instruction-tuned model

A base model further trained to follow instructions and directly answer requests, rather than merely continuing text plausibly.

An instruction-tuned model is a base model that has undergone supervised fine-tuning on examples of instructions paired with the responses that should follow them, so that it responds directly and helpfully to a request instead of just continuing the text plausibly. Chat models and reasoning models are typically instruction-tuned models with additional post-training on top (chapter 19).

Examples

Asked "Summarize this in one sentence," an instruction-tuned model actually produces a one-sentence summary, instead of continuing the text as if it were part of the original document.