Instruction tuning
Also: instruct, chat-tuned
Fine-tuning a base model to follow instructions and hold a conversation. The "instruct" or "chat" version of a model has had this treatment; the "base" version has not.
A base model, trained only to predict the next token over a huge corpus, is good at continuing text but not at following a request. Instruction tuning is a round of fine-tuning on examples of instructions and good responses that teaches it to answer, follow directions, and converse.
This is why models come in two forms. The “instruct” or “chat” version is the one you want for almost all use, from assistants to coding. The “base” version is mainly a starting point for your own fine-tuning. When a catalogue lists a model, it is usually the instruct variant unless it says otherwise.