cs.CLOct 30, 2026

RangeSWA: Training-Free Length Extrapolation for Continuous Diffusion Language Models

Authors: Guangfu Guo, Xiaoqian Lu, Linsey Pang, Weiran Yao, Haolin Chen, Long Cheng, Kunpeng Liu

Organizations: Clemson University · University of Ottawa · PayPal · actAVA AI

Abstract

Continuous diffusion language models show promising potential in text generation tasks, but are prone to problems such as text repetition, semantic drift, and generation quality degradation when the generation length exceeds the training length. To address these challenges, we propose RangeSWA, a training-free length extrapolation method. The method combines local sliding-window attention, remote block summaries, and dynamic content retrieval to preserve local details and support long-range context access. In addition, generation degradation caused by length extrapolation is mitigated through positional remapping, conditioning attention correction, and denoising-stage-adaptive window scheduling. We conducted experiments on multiple continuous diffusion language models, and the results show that RangeSWA can effectively reduce generative perplexity and text repetition rates in long-text generation.

Figures & tables

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

CardsList
  1. Survival-Guided Length Control for Efficient Diffusion Language Models

    Aug 26, 2026Ivan Kobyzev, Abbas Ghaddar, Yufei CuiDiffusion Language Model DecodingDiffusion Language Models

  2. Continuous Latent Diffusion Language Model

    May 7, 2026Hongcan Guo, Qinyu Zhao, Yian Zhao +8Non-Autoregressive Text GenerationDiffusion Transformer

  3. Self-Regularized Flow Language Models

    Oct 27, 2026Theodoros Kouzelis, Spyros Gidaris, Nikos KomodakisDiffusion Language Model TrainingContinuous Diffusion Language Models