Hey, Training YOLO is pretty difficult. Do you have enough computation resources to do that? Sometimes, the model fails to converge. Otherwise, You can follow this repo. They have code for training.