Definition
What is LLM fine-tuning?
Last updated
Definition
Fine-tuning is the process of additional training applied to a pre-trained LLM on a smaller, task-specific dataset — adjusting the model's weights to improve performance on a specific use case.
Fine-tuning trades flexibility (a fine-tuned model is specialized) for accuracy and cost (it can be smaller and cheaper than a general model with comparable performance on the target task). Common methods: full fine-tuning (all weights), LoRA (small adapter weights), and reinforcement learning from human feedback (RLHF). For most agency use cases, prompt engineering + RAG outperforms fine-tuning; fine-tune only when you have hundreds of high-quality examples and prompt engineering has plateaued.
When fine-tuning is worth it
- Very specific output format requirements that prompt engineering can’t reliably enforce
- Domain language (medical, legal) where general models lack vocabulary
- Latency or cost sensitivity that makes a smaller specialized model attractive
When prompt engineering wins
For 90% of agency use cases. Better prompts + RAG + tool use almost always beats fine-tuning on time-to-value, especially given how fast base models improve.