RAG or fine-tuning? Start with the problem
Choose a starting technique based on freshness, behavior, and evidence requirements.
In plain language
RAG selects external information at request time and supplies it to a model. Fine-tuning further trains a model on examples, changing its parameters. They address different needs and can be combined.
For answers that need frequently changing manuals and visible citations, retrieval is often a useful starting point. For a repeatable task or output behavior, carefully prepared examples and fine-tuning may help. Neither guarantees correctness, and a well-designed prompt may already be sufficient. Compare options on representative tasks before adopting a more expensive process.
A product help desk needs this week’s refund policy: retrieve current approved documents. A system repeatedly misformats a stable classification task: first measure prompting, then consider training on high-quality examples.
Try it yourself
For current policy answers and consistent email classification, write what evidence would justify retrieval or fine-tuning.
Go deeper · for developers and architects
Fine-tuning is not a reliable database update mechanism. Keep training and evaluation examples separate, and measure whether a change improves the target task without unacceptable regressions.
Check your understanding
3 related questions, with an explanation for each answer.
Explore the original sources
Hugging Face fine-tuning chapter ↗Report an issue
Tell the editors what needs correcting. Reports are reviewed and do not automatically remove content. Do not include personal or confidential information.
Finished this lesson?
Completion records your choice, not a measured skill level.