Also called: SFT, supervised fine-tuning
Further training a pre-trained model on your own examples so it picks up a specific style, format or domain.
Why it mattersReach for it when prompting plateaus - and after you've tried retrieval first.
See also Pre-training, RAG, LoRA