What inference is

Using the trained model to generate text.

Inference is the model in action: it takes your prompt and produces a response one token at a time. Training builds the model; inference runs it.