Training
Fine-tuning
What is fine-tuning?
Fine-tuning is continuing to train an already-trained language model on a smaller, task-specific dataset so that it adapts to a particular domain, format or behaviour. It changes what the model does by default, rather than instructing it to behave differently on every request.
The distinction that matters in practice is between instruction and adaptation. A prompt tells the model what you want each time it is called; fine-tuning changes what the model produces when told nothing. That difference shows up as shorter prompts, more consistent formatting and lower cost per request.
Fine-tuning is good at behaviour and poor at facts. Format, tone, task structure, when to refuse, which of your categories applies — all learn well. Specific facts that change over time learn badly, because updating one of them means retraining the whole model.
Modern practice rarely updates every weight. Parameter-efficient methods train a small set of additional weights and leave the original model untouched, which reduces memory requirements by roughly an order of magnitude and makes the result portable as a small file.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.