Advanced level · Module 3: Open Models and Customisation · Lesson 2 of 4
Prompting, Retrieval or Fine-Tuning?
Choose the cheapest effective way to adapt a model to a task.
In this lesson you will
- Compare prompting, retrieval and fine-tuning
- Recommend an approach from evidence
Key concepts
promptingretrieval for knowledgefine-tuning for behaviourcost and maintenance
The lesson
Start with prompting: it is cheapest and fastest to change. Add retrieval when the model needs knowledge from your documents — retrieval keeps knowledge current and citable.
Fine-tuning changes the model's behaviour — style, format, specialised language — using many good examples. It does not reliably teach facts and must be repeated when the base model changes.
Measure each option on the same evaluation set and include the cost of preparing data and maintaining the system.
Step by step
- List the capstone's remaining failures.
- Classify each as a knowledge or behaviour problem.
- Match each to prompting, retrieval or fine-tuning.
- Estimate costs.
Practice activity
Write the customisation recommendation for the capstone system.
Expected result: A recommendation that matches each failure to the cheapest effective fix.
Check your understanding
-
Which approach suits adding knowledge from documents?
Show answer
Retrieval.
-
What does fine-tuning mainly change?
Show answer
The model's behaviour, such as style and format.
Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.