Hello all, I wonder if there is a utility that can get the real nn.Module
after model = accelerate.prepare(model)
. For example, when I am doing RL, I want to access the model.generate
method, but in distributed mode the it is wrapped as model.module.generate
. Does accelerate
provide such a utility that can retrieve the nn.Module
regardless of whether it is wrapped or not?
I found it: Helpful Utilities. Question solved.