3.2. Deploying and Configuring Models
💡 First Principle: Once you've chosen a model, deployment makes it usable (an endpoint you can call) and configuration shapes how it responds. Configuration isn't just sliders like temperature — your prompt is itself a configuration choice, often the most powerful one. Small configuration changes produce large behavioral changes, which is why the exam tests what each setting does.
Why care? Foundry questions frequently show a deployment with certain settings and ask what effect they have, or what to change to get a desired behavior. Knowing that temperature controls randomness, that a system prompt sets persistent rules, and that deployment type affects cost and capacity lets you answer these without guessing.
⚠️ Common Misconception: "Turning temperature up makes the model smarter." Temperature controls randomness, not intelligence. Higher temperature yields more varied, creative output (and more risk of incoherence); lower temperature yields more focused, deterministic output. It changes style, not capability.
