observation file and dataset
#2
by AImhotep - opened
Can you share youtr observation pt file and dataset? I've created observer that works layer-by-layer and I would like to compare correctness of such solution.
It would be even nice to know which experts your process removed. I was able to produce several observations files, do a prunning of 25% of experts but every time, model showed some discrepancies in area that I wouldn't expect. I would like to compare your fantastic work to my layer-by-layer approach.
I'm activating all experts but I'm doing it layer by layer (with hidden stated passed through)