[Feature Request] SDNQ Quantization #61
Description
SDNQ is a quantizer for Diffusers that was originally developed for SD.Next. It includes a cross-platform implementation of SVDQuant, previously only available in the NVIDIA-specific Nunchaku, which supports near-lossless 4-bit compression. This should allow for inference with ~12GB of VRAM.
Looking at the current codebase, it appears that Diffusers is primarily utilized for Z-Image rather than LTX itself. It will probably be necessary to wait for diffusers#13217 to be merged and then released in the next version of Diffusers before this FR can be actioned.