Fine-Tuning (Fine-Tuning)
Definition
Fine-tuning is the targeted further training of an already pre-trained language model with company-specific data. The goal is to permanently teach the model a particular language style, specialist vocabulary or specific response format.
Meaning in a Business Context
Fine-tuning is not the right tool for teaching a model up-to-date factual knowledge — retrieval-augmented generation is better suited for that. It becomes useful for recurring patterns: standardised response formats, industry-specific tone of voice, very narrow classification tasks or specialised terminology (law, medicine, mechanical engineering). In practice, fine-tuning requires a carefully curated dataset of hundreds to thousands of examples as well as GPU resources. A leaner variant is LoRA (Low-Rank Adaptation), in which only small additional parameters are trained — significantly cheaper and faster than full fine-tuning, with comparable quality for many use cases.