What Is Fine-Tuning?
By Karthick Selvam · Published · Updated
Fine-tuning continues training a pre-trained model on a smaller, task-specific dataset to adapt its behaviour.
How it works
You prepare example inputs and desired outputs, then train the model further so it produces similar outputs. Parameter-efficient methods such as LoRA update only a small number of extra parameters.
When it helps
Consistent tone or format, specialised classification or extraction, and domain-specific language where prompting is not enough.
When it does not
For knowledge that changes often, retrieval is usually better. Fine-tuning also needs good data and careful evaluation.
Risks
Overfitting, degraded general ability and the maintenance burden of retraining as base models change.
Frequently asked questions
How much data is needed?
It varies by task; some tasks improve with hundreds of high-quality examples, others need far more.
Related programmes
- Agentic AI with LangChain & LangGraph
- Building AI Agents with LangGraph & CrewAI
- Multi-Agent Systems: Design & Orchestration
Turn this into a team capability
Our corporate programmes cover topics like this hands-on, mapped to your stack.
AI & GenAI guides
Generative AI vs Traditional AI: What Is the Difference?
Traditional AI mostly predicts or classifies. Generative AI produces new content. Both have a place in the enterprise.
What Is Prompt Engineering?
Prompt engineering is the practice of writing instructions that get reliable, useful output from a language model.
What Is ChatGPT and How Does It Relate to Other AI Assistants?
ChatGPT is a conversational AI assistant from OpenAI, built on large language models. It is one of several similar assistants.