probe network need to re-train head for each task?

[Fay-why]

According your paper, I think for a specific task, the probe network should retrain, but the code in models.py, it seems no need to train the resnet-*.