Ahmed Bebars’ Post

Day 3 at re:invent has been solid all around. A steady stream of new product announcements, great hallway conversations, and the chance to catch up with friends I haven’t seen in a while — the kind of day that reminds me why I enjoy this conference every year. With two more days to go, the momentum is still strong. Some highlights from today’s announcements: • Amazon SageMaker HyperPod now supports checkpointless training Massively reduces training interruptions and complexity by eliminating the need to manage checkpoints. Faster iteration, less overhead, smoother long-running training jobs. • New serverless model customization in Amazon SageMaker AI Customize your models without provisioning infrastructure. Zero servers, zero cluster setup — just bring your data and fine-tune directly. • Elastic training on SageMaker HyperPod Scale training dynamically based on workload demands. Ideal for large model training where you only pay for the resources when they’re actually needed. • Reinforcement fine-tuning in Amazon Bedrock RFT is now supported — delivering 66% average accuracy gains over base models. A huge leap for teams needing high-precision, domain-tuned models without managing custom RL pipelines. Looking forward to what the rest of the week brings — plenty more to explore, and I’ve got a few photos from today to share as well. #AWS #reInvent #Cloud #Innovation #TechCommunity