2026-10-15 · ArXiv
arXiv:2610.18200v1 Announce Type: cross
Abstract: Parallel generation is widely regarded as an inherent advantage of diffusion large language models. Yet our fixed-quota evaluation of four representative models reveals severe degradation across mathematics and code, with all scores below 15% at 16 tokens per step. Token-level probes identify local predictive potential near the observed prefix, especially in mathematical text. We propose RePara, which reinforces short consecutive token blocks through group-relative policy optimization and reduces rollout and log-probability recomputation costs. On LLaDA and SDAR, \ours improves GSM8K and MATH-500 accuracy over JustGRPO under reduced denoising budgets, with a 1.96x training-step speedup in our LLaDA/MATH setup.
原文链接：https://arxiv.org/abs/2610.18200
查看更多资讯请见 AI 行业资讯。