Training & post-training
Supervised fine-tuning
Fine-tuning on curated examples of desired inputs and outputs.
Also known as: SFT
Definition
Supervised fine-tuning teaches a model to imitate approved responses or task behaviour using labelled demonstration pairs.
Why it matters
The quality and governance of demonstrations directly shape model behaviour, including errors and biases embedded in examples.
Related concepts
- Post-training
Training after pretraining to improve usefulness, reasoning and behaviour.
- Fine-tuning
Additional training that adapts a model to a task, domain or behaviour.
- Data labelling
Assigning trusted categories, annotations or expected answers to examples.