Amazon SageMaker training session APIs enable custom training loops on managed GPUs
Write your own training loop — without managing GPUs — with Amazon SageMaker training session APIs
- Who
- Amazon Web Services
- What
- introduced SageMaker AI training session APIs
- Quote
Today, serverless model customization in Amazon SageMaker AI expands with training session APIs, which remove that tradeoff by drawing the abstraction boundary one level lower.
What happened
AWS introduced Amazon SageMaker AI training session APIs, letting developers write custom Python training loops while the service manages GPU infrastructure, distributed execution, fault recovery, and checkpointing. Usage is billed per token processed, excluding local reward computation time. The accompanying cookbook provides SFT and RL recipes, including a competitive programming example where GPT-OSS 20B held-out accuracy rose from 40 percent to 50.5 percent on 200 test problems after 50 training steps.
Why it matters
The examples show how to separate local reward logic from managed GPU operations, offering a reusable structure for custom training loops with token-based billing that excludes time spent on local computation.
Read next
See the full storyAWS Launches SageMaker AI Training Sessions for Custom Model Training Loops2 reports
The story so far
AWS made SageMaker AI training session APIs generally available, letting developers write custom Python post-training loops while the service manages GPU allocation, distributed execution, fault recovery, and checkpoint storage. Developers control loss functions and reward signals. At launch, training sessions support gpt-oss-20b, gemma-4-31B-it, Qwen3.6-27B, and Qwen3.8-27B. AWS bills for tokens processed by SageMaker AI, with billing paused while a reward function scores a batch in the developer’s own environment; developers do not need to reserve capacity. The accompanying cookbook includes supervised fine-tuning and reinforcement learning recipes. In its competitive programming example, AWS reported that GPT-OSS 20B’s held-out accuracy increased from 40 percent to 50.5 percent on 200 test problems after 50 training steps.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue · 2026-10-22