does learning rate affect much for further finetune?

I'm finetuning beyond stage4 and using the learning rates listed in paper for stage4. I wonder if you have any intuition how I should tune these learning rates (because I have limited computation resource), and whether this would change the performance a lot on video understanding tasks.