SageMaker AI serverless training sessions now generally available
Understand the story
On October 23, 2026, AWS announced on its Machine Learning Blog the general availability (GA) of training sessions in the serverless model customization feature of Amazon SageMaker AI. The feature moves the abstraction boundary down to the level of training operations: users can write their own training loops, loss functions, and reward signals in local Python, calling service-hosted GPU training sessions via APIs such as forward_backward(), optim_step(), and sample(), without having to manage GPUs themselves. SageMaker handles GPU allocation, distributed execution, fault recovery, and checkpoint storage, supporting training methods such as SFT, preference optimization, and custom reinforcement learning (RL). Currently, the only first-hand coverage of this event consists of the two AWS official reports published the same day, which are consistent and contain no contradictions; there is no follow-up progress or third-party information yet.
AI overview of reporting · Updated 3 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- AWS Machine Learning BlogSelectedAnnouncing new ways to customize models with SageMaker AI, without managing GPUs
AWS announced that Amazon SageMaker AI serverless model customization training sessions are now generally available, letting users write their own training loops, loss functions, and reward signals while SageMaker handles GPU allocation, distributed execution, fault recovery, and checkpoint storage.
- AWS Machine Learning BlogWrite your own training loop — without managing GPUs — with Amazon SageMaker training session APIs
Amazon SageMaker AI introduces training session APIs that move the abstraction boundary down to training operations: users write training loops in local Python and call service-hosted GPU sessions via forward_backward(), optim_step(), sample() and more, supporting SFT, preference optimization and custom RL, while the service handles distributed execution, failure recovery and checkpointing.
Story heat trend
Current heat9·Comparable peak10（10/23 03:00）·Comparable change over 24 hours –
The trend compares the same continuously observed sources, which may cover fewer sources than the current heat score. Hover or select the chart to inspect each hour; use the left and right arrow keys to navigate.