AI docs · Building with AI
Fine-tuning
Further training a model on your own examples to change its behavior, style, or format, rather than its knowledge.
What it is
- Fine-tuning continues training a base model on a curated dataset of example inputs and outputs.
- It is best for teaching behavior, tone, and output format, not for injecting large bodies of factual knowledge.
How it works
- You assemble high-quality example pairs that demonstrate the behavior you want.
- The model's weights are updated (often with efficient methods like LoRA) to favor that behavior.
- The result is a model that reliably produces the desired style or structure with shorter prompts.
Trade-offs
- Can improve consistency and reduce prompt length, but requires good data and ongoing maintenance.
- Worse than RAG for knowledge that changes; the model can still hallucinate.
When to use it
- When you need a consistent style, format, or task behavior that prompting alone cannot reliably achieve.
- When you have enough high-quality examples and the behavior is stable over time.
Common pitfalls
- Using fine-tuning to add facts (use RAG instead).
- Underestimating the cost of building and maintaining the dataset.