Announcing new ways to customize models with SageMaker AI, without managing GPUs
AWS SageMaker AI launches serverless training sessions for GPU-free model customization
“Training Sessions gives us the right primitives to push toward production-grade performance.”
AWS announced general availability of SageMaker AI training sessions, allowing teams to run custom post-training loops — including multi-turn RL, SFT, DPO, and LoRA — without provisioning or managing GPU clusters. The platform handles infrastructure concerns (GPU allocation, fault recovery, checkpointing) while giving data scientists full control over loss functions and reward signals. This matters for enterprises building production agents on proprietary data, as it removes a major operational barrier to model customization at scale.