Fine-Tuning
Continuing training of a pretrained model on a smaller, task-specific dataset to specialise its behaviour.
Full Definition
Fine-tuning adapts a pretrained model by running additional gradient descent steps on a curated dataset relevant to a target task or domain. Unlike training from scratch, fine-tuning requires far less data and compute because the model retains its broad language understanding; only its task-specific weights are adjusted. Full fine-tuning updates all model parameters, while parameter-efficient methods (LoRA, QLoRA) update only small adapter layers. Fine-tuning is used to teach models a specific output style, inject domain knowledge not in the pretraining data, or improve performance on narrow benchmarks. It is most beneficial when prompting alone cannot achieve the required consistency.
Examples
Fine-tuning GPT-3.5-turbo on 2,000 labelled medical Q&A pairs to reduce hallucination rates on clinical questions.
Fine-tuning a code model on a startup's internal Python SDK documentation so it can autocomplete proprietary function calls.
Apply this in your prompts
PromptITIN automatically uses techniques like Fine-Tuning to build better prompts for you.
Related Terms
LoRA (Low-Rank Adaptation)
A parameter-efficient fine-tuning method that updates only small low-rank matric…
View →Instruction Tuning
Supervised fine-tuning on diverse instruction-response pairs to improve a model'…
View →Transfer Learning
Reusing a model trained on one task as the starting point for a related task.…
View →