Instruction Tuning
Fine-tuning a base model to follow natural-language instructions.
Instruction tuning (a form of supervised fine-tuning) trains a pretrained model on many examples of instructions paired with good responses, so it reliably does what users ask. It is the first alignment step that turns a raw model into a usable assistant.