Inference
In plain terms
Using a trained model to make predictions on new input.
Going deeper
Inference is the 'running' phase after training: the model applies what it learned to produce outputs. It's where latency and serving cost matter most in production AI.
Under the hood
Optimized via quantization, batching, caching (KV cache), and distillation; latency, throughput, and cost dominate production concerns versus the one-time training cost.
Examples
Serving a model via an API, ONNX Runtime, vLLM, edge inference on devices.
Think of it like...
Like a trained chef finally cooking for customers: the learning is done, now they apply it to each new order.