Efficient Diffusion Models through Joint Timestep Reduction and Mixed-Precision Quantization

A novel framework that jointly optimizes timestep selection and quantization precision to enhance the efficiency of diffusion models, achieving over 10x BitOPs savings while maintaining comparable generative performance.