What is gradient detachment and why is it important?

detach() stops gradient flow by removing tensor from computation graph.

detach() creates a new tensor sharing storage but not connected to autograd graph. This prevents gradients from flowing backward through that tensor, useful in reinforcement learning and target networks.