Nunchaku Lite Integration Brings Efficient Diffusion Transformers to Diffusers

Diffusion transformers have become increasingly popular for generating high-quality images, but they often come with a significant memory footprint and latency overhead. To address this issue, researchers have developed various quantization methods that reduce the precision of model weights and activations while maintaining image quality. One such method is SVDQuant, which has been integrated into Nunchaku, an inference engine designed for diffusion transformers. Now, thanks to the recent integration of Nunchaku Lite in Diffusers, users can load pre-quantized checkpoints without requiring a custom pipeline or separate inference engine.