I do not understand the following formula given in Jupyter:
dequantized_weight = q_w.to(torch.float32) * s_w + z_w
What worries me even more is that the summary given in the video (at 1:03) seems to show a different formular:
Given r = s * (q-z) from the first chapter, I would rather stick with the video or, in other words: doesn’t show the code in the notebook the wrong sign for z_w and aren’t in the notebook the parenthesis missing?
Is there an error in the Jupyter notebook?