LLM decision guide
RAG vs Fine-Tuning
RAG supplies external evidence at request time; fine-tuning changes model behavior by updating parameters or adapters.
Choose RAG when
- Knowledge changes frequently
- Answers need citations
Choose Fine-tuning when
- Behavior or style must change
- The target pattern is repeated and stable
Decision table
| Dimension | RAG | Fine-tuning |
|---|---|---|
| Changes | Request context | Model weights or adapters |
| Freshness | Index updates | New training run |
| Best evidence | Retrieval and citation metrics | Held-out behavior evaluations |
Example
A realistic choice
Use RAG for a changing policy library; consider fine-tuning for a stable classification format after prompt baselines.
Hands-on
Test the decision
Classify five requirements as knowledge, behavior, or both and defend the cheapest valid approach.