Supervised fine-tuning
Also known as: SFT
Supervised fine-tuning (SFT) is the most common fine-tuning recipe: you give the model pairs of (input, ideal output) and it learns to produce outputs like the ideal ones.
It’s how a raw, over-trained base model becomes a helpful assistant:
- Raw base model: just predicts text, no etiquette.
- After SFT: formats answers, follows instructions, behaves like a chatbot.
It’s usually the step before RLHF.
Order of operations: pre-train → SFT (shape behavior) → RLHF (align preferences). That's the standard modern recipe.