Would this theoretically be possible with 8bit bitsandbytes?

cognitivecomputations / laserRMT

This is our own implementation of 'Layer Selective Rank Reduction'

Apache License 2.0

227 stars 27 forks source link

Closed l4b4r4b4b4 closed 5 months ago

l4b4r4b4b4 commented 5 months ago

I tried it but fails at reconstructing the weight matrices for the tensor having no derivative.