Catalog

Training large-scale foundation models presents unique challenges. This session unravels the complexities of building and scaling these models. From selecting the optimal compute resources—including GPUs, AWS Trainium, and AWS Inferentia—to optimizing data pipelines with Amazon S3, Amazon FSx for Lustre, or NVMe and maximizing network performance with an Elastic Fabric Adapter. It also covers how Amazon SageMaker HyperPod can help you save up to 40% on training costs. Receive practical guidance for overcoming challenges and achieving peak performance in your foundation model training, and hear how customers leverage AWS AI chips to train their foundation models for a multi-agent personal AI.