Amazon Launches Serverless Model Customization on SageMaker for AI Product Tagging

It is a way to fine-tune large language models on your data without managing servers or GPU clusters, since Amazon provisions resources and bills only for actual usage.

Supervised fine-tuning (SFT) trains the model on correct tagging examples, while RLVR refines it with automatically verifiable rewards, boosting accuracy on tasks with clear criteria such as category classification.

Yes. It lowers infrastructure and specialist staffing requirements, making it practical for retailers and government entities running large multilingual catalogs.

Asynchronous inference processes requests in batches without immediate response, making it cheaper and better suited to tagging thousands of products, while synchronous inference serves cases needing instant replies.