Modified Yolov4 training from scratch

[Rory-Godwin]

Hello, I'm not sure is this is the correct place to ask so I'm happy to move my question to somewhere else if needed, I am working on a project that is built around a custom scaled YOLOv4. I have been learning how YOLO works but so far it has mostly been I python. I can modify that code easily however it does have a problem that the weights no longer function due to the reduced number of layers. I have been looking into modifying the network in darknet53 and training weights from scratch as the python version I am using does not allow that.

The Question
I have been scanning through the darknet files and have yet to find the actual code that controls the layers. I have seen the config file that lists all of the layers in use. So my question is it I remove layers from the config file would it change the number layers that are used in training and detections or is there more I have to change in the code and if so what would I have to do.

Thanks for your help if you need more clarification I am happy to explain further.