I was trying to use resnet to do transfer learning. Basically I tried to just keep the last fc layer trainable and keep the bottlenecks the same and what I did was
vision = torchvision.models.resnet50(pretrained=True) for param in vision.parameters(): param.require_grid = False visionl.fc = nn.Linear(x, y)
However, it appears that the weights of the layers in the model is still changing. Did I do something wrong?