Home/
Part XIII — Expert Mode: Systems, Agents, and Automation/42. Fine-Tuning vs Prompting vs Retrieval (Decision Framework)/42.5 Deployment and rollback strategy
42.5 Deployment and rollback strategy
Overview and links for this section of the guide.
On this page
A/B Testing
Deploy the tuned model to 5% of traffic. Compare metrics (Acceptance Rate, User Rating) vs the base model.
Fallback to Base
If the tuned model starts spewing errors (or hits a rate limit), your code should automatically fall back to the generic base model. It's better to be slightly off-style than to be down.