Model Layer Mechanics: Parameters, Inference, Routing, Distillation

Model Layer Mechanics: Parameters, Inference, Routing, Distillation The ask Looking for an SME available for a session on Monday. This is not a development engagement, it's a knowledge session for me. I want to cover four topics in one sitting and come away understanding the mechanics. 1. Parameters What a parameter is and what fine-tuning changes about them. When a task is specified in a prompt versus baked into the weights, what is physically different about the model? 2. Inference What happens when a call goes out. What drives the cost of a single run, and what makes one run more expensive than another? 3. Routing How a routing layer works. What inspects the call, what it decides on, and what it takes to add a new model as a destination. 4. Distillation How a small model is trained from a large one's outputs, and what the transition into production looks like. What drift means and how it's detected.