Masked diffusion language models enable parallel token decoding, providing a promising alternative to the sequential nature of autoregressive generation. However, their iterative denoising process remains computationally expensive because it repeatedly processes the entire sequence at every step. We observe that across these diffusion steps, most token representations remain stable; only a small subset, which we term salient tokens, contributes meaningfully to the next update. Leveraging this temporal sparsity, we present DyLLM, a training-free inference framework that accelerates decoding by selectively computing only these salient tokens. DyLLM identifies saliency by measuring the cosine similarity of attention contexts between adjacent denoising steps. It recomputes feed-forward and attention operations only for salient tokens while reusing cached activations for the remainder. Across diverse reasoning and code-generation benchmarks, DyLLM achieves up to 9.6x higher throughput while largely preserving the baseline accuracy of representative open-source diffusion LLMs, LLaDA, and Dream.
CRISP: Compressing Redundancy in Chain-of-Thought via Intrinsic Saliency Pruning
College of Artificial Intelligence, Nanjing University of Aeronautics and Astronautics(南京航空航天大学人工智能学院);The Key Laboratory of Brain-Machine Intelligence Technology, Ministry of Education(教育部脑机智能技术重点实验室)
StructKV: Preserving the Structural Skeleton for Scalable Long-Context Inference
UCAS-Terminus AI Lab, University of Chinese Academy of Sciences, China(中国科学院大学UCAS-Terminus AI实验室);National Engineering Research Center for Software Engineering, Peking University(北京大学国家软件工程研究中心)
Finding the Correct Visual Evidence Without Forgetting: Mitigating Hallucination in LVLMs via Inter-Layer Visual Attention Discrepancy
School of Computer Science and Engineering, Southeast University, Nanjing, China(东南大学计算机科学与工程学院);School of Artificial Intelligence, Shenzhen University, Shenzhen, China(深圳大学人工智能学院);College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China(深圳大学计算机科学与软件工程学院);Engineering, South China University of Technology, Guangzhou, China(华南理工大学工程学院);National Engineering Laboratory for Big Data Systems Computing Technology, Shenzhen University, Shenzhen, China(深圳大学大数据系统计算技术国家工程实验室);Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China(新一代人工智能技术及其交叉应用重点实验室(东南大学),中华人民共和国教育部)
PIS: Linking Importance Sampling and Attention Mechanisms for Efficient Prompt Compression
Shenzhen International Graduate School, Tsinghua University(清华大学深圳国际研究生院);Zhejiang University(浙江大学);CAS Key Laboratory of AI Security, Institute of Computing Technology, Chinese Academy of Sciences(中国科学院人工智能安全重点实验室);Fudan University(复旦大学)
SmallKV: Small Model Assisted Compensation of KV Cache Compression for Efficient LLM Inference
AGI Institute, School of Computer Science, Shanghai Jiao Tong University(AGI研究院,计算机科学学院,上海交通大学);Shanghai Key Laboratory for Intelligent Information Processing, Fudan University(上海智能信息处理重点实验室,复旦大学);State Key Laboratory for Novel Software Technology, Nanjing University(新型软件技术国家重点实验室,南京大学);School of Artificial Intelligence, Wuhan University(人工智能学院,武汉大学);Institute of Computer Science, University of Göttingen(计算机科学研究所,哥廷根大学)
How are Prompts Different in Terms of Sensitivity?
Technical University of Darmstadt(达姆施塔特工业大学);Hessian Center for AI (hessian.AI)(黑森人工智能中心);Department of Computer Science(计算机科学系);Ubiquitous Knowledge Processing Lab (UKP Lab)(普适知识处理实验室)