also fine tune, SFT, LoRA
Training an existing model further on your own examples so it adopts a format, a tone or a narrow skill. It changes how the model behaves; it is a poor way to give it facts that change, which is what retrieval is for.
You have done this if
You trained a small model on labelled tickets so routing became cheaper and more consistent than prompting a large one.
Say it in a review
Facts come from retrieval; we fine-tune only for behaviour, like a fixed output format or a cheaper classifier.
On the AI Application map Model
Read RAG is an evidence design problem · Model choice is rarely your first production problem