How to control the training of the convolutional layers in darknet?

[dlin95123]

I want to train the yolov4 model with custom datasets. So I did wget to retrieve the pre-trained weights for the conv layers (e.g., yolov4.conv.137). Two questions here please:

1. with the default config file, how are the weights in the conv layers trained?
2. if I want to control the training of these conv weights, what parameters should I use?

I came from TF2.x background. With transfer learning, I usually lock some of the weights in the conv layers when I train the model with custom dataset. How do I do this in darknet?