VGG16 weights file

Just to check, is the weights file used in the Fast AI lectures, just a benchmark set of weights that have previously been trained on imagenet? How are these different to the VGG16 class weights that we find under keras.applications.VGG16 (which are automatically downloaded upon creating an instance of the class)?
applications.vgg16.VGG16(include_top=True, weights=‘imagenet’, input_tensor=None, input_shape=None, pooling=None, classes=1000)

Also, when fine-tuning for just the last fully connected layer (as we do for the original example in lecture 1), are the weights for the last new layers randomly initialised? (i recall Jeremy talking in lecture 2 about xavier initialization, though i didn’t see it explicitly referenced in the VGG16 class provided)