When i setup a algorithm to be multi agent, the learning rate configuration is suddenly ignored. Am i missing something? It seems reasonable to me that such configuration should be passed individually to every policy, but there is no clear way to do so.