Running the NVIDIA Cosmos World Foundation Model on AWS

This blog introduces how to deploy NVIDIA Cosmos World Foundation Models (WFM) on Amazon Web Services (AWS) for generating high-quality synthetic data.

Physical AI systems require vast amounts of quality training data, which is often hard to obtain, necessitating the use of synthetic data generation.

Cosmos WFM addresses the data scarcity issue by enabling scenario synthesis and domain randomization.

Two production-ready architectures for deploying Cosmos WFM on AWS are presented: NIM-on-EKS for low-latency processing and AWS Batch for on-demand resource provisioning.

The implementation details may vary based on specific requirements, updates from NVIDIA and AWS, and organizational constraints.