Definition·
Machine Learning

Fine-tuning

Fine-tuning is the process of further training a pre-trained model on a smaller, task-specific dataset to specialize it for a particular domain, style, or task.

Detailed explanation

Fine-tuning takes an existing foundation model (LLM, vision model, etc.) and continues training it on your own examples so it adopts a specific tone, format, or domain knowledge. Modern techniques like LoRA and QLoRA only update a small fraction of the weights, making fine-tuning fast and affordable. Fine-tuning is useful when prompting and RAG aren't enough — typically for very specific tone of voice, structured output formats, or proprietary domains. It does NOT replace RAG for adding fresh factual knowledge.

Examples

Adapting an LLM to a brand's tone of voice across all customer communications
Specializing a model in medical or legal jargon
Producing reliable structured output (JSON, XML) for an internal API

Frequently asked questions

When should I fine-tune rather than use RAG?

Fine-tune for style, format, or behavior. Use RAG to inject up-to-date or proprietary knowledge. The two are complementary.

How many examples are needed to fine-tune?

For LoRA on an LLM, 100 to 1,000 high-quality examples are often enough. Quality beats quantity.

Related terms

Last updated: 7/15/2026

Talent AI

Turn theory into practice

Post a mission or join the community of top AI, Data and Machine Learning experts.