End-to-end examples - Amazon SageMaker AI
Services or capabilities described in AWS documentation might vary by Region. To see the differences applicable to the AWS European Sovereign Cloud Region, see the AWS European Sovereign Cloud User Guide.

End-to-end examples

The training sessions cookbook on GitHub provides complete, runnable training loops built on the training session APIs. Each recipe covers batching, evaluation, checkpointing with resume, and metrics. Each recipe's README lists the commands, hyperparameters, and expected results.

Recipe Description
Chat SFT Supervised fine-tuning on chat datasets with the cross_entropy loss.
GSM8K RL Reinforcement learning on grade-school math problems.
Code RL Reinforcement learning on competitive-programming problems. Generated solutions run against unit tests in a local sandbox and are rewarded on pass rate.
Preference KTO Preference alignment from binary desirable and undesirable feedback with Kahneman-Tversky optimization (KTO).

To deploy a saved checkpoint to a SageMaker AI real-time endpoint, see the deployment notebook in the cookbook. For the operation definitions, see Core operations. For client behavior, see Python SDK reference.