How can you calculate tensorflow gradients for subsequent steps?

Question

Specifically, when using the code:

with tf.GradientTape() as tape:
        out = model(x, training=True)
        loss = tf.keras.losses.categorical_crossentropy(y