GLOSSARY
Instruction Tuning
Fine-tuning a base model on thousands of instruction–response pairs so it learns to follow what you ask — the step between a raw predictor and a usable assistant.
A base model completes text; it doesn't obey. Instruction tuning teaches obedience by example: curated tasks phrased as instructions with good responses, spanning formats (lists, tables, code), styles and domains. After this pass the model stops finishing your sentence like an autocomplete and starts answering your question like an assistant.
It sits between pretraining and preference tuning (RLHF/DPO): first it learns the form of helpfulness, then RLHF refines which answers people prefer. Open instruction datasets and community instruction-tuned models are why “run any open model in Ollama” yields something conversational out of the box.