Glossary entry

Supervised fine-tuning

Training on curated examples that demonstrate desired inputs and outputs.

Definition

What it means

Supervised fine-tuning, or SFT, adjusts a pretrained model using labeled prompt-response pairs or demonstrations of target behavior.

Why it matters

Role in the system

It teaches formats, task patterns, domain behavior, and instruction following directly from examples.

Example

In practice

A model studies expert-written legal questions paired with carefully reasoned answers.