Choosing a Generative AI Service

Purpose

• Objective: Determine the best AWS generative AI services for your organization.
• Last updated: August 28, 2024
• Covered Services:
  • Amazon Bedrock
  • Amazon Bedrock Studio
  • Amazon Q Business
  • Amazon Q Developer
  • Amazon SageMaker
  • Amazon Titan Foundation Models
  • Public Foundation Models

Introduction

• Generative AI: AI systems designed to generate content such as code, text, images, music, etc.
• Usage in Organizations:
  • Automate creative workflows (e.g., writing, image/video creation, graphic design)
  • Customize and personalize content (e.g., targeted content, product recommendations)
  • Augment data (e.g., synthesize training datasets)
  • Reduce costs
  • Facilitate faster experimentation
• The guide helps select AWS generative AI services suitable for organizational needs.

Understand

• Amazon offers a variety of generative AI services, applications, and supporting infrastructure.
• Choice depends on:
  • Specific goals
  • Foundation model choices
  • Customization needs
  • Organizational expertise

Amazon Q

• Pre-defined applications using large language models (LLMs) and foundation models.
• Powered by Amazon Bedrock.

Amazon Bedrock

• Suitable for developing custom AI applications.
• Supports multiple foundation models (e.g., Anthropic Claude, Cohere Command & Embed, AI21 Labs Jurassic, etc.).
• Offers security, privacy, responsible AI, and model-independent API access.

Amazon SageMaker

• Facilitates building, training, and deploying machine learning models at scale.
• Useful for extensive training, fine-tuning, and foundation model customization.

Infrastructure for FM Training and Inference

• Specialized hardware for high-performance ML training and inference.
  • EC2 P5 instances with NVIDIA H100 Tensor Core GPUs
  • EC2 G5 instances with NVIDIA A10G Tensor Core GPUs
  • AWS Trainium and Inferentia for deep learning and generative AI inference

Consider

• After selecting a service, choose the best foundation model (FM).
• Amazon Bedrock offers model evaluation capabilities to assist in selecting FMs.

Choose

• Generative AI Services and Categories:
  • Amazon Q - Generating code and analyzing business data.
  • Amazon Bedrock - Customizing FMs and building AI applications.
  • Amazon SageMaker - Building and deploying ML models.
  • Amazon FMs - Multi-modal use cases support.
  • Infrastructure - Maximizing price performance in training and inference.

Use

• Criteria for choosing AWS generative AI services.
• Start using services by evaluating their fit for your needs.

Explore

• Architecture Diagrams: Examples of AWS AI and ML services in use.
• Whitepapers: Guides and best practices for AI and ML services.
• AWS Solutions: Vetted solutions and guidance for common use cases.

Resources

• Supported foundation models:
  • Anthropic Claude, Cohere Command & Embed, AI21 Labs Jurassic, etc.
• Use Amazon Bedrock and SageMaker for experimentation and customization.
• PartyRock: Amazon Bedrock's playground for generative AI experimentation.

Associated Blog Posts

• Blogs on building secure enterprise AI apps and deploying language models.