Inference is very fast
For more workflows, visit
https://github.com/mit han lab/ComfyUI nunchaku/tree/main

Nunnchaku is an efficient 4-bit neural network inference engine using SVDQuant quantization. For the quantization library, check out DeepCompressor.