I have to compute gradient two time for implementing an optimizer, but the second gradient (g_n) returns None
for i in range(1): with tf.GradientTape(persist