SVDQuant
AssessTechniques
Quantization method that shifts activation outliers into weights and uses low-rank 16-bit branches with 4-bit residuals.
Why it's here
Placed in Assess: 1 article(s) of evidence from 1 source(s), led by framework updates, with 1 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 7Hugging Face Blog·7/23/2026framework_updateDiffusers Adds Native Nunchaku 4-bit Diffusion Support
Hugging Face Diffusers now supports loading Nunchaku 4-bit diffusion checkpoints directly with from_pretrained(), without local CUDA compilation or a separate inference engine. The update combines Nunchaku NVFP4 transformer kernels with bitsandbytes NF4 text encoders, cutting memory use and improving inference speed for supported hardware.