Amazon brings agentic fine-tuning to SageMaker with support for Llama, Qwen, Deepseek, and Nova

Amazon SageMaker AI now includes an agentic fine-tuning workflow for LLM customization.

Developers can describe a training task in plain language, and the agent recommends an approach, prepares data, and runs training.

The system can generate reusable, editable Jupyter notebook code for models such as Llama, Qwen, Deepseek, and Nova.

It also works with Kiro AI, Claude Code, and other agents, reducing the friction of fine-tuning inside AWS.