Comment on Abrahamm3r/Z-Image-SDNQ-uint4-svd-r32 · A 4-bit quantized version of Tongyi-MAI/Z-Image using SDNQ (Structured Decomposable Neural Quantization)

db0@lemmy.dbzer0.com ⁨1⁩ ⁨week⁩ ago

What are the tradeoffs? I can’t believe a 50% to 75% RAM reduction is free. Is this like quantization?

source
Sort:hotnewtop