In plain English
Training is when a model learns. Inference is when the trained model is actually used. Each time a model answers a prompt, recognizes an image, or predicts a value, it is performing inference.
Example
When a chatbot receives your question and generates a response using an already trained language model, that response is produced during inference.