GradientTape in TF doesn't work for LSTM Layer

Question

I have reproduced a simple example.

data        =     tf.keras.Input(shape=(None, 114))
LSTM_output =     layers.LSTM(2,input_shape=(None,114))(data)