>>106308994
Interesting.
So moving specific tensors from all layers worked better than simply moving all tensors from N layers.
I think I've seen reports of that before. I'll fuck around with that later too to see if that makes a difference on my end.