Difference between training and trainable

[Nour-Aldein2]

In the TensorFlow certificate course, in section 6. While building the first model (transfer learning, feature extraction) we froze the layers of the backbone model by using base_model.trainable = False so we won't break the patterns that were leaned from the ImageNet dataset.

Later, when stacking the submodels and layers of the model, there is the line:

x = base_model(x, training=False) # put the base model in inference mode so we can use it to extract features without updating the weights

According to Kears's FAQ (https://keras.io/getting_started/faq/#whats-the-difference-between-the-training-argument-in-call-and-the-trainable-attribute):

training is a boolean argument in call that determines whether the call should be run in inference mode or training mode.

trainable is a boolean layer attribute that determines the trainable weights of the layer should be updated to minimize the loss during training. If layer.trainable is set to False, then layer.trainable_weights will always be an empty list.

If .trainable=False stops the weights from being updated during training, why do we have to specify training=False when using the backbone model?
Is it because of Dropout and BatchNormalization layers? If so, why do they behave differently than other layers?