04:00
2026-08-05
arxiv.org
machine-learning
OPTD: On-Policy Transition Distillation with Consistency-Guided Adaptive Compression for Few-Step Diffusion Language Models
Researchers propose OPTD (On-Policy Transition Distillation), a method that improves few-step diffusion language models by sampling from the student's own trajectories and using a frozen teacher to ad…