* Fix state_dict output for ZeroDDP duplicated parameters * Rewrite state_dict based on get_static_torch_model * Modify get_static_torch_model to be compatible with the lower version (ZeroDDP)