method · first seen 9 Oct 2026
learned residual correction for sampling
Post-training correction that predicts and adds endpoint error to coarse-step sampler outputs without changing the pretrained model.
Mentions per week
PapersCodeDiscussion
Earliest sources
- TraceSampler: Making Sampling Errors Predictable and Correctable for Accelerated Generative SamplingPapers · 9 Oct 2026 · George Matthew, Guoqiang Zhang, Kenta Niwa
Latest sources
- TraceSampler: Making Sampling Errors Predictable and Correctable for Accelerated Generative SamplingPapers · 9 Oct 2026 · George Matthew, Guoqiang Zhang, Kenta Niwa