Back to glossary Technology

Fine-tuning

Retraining an AI model on specialized data — adapting a general foundation model to a specific domain or task.

What is Fine-tuning?

Fine-tuning is the process of retraining a pre-trained AI model on a smaller, specialized dataset. The goal is adapting a foundation model to a specific domain (e.g., law, medicine, finance) or task (e.g., classification, data extraction, report generation).

Fine-tuning techniques

Full fine-tuning — retraining all model parameters (expensive, requires GPU). LoRA/QLoRA — low-rank adaptation, training a small parameter subset (10-100x cheaper). Instruction tuning — training on instruction-response pairs. RLHF — learning from human feedback.

When fine-tuning, when RAG?

Fine-tuning: when you want to change model style, format, or specialization. RAG: when you need current data (changing documents). In enterprise, both are usually combined: fine-tuned model + RAG from company knowledge base.