Thanks for your great work. How many trainable parameters are in the student network? And what about the encoder and the LoRA Model?