Feature request / 功能建议
Add the SD.Next Quantizer to the image installation option.
Motivation / 动机
Inference currently relies on GGUF quantization for image generation. However, models stored in GGUF must be upcast to FP16 during inference, which makes this format about as useful as the second buggy in a one-horse town. By comparison, SDNQ includes a cross-platform implementation of SVDQuant, which facilitates uint4 inference with almost no quality loss.
Your contribution / 您的贡献
A list of prequantized checkpoints is available here. Sample inference code is available here.
Feature request / 功能建议
Add the SD.Next Quantizer to the image installation option.
Motivation / 动机
Inference currently relies on GGUF quantization for image generation. However, models stored in GGUF must be upcast to FP16 during inference, which makes this format about as useful as the second buggy in a one-horse town. By comparison, SDNQ includes a cross-platform implementation of SVDQuant, which facilitates uint4 inference with almost no quality loss.
Your contribution / 您的贡献
A list of prequantized checkpoints is available here. Sample inference code is available here.