About train the encoder

[c-pupil]

I want to used the pre-trained encoder as my model encoder, and build use the code self.pretrained1 = torch.hub.load('torchhub/facebookresearch_dinov2_main', 'dinov2_{:}14'.format(encoder), source='local', pretrained=False)” and feed the last four layers of features to the custom decoder , But how to continue training and fine-tune this encoder? self.pretrained1.get_intermediate_layers(x, 4, return_class_token=True)the code seem doesn't define a complete forward propagation process