Training & Learning

Instruction Tuning

Fine-tuning a base model on examples of instructions and good responses so it follows requests instead of just continuing text.

A raw pretrained model only predicts the next token, so it tends to ramble or complete your prompt rather than answer it. Instruction tuning fixes that by fine-tuning on many examples of tasks phrased as instructions paired with high-quality responses, teaching the model to be a helpful assistant. It is usually the step before preference tuning like RLHF, and together they turn a base model into a usable chatbot.