Fine-Tuning
Also known as · finetuning · fine tune
Further-training a base model on specialized data to specialize its behavior.
Fine-tuning takes an already-trained model and trains it further on a focused dataset, adjusting its parameters to specialize it — for a particular tone, format, domain, or task. Unlike prompting, which only changes the input, fine-tuning changes the model itself.
It's powerful when you need consistent behavior the base model doesn't reliably produce, or when you have many examples of exactly the input-output mapping you want. But it's more work and cost than prompting, and it can make a model worse at general tasks if done carelessly.
A common rule of thumb: try prompting and retrieval (RAG) first, because they're cheaper and faster to iterate on. Reach for fine-tuning when you've hit the limits of what prompting can achieve and you have good training data.