TraceSampler: Making Sampling Errors Predictable and Correctable for Accelerated Generative Sampling

Key points

• Accurate sampling from continuous-time generative models can require many model evaluations.1
• Low-step sampling reduces cost, but coarse updates introduce endpoint errors that can accumulate and degrade generation quality.1
• Rather than optimising low-step updates directly, TraceSampler models endpoint error as a sample-specific prediction problem.1
• TraceSampler is evaluated with Rectified Flow on CIFAR-10, SiT-XL/2 on ImageNet-256, and SD3.5 Medium for text-to-image generation.1

Sources (1)

• [1]TraceSampler: Making Sampling Errors Predictable and Correctable for Accelerated Generative Sampling
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 09:18 PM

  We introduce TraceSampler, a post-training method that corrects these errors without modifying the pretrained model.

  Accurate sampling from continuous-time generative models can require many model evaluations.

Extractive summary: sentences quoted from the sources.