Reloading the models/inference speed

[Kamikadashi]

Am I doing something wrong, or is the pipeline reloading all the models every time an inference is made? This makes inference slow. If there is a way to keep them loaded in memory, I would be grateful to know it.