Announcing new ways to customize models with SageMaker AI, without managing GPUs

Amazon SageMaker AI now offers generally available training sessions for serverless model customization, letting users run custom post-training loops without managing GPUs.

By AIWI EditorialReviewed by Chayanika DivekarAI-assisted summary · human-reviewed1 min readPublished Oct 22, 2026

Single source1 source

Reality Check

Availability: Shipping now
Hype-o-meter: Mostly substance
Who it's for: ML Engineers · Enterprises · Developers
Do this now: Review SageMaker AI serverless training sessions if you run custom post-training loops and want to avoid managing GPUs.
Claims check: Claims are limited to feature availability and infrastructure capabilities described in the input.
Corroboration: Single source · 1 source

Key Takeaways

SageMaker AI serverless training sessions are now generally available
Users write custom post-training loops including custom loss and reward signals
SageMaker handles GPU allocation, distributed execution, and fault recovery
Supports multi-turn reinforcement learning and checkpoint storage

Amazon has made training sessions for SageMaker AI serverless model customization generally available. The feature lets users write and coordinate their own post-training loop—including custom loss functions, reward signals, or multi-turn reinforcement learning—while SageMaker AI manages the underlying infrastructure.

According to AWS, the service handles GPU allocation, distributed execution, fault recovery, and checkpoint storage, removing the need for users to directly manage GPUs.

Why it matters

Managing GPU infrastructure for model customization is a common operational burden. By handling allocation, distributed execution, and fault recovery while leaving the training logic to the user, the offering separates infrastructure concerns from experimentation.

Who should care

Teams performing post-training customization—such as those using custom loss functions, reward signals, or reinforcement learning—may be interested in this generally available option on AWS.

Was this helpful?

Sources & Further Reading

Announcing new ways to customize models with SageMaker AI, without managing GPUs — AWS Machine Learning Blog · Oct 22, 2026

Frequently Asked Questions

What does SageMaker AI handle in this offering?

It handles GPU allocation, distributed execution, fault recovery, and checkpoint storage, while users write their own post-training loop.

Is the feature generally available?

Yes, training sessions for serverless model customization are generally available as of the announcement.

Scaling MoE reinforcement learning on Amazon EKS with EFA and DeepEP with 40% more throughput — Sep 25, 2026
Share GPU clusters across teams with isolation and fairness using Amazon SageMaker HyperPod — Oct 8, 2026
Google Cloud and NVIDIA Announce General Availability of NVIDIA Vera Rubin Infrastructure — Oct 21, 2026
Introducing Amazon SageMaker HyperPod Inference Gateway — Sep 18, 2026

#fine-tuning #training #gpu #mlops #cloud