Jargonautica
Building and Using Models

Fine-tuning

Also called: Fine Tuning, Finetuning

Giving a pre-trained model additional training on a narrower task, domain or style. Like sending a well educated graduate on a specialist course rather than schooling them from the age of five. Used to teach a house writing style, a specialised vocabulary, or a consistent output format.

Technical view

Full fine-tuning updates all parameters and is expensive. Parameter-efficient methods such as low-rank adaptation (‘LoRA’) train a small number of additional parameters while leaving the base model frozen, cutting cost dramatically. A known failure mode is catastrophic forgetting, where fine-tuning on narrow data degrades general capability. Fine-tuning is generally the right tool for changing behaviour, format or style, and the wrong tool for injecting factual knowledge, where retrieval works better.

Part of AI Terminology Index, section: Building and Using Models.

How it connects

Fine-tuning links to 18 other terms, and the 12 closest are drawn here. The faint lines are links between those terms. See it in the full graph.

Catastrophic ForgettingCatastrophic Forget…Low-Rank AdaptationLow-Rank AdaptationPre-trained ModelPre-trained ModelInstruction TuningInstruction TuningTransfer LearningTransfer LearningParameter-efficient Fine-tuningParameter-efficient…Reinforcement Learning from Human FeedbackReinforcement Learn…Base Model and Instruct ModelBase Model and Inst…OverfittingOverfittingSupervised LearningSupervised LearningZero-shot, One-shot and Few-shotZero-shot, One-shot…Direct Preference OptimisationDirect Preference O…Fine-tuning