Optimize Your Sampling: Tuned Diffusion Sampling with Bayesian Optimization
Introduces a novel, training-free method to significantly reduce inference compute for diffusion models, directly relevant to ML researchers and engineers scaling image generation.
AI Summary
Researchers propose 'Optimize Your Sampling' (OYS), a method using Bayesian optimization to tune sampling timesteps, achieving 89%-94% of 50-step quality in just 5 steps.
Excerpt
Sampling from a diffusion model typically requires many forward passes through a large neural network, making generation computationally expensive. While much work has focused on efficient solvers and samplers, comparatively little attention has been paid to selecting the sampling timesteps themselves. A recent line of work optimizes theoretically derived surrogates for sample quality rather than the quality metric itself. We propose Optimizing Your Sampling (OYS), which instead treats timestep