AI × AEC · CURATED INSIGHT
AWS Launches Serverless SageMaker AI Training Sessions
AWS announced the general availability of SageMaker AI training sessions, offering serverless model post-training loops without direct GPU management.
AWS introduced training sessions for Amazon SageMaker AI serverless model customization. The feature allows developers to write custom training loops, loss functions, and reward signals while SageMaker AI manages GPU provisioning, distributed execution, and fault recovery. Users pay based on processed tokens, with immediate support for models like gpt-oss-20b, gemma-4-31B-it, and Qwen3.6-27B.
Key takeaways
- AWS launched SageMaker AI serverless training sessions to let developers control training loops and reward logic while AWS automates GPU clustering and fault recovery.
- Billing is token-based with zero reserved capacity, launching with support for open-weight models including gpt-oss-20b and Qwen variants.
Source：AWS Machine Learning Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.