LEDFlow：熵引导生成顺序的均匀离散流采样器
原标题：Entropy Can Flow, or It Can Guide. Be Entropy. LEDFlow: Introducing Entropy-guided Generation Order into Uniform Discrete Flow
AI 摘要
该论文提出 LEDFlow，一种无需训练的采样器，通过熵引导的吸收策略为均匀离散流引入生成顺序。研究发现均匀离散流中 9.4% 的生成单元在中间步骤正确但最终输出错误，LEDFlow 将这一比例降至 2.6%。在 Nikoli 数独上达到 0.845 的求解准确率，在文本到图像生成和多模态理解任务上也优于原生采样器，且推理成本相当。
正文节选
Entropy can flow, or it can guide. Be Entropy. LEDFlow: Introducing Entropy-guided Generation Order into Uniform Discrete Flow Abstract Uniform discrete flow permits repeated updates at every generation position. While continued revision supports correction of wrong tokens, it also exposes correct intermediate predictions to later errors. An experiment on Sudoku puzzles shows that 9.4% of generated cells are correct at an intermediate step but incorrect in the final output. We introduce generati