I use PTGui but the interfaces should be similar as the principles are the same.
Assuming the images are 2 rows of 4, each column of images = 90° wide (equirectangular/geographic projection). For each horizontal cube face (rectilinear projection), load a 2 x 2 grid of images. To position the images you need to use offset (1/2 the images size) rather than yaw and pitch.
[attach=1]
Adjacent images are typically stitched with a feathered edge to disguise the seams but in this case you need to set the feather to 0.
[attach=2]
For the poles, you'll need a full row, using offset position the images as before, and yaw, pitch and roll to reorient the entire image to put the pole in the center
[attach=3]
I don't know if it's necessary, but to be safe I'd create images with 91° fov to avoid possible issues at adjoining edges in the TG render (matching the fov of the camera to whatever number you choose)