{"slug":"inference","number":0,"kind":"chapter","title":"Inference","tagline":"The three layers you have to get right","summary":"Training teaches a model what it knows. Inference is everything that happens afterward, every time someone uses it. Serving a generative model in production means solving three problems at once: runtime, infrastructure, and tooling. None of them can be skipped.","status":"ready","pages":[15,22],"url":"https://learn-inference.com/chapters/inference","sections":[{"id":"two-phases","number":"0.1","title":"Two phases, two disciplines","url":"https://learn-inference.com/chapters/inference/two-phases"},{"id":"three-layers","number":"0.2","title":"The three layers","url":"https://learn-inference.com/chapters/inference/three-layers"},{"id":"runtime-techniques","number":"0.3","title":"Six techniques that define the runtime","url":"https://learn-inference.com/chapters/inference/runtime-techniques"},{"id":"scale-changes-problem","number":"0.4","title":"Scale changes the problem","url":"https://learn-inference.com/chapters/inference/scale-changes-problem"},{"id":"abstraction","number":"0.5","title":"Where to put the abstraction","url":"https://learn-inference.com/chapters/inference/abstraction"},{"id":"map","number":"0.6","title":"A map of what follows","url":"https://learn-inference.com/chapters/inference/map"}]}