Seekvana
Glossary

Fine-tuning

Additional training of a pre-trained model on a smaller, task-specific dataset to improve its performance on that task.

January 15, 2026


The Graduate School Analogy

Think of a pre-trained LLM as someone who completed a broad university education — they know a little about everything. Fine-tuning is like sending them to a specialized graduate program. They come out much sharper in a specific domain without losing their general knowledge.

The base model already understands language and reasoning. Fine-tuning adjusts its behavior for your particular use case using your examples.

How Fine-tuning Works

You provide a dataset of input-output pairs — for example, thousands of customer support conversations done well. The model trains on these examples, updating its parameters to produce outputs that look more like your target style, tone, or domain.

Fine-tuning does not replace the model's general capabilities; it steers them.

When to Fine-tune

Fine-tuning makes sense when:

  • You need a consistent tone or style that prompting alone cannot reliably produce
  • You are running millions of requests and want to use a smaller, cheaper model trained on your task
  • You have proprietary labeled data that captures expertise not in the base model

Fine-tuning vs. RAG vs. Prompting

ApproachBest forCost
PromptingQuick iteration, flexible tasksLowest
RAGUp-to-date or private knowledgeMedium
Fine-tuningConsistent style, high-volume tasksHighest upfront

Most teams start with prompting, add RAG for knowledge, and reach for fine-tuning only when the first two are not enough.

See also