[jonplackett]

Inference just means using the model, rather than training it.

As far as I know Nvidia still has a monopoly on the training part.