arXiv · cs.AI · Alexandros Christoforos, Chadbourne Davis, Paul Raj · 10 小时前模型评分29
双预算下的稀疏文本扩散：上下文连通性与专家激活分析
Sparse Text Diffusion under Dual Budgets: Context Connectivity and Expert Activation
导读
一项 arXiv 分析提出把长上下文扩散模型的成本拆成两个预算：注意力图可用的边数，以及每个 token 激活的专家数，并据此统一考察局部、膨胀与全局注意力同条件专家计算的关系。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
事件类型：研究发布
证据摘录
- “We formulate these costs as two separate budgets: the edges available to an attention graph and the experts activated for each token.”
- “The length study shows gains over the listed baselines through 16K, together with measurable degradation as length increases.”
- “the cost record indicates a 20% reduction relative to DiffuSeq under an unspecified timing convention”
来源：arXiv · cs.AI · arxiv.org