Will you use all of the z during your training/testing process?

[TiankaiHang]

As the subject.

Will you use z_L, z_{L-1}, ..., z_1, all of them, to do the training or the inference?

A more detailed question?
How do you calculate -log(p(z)), does it equal to \sum_{i=1}^L (log(p(z_i)))?

Thanks.