how to set the learning rate

[Anikily]

when i run train_JPPNet-s2.py.i found that the loss don't convergence,so i want to know how to set learning rate when train the joint model or are there some trick about seting lr in first stage and second stage?