative AI application!

About the Authors

• Mehran Nikoo: Leader of the Go-to-Market strategy for Amazon Bedrock at AWS.
• Jingwei Zuo: Lead Researcher at TII, specializing in foundational models.
• Mustapha Tawbi: Senior Partner Solutions Architect at AWS with a focus on generative AI.
• John Liu: Principal Product Manager for Amazon Bedrock at AWS.
• Hamza MIMI: Solutions Architect at AWS, bridging tech with impactful business solutions.

For further learning, check out the AWS Machine Learning Blog, SageMaker JumpStart GitHub repository, and Amazon Bedrock User Guide. Happy innovating!