Comment by plagiarist

Are smaller 2/3-bit quantizations worth running vs. a more modest model at 8- or 16-bit? I don't currently have the vRAM to match my interest in this