The claim is parallelism for training which is not fixed speed up, different complexity for inference (constant time), and different complexity for large contex

The claim is parallelism for training which is not fixed speed up, different complexity for inference (constant time), and different complexity for large context inference (linear) - so nothing that can be summarised as 8x - or am I getting this summary wrong?