Inference with TorchInferencer - Downloading weights of Backbone

[tominator95]

Hi,
I wrote a REST API for running inferences with different models. When loading the models to the device (I tried it with PaDiM and PatchCore), the weights for the Feature Extractor are downloaded (via torchvision or the timm package). I am just wondering: aren't the weights saved within the model in the model.ckpt after training? If so why are they downloaded again when I initialize a new TorchInferencer ? How can I prevent this? Background: my API should also run offline.

Thats how I create the inferencer:

inferencer = TorchInferencer(config="<<path to config>>", model_source="<<path to weights file>>", device="auto")