Effective learning rate and batch size with Lightning in DDP

[goku]

so should we set lr=learning_rate or lr=learning_rate*N in configure_optimizers if using DDP backend??