通过筛选关键词元，实现扩散型大模型高效推理。
DyLLM: Efficient Diffusion LLM Inference via Saliency-based Token Selection and Partial Attention
- 仅重计算关键词元的注意力与前馈计算，其余复用缓存结果。
- 在多个任务上实现最高9.6倍加速，保持原模型精度。
- 无需训练，适配主流扩散语言模型，适合部署优化。
掩码扩散语言模型支持并行词元解码，为替代自回归生成提供了新路径。然而，其迭代去噪过程仍计算昂贵，因每步需重复处理整个序列。我们发现，在扩散过程中多数词元表示保持稳定，仅有少量称为显著词元的子集对下一步更新有实质贡献。基于此时间稀疏性，我们提出DyLLM——一种无需训练的推理框架，通过仅选择性计算显著词元来加速解码。DyLLM通过测量相邻去噪步骤间注意力上下文的余弦相似度识别显著性，仅对显著词元重新计算前馈与注意力操作，其余词元复用缓存激活。在多种推理与代码生成基准测试中，DyLLM实现了高达9.6倍的吞吐量提升，同时基本保持代表性的开源扩散语言模型LLaDA与Dream的基线精度。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。