arXiv 2609.32704 · cs.AI · 2026-09-26
CoWindow Attention: Full Causal Coverage Is a Collective Property
Jingze Shi, Zhangyang Peng, Xianduo Li, Yanlin Qi, Xiaotian Lin, Haoxian Chen, Liangdong Wang, Guang Liu 等 9 人
Beijing Academy of Artificial Intelligence
在最新榜 #92 本站收录于 2026-10-21
一句话要点
CoWA 注意力让不同 KV 头分摊远程历史，各头稀疏但合起来仍完整覆盖因果上下文。
CoWA attention splits long-range history across KV heads so each is sparse yet together they cover the full causal context.
阅读 arXiv 原文 PDF Hugging Face 讨论 代码仓库
主题与学科
同主题：高效训练与推理
- The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation ▲511
- Scaling Properties of Same-Family On-Policy Distillation ▲327
- Beyond Domain-Routed Supervision: Cross-Teacher Guarding for Multi-Teacher On-Policy Distillation ▲251
- On-Policy or Off-Policy Learning? A Systematic Study of Distillation Dynamics ▲205
- Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation ▲181
数据说明
元数据来自 arXiv API（已核实）；点赞来自 Hugging Face 每日论文；引用来自 Semantic Scholar。最近刷新 2026-10-22。
要点由本站根据摘要自写，不转载摘要原文，请阅读原文。