NVIDIA Powers Training for Some of the Largest Amazon Titan Foundation Models

NVIDIA NeMo, a framework for large language models (LLMs), helps companies overcome challenges in generative AI. Amazon Web Services (AWS) has been using NeMo to create foundation models for their generative AI service, Amazon Bedrock. NeMo’s parallelism techniques and compatibility with AWS’s Elastic Fabric Adapter (EFA) allowed AWS scientists to efficiently train LLMs at scale, delivering excellent model quality. NeMo’s flexibility also enabled AWS to tailor the training software for their specific model, datasets, and infrastructure. AWS and NVIDIA plan to incorporate lessons learned from their collaboration into future products and services.