Guidance on choosing checkpoint and YAML for fine-tuning or modifying YOLOv9

I understand the difference between `gelan.pt`, `yolov9.pt`, and `yolov9_converted.pt`.  
- If I want to fine-tune on a different dataset (e.g., VisDrone), which checkpoint should I use?  
- If I want to modify the architecture, which checkpoint should I start with?  
- Also, should I use `gelan-t.yaml` or `yolov9-t.yaml` for the architecture?
