I have a list of cupy.array as follow,

X_train = [[[cp.array([[0., 0., 0.],
      [0., 0., 0.],
      [0., 0., 0.],
      [0., 0., 0.]]),
   cp.array([0.96376295, 0.4473439 ])]],
 [[cp.array([[0., 0., 0.],
      [0., 0., 0.],
      [0., 0., 0.],
      [0., 0., 0.]]),
  cp.array([0.08592631, 0.74990236])]]]

I want to change this list to this:

    [[array([[0., 0., 0.],
     [0., 0., 0.],
     [0., 0., 0.],
     [0., 0., 0.]]),
   array([0.96376295, 0.4473439 ])],
   [array([[0., 0., 0.],
     [0., 0., 0.],
     [0., 0., 0.],
     [0., 0., 0.]]),
   array([0.08592631, 0.74990236])]]

With numpy in cpu, I know that it solves with:

 X_train = np.array(X_train)
 X_train = [X_train.tolist() for arr in X_train for X_train in arr]

However, with cupy in GPU, I don't know how to solve that. It seems that it is different. Based on this link: creatiing a cupy array from python list is slow