Leveraging predictions in smoothed online convex optimization via gradient-based algorithms