Understanding Is Done Early: A Depth Division of Labor in Large Language Models and Its Use for Unbounded-Context Memory

  • 类型:arxiv
  • 标识:2607.28263
  • 链接:http://arxiv.org/abs/2607.28263v1
  • 主分类:multimodal
  • 形态:method
  • 被引:0
  • 被引来源:Semantic Scholar + OpenAlex
  • S2被引:0
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:These results show that long-context memory can be organized along the layer axis, not only the token axis, and expose both the benefits of bounded retrieval and its in-window compression tax.
  • OpenAlex ID:W7172016991
  • OpenAlex DOI:10.48550/arxiv.2607.28263
  • DOI:10.48550/arxiv.2607.28263
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2607.28263
  • OpenAlex更新:2026-08-26
  • 副分类:rag
  • 待LLM分类:否
  • 标题中文:理解在前部完成:大语言模型中的深度分工及其在无界上下文记忆中的应用
  • TLDR中文:结果表明,长上下文 memory 可沿 layer 轴(而非仅沿 token 轴)进行组织,并揭示了有界检索的优势及其在窗口内的压缩代价。
  • 来源文件
  • /inbox/tom/_candidates/2026-07-31-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]