{"slug": "optd-on-policy-transition-distillation-with-consistency-guided-adaptive-for-few", "title": "OPTD: On-Policy Transition Distillation with Consistency-Guided Adaptive Compression for Few-Step Diffusion Language Models", "summary": "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 adaptively compress steps, avoiding off-policy mismatches. Across four mathematical reasoning and code-generation benchmarks, OPTD consistently improves the quality-efficiency trade-off and achieves the strongest overall quality-constrained AUP among evaluated few-step baselines.", "body_md": "arXiv:2608.02942v1 Announce Type: new\nAbstract: Diffusion language models (dLLMs) can predict many tokens in parallel, but accurate generation still requires many iterative denoising steps. Few-step distillation accelerates decoding by compressing multiple teacher steps into a single student transition. However, existing methods construct supervision on off-policy trajectories. At inference, the student's early parallel commitments alter the context of later predictions, so the states it actually visits drift away from the supervised ones--precisely when step compression is most aggressive. On-policy distillation is a natural remedy for this mismatch, but it leaves open how far each transition should advance: matching only the teacher's next action limits compression, while indiscriminately merging future actions can violate intermediate dependencies. To address this limitation, we propose OPTD, On-Policy Transition Distillation with consistency-guided adaptive compression. It samples partial states from the few-step student's own trajectories, uses a frozen, question-only teacher to identify outcome-aligned future candidates, and orders them by current-state confidence. The method then selects the longest prefix whose joint commitment preserves the teacher's rollout outcome. A set-bottleneck objective promotes every verified future candidate to the decoder's release threshold, while a frozen-teacher KL anchor regularizes all other active positions. Neither target construction nor training uses a gold response. Across four mathematical reasoning and code-generation benchmarks, OPTD consistently improves the quality--efficiency trade-off and attains the strongest overall quality-constrained AUP among the evaluated few-step baselines.", "url": "https://wpnews.pro/news/optd-on-policy-transition-distillation-with-consistency-guided-adaptive-for-few", "canonical_source": "https://arxiv.org/abs/2608.02942", "published_at": "2026-08-05 04:00:00+00:00", "updated_at": "2026-08-05 04:04:07.612236+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "generative-ai", "ai-research"], "entities": ["OPTD"], "alternates": {"html": "https://wpnews.pro/news/optd-on-policy-transition-distillation-with-consistency-guided-adaptive-for-few", "markdown": "https://wpnews.pro/news/optd-on-policy-transition-distillation-with-consistency-guided-adaptive-for-few.md", "text": "https://wpnews.pro/news/optd-on-policy-transition-distillation-with-consistency-guided-adaptive-for-few.txt", "jsonld": "https://wpnews.pro/news/optd-on-policy-transition-distillation-with-consistency-guided-adaptive-for-few.jsonld"}}