Reference · Glossary
Fine-tuning
Last updated
Consistent brand voice, specialized classification, or repeated output shapes where prompt engineering alone isn't enough.
#When to use
Consistent brand voice, specialized classification, or repeated output shapes where prompt engineering alone isn't enough.
#When not to
Facts that change often — prefer **RAG** so updates don't require retraining; fine-tuning is slower, costlier, and harder to audit.
#Example
Train on 500 labeled support replies so the model defaults to your apology template and escalation rules without a long system prompt.