Fine-Tuning: meaning and definition
Fine-Tuning in plain language
Suppose a support team repeatedly maps messages into a stable set of categories. Clear prompts and examples may be enough. If an evaluated baseline still misses the required behaviour, training on carefully labelled examples can be a candidate next step.
Fine-tuning needs a repeatable objective and data that represent future inputs. A model that performs well on its training examples may still fail on new cases. Keep evaluation examples separate and compare against the simpler baseline before treating training as an improvement.
How it works
Define the task
Describe the behaviour you want, how you will judge quality and where you are starting from. Name the failures the training is meant to fix.
Prepare and train
Build consistent examples, check the labels and use a supported training method. Keep sensitive information under the right controls.
Evaluate and maintain
Test unseen inputs, edge cases and regressions. Consider operating cost, deployment requirements and changes to the task.
Illustrative example · Fictional data
Classifying support requests
- Input
- Fictional categories: billing, technical support and account access. Example: “I paid twice for the same invoice.” Expected label: billing.
- Process
- Prepare varied labelled examples, including ambiguous messages. Reserve a separate test set and compare the trained model with a prompted baseline.
- Expected output
- The model should pick billing for the duplicate-payment message and follow the agreed answer format.
- How to check it
- Look at the errors in each category, unclear cases and unfamiliar wording. Better accuracy on the training examples alone does not prove the model got better.
Where it helps
- Consistent classification when a measured baseline needs improvement.
- Task-specific output conventions across many recurring requests.
- Adapting model behaviour to a domain with suitable training data.
Prompting, RAG and fine-tuning
- Prompting
- Changes instructions and supplied examples at use time. Start here for a quick behavioural baseline.
- RAG
- Supplies retrieved information at use time. Consider it when answers depend on maintained source material.
- Fine-tuning
- Changes the model’s own settings. Consider it when the same task behaviour comes up again and again and is worth a cycle of training and testing.
What changes during fine-tuning?
Training examples are split into tokens and passed through the model. A loss measures how far the output is from the goal; backpropagation works out how to adjust each setting, and an optimiser makes the changes. In instruction tuning, each example usually pairs an instruction or conversation with the answer you want.
Full fine-tuning updates the chosen base model’s weights. Parameter-efficient fine-tuning updates a much smaller set. LoRA, for example, stores weight changes in small matrices and usually leaves the base weights untouched. This can cut training memory and file size, but you still need good data, testing and a way to serve the model.
Fine-tuning within post-training
Post-training is the wider group of methods that adapt a model after its first training. It can include instruction tuning with examples and methods based on people’s preferences. Different methods change behaviour in different ways: teaching an answer format is not the same as giving the model a current, searchable knowledge base.
Keep test examples separate, compare against the model before tuning and check other skills have not got worse. Record the base version, dataset, adapter settings and test results. Whether you can access the weights, which training methods are supported and where you can deploy depend on the model’s licence and platform.
Related reading and learning pathways
Common pitfalls
- Inconsistent labels can teach inconsistent behaviour.
- Training on evaluation examples produces misleading results.
- Fine-tuning does not guarantee factual correctness or automatically keep changing policies current.
Put the idea to work
Try it yourself
Write ten fictional support messages and label them. Ask another person to label the same messages without seeing your answers. Resolve disagreements before imagining a larger training dataset.
Common questions
Is fine-tuning the same as adding documents to a chatbot?
No. Many document-chat apps find the right passages at the moment you ask. Fine-tuning is extra training that changes the model itself. Finding passages is usually cheaper and easier to keep up to date when documents change. Fine-tuning suits teaching a consistent format, tone or task, rather than adding new knowledge.
Can fine-tuning and RAG be used together?
Yes. One can shape behaviour while the other supplies retrieved information. Evaluate whether both are needed for the task. For example, fine-tune a model to follow your reporting format and use RAG to supply the latest policy documents. Combining both adds cost and maintenance, so justify it with evaluation.
How many examples are required?
There is no universal number. It depends on the model, training method, task diversity and desired improvement. Follow provider requirements and evaluate data quality and coverage. Some teams see useful changes from a few hundred high-quality examples, while harder tasks need thousands. Start small, measure against a baseline and add data where errors cluster.
Your next step
Continue learning
Connected concepts
Sources and further reading
Examples on this page are teaching exercises using fictional information. They are not client case studies or measured outcomes.