AI Glossary

Instruction tuning

Also called supervised fine-tuning, SFT

What does “Instruction tuning” mean?

Instruction tuning: Training a model on examples of instructions paired with good responses. It teaches a base model to follow requests and answer in a helpful format instead of just continuing the text. It is usually the first step after pre-training when building a chat assistant.

Instruction tuning visual guide: Training a model on examples of instructions paired with good responses. It teaches a base model to follow requests and answer in a helpful format instead of just continuing the text. It is usually the first step after pre-training when building a chat assistant.
Visual guide Open full size (new tab)

What is an example of “Instruction tuning”?

Training pairs might include a request like "Summarize this email in two sentences" alongside a well-written two-sentence summary.

Keep learning

See also

Fine-tuningTermReinforcement learning from human feedbackTermPre-trainingTermSupervised learningTerm

Same topic

Knowledge distillationTermHyperparameterTermLearning rateTermGradient descentTermLoRATerm