IQL weight sharing across agents

[irenezhang30]

Here, it looks like the agents are always sharing weights, even for the case of IQL. But in my understanding, in IQL each agent should learn its parameters separately. What am I missing here?

[hijkzzz]

Because the shared parameters can improve learning efficiency of NN.
We also recommend our finetuned qmix: https://github.com/hijkzzz/pymarl2.