How can i use torch.optim.lr_scheduler.MultiStepLR with Trainer?

is this final solution?