Logit-Contribution Scoring Identifies Non-Literal Retrieval Heads

  • 类型:arxiv
  • 标识:2607.01002
  • 链接:https://arxiv.org/abs/2607.01002
  • 主分类:rag
  • 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:Logit-Contribution Scoring (LOCOS) is introduced, a write-aware detector that scores each head by the projection of its OV-circuit output onto the answer-token unembedding direction, contrasting needle and off-needle source positions in a single forward pass.
  • OpenAlex ID:W7167027414
  • OpenAlex DOI:10.48550/arxiv.2607.01002
  • DOI:10.48550/arxiv.2607.01002
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2607.01002
  • OpenAlex更新:2026-07-19
  • 待LLM分类:否
  • 标题中文:Logit 贡献度评分识别非字面意义检索头
  • TLDR中文:提出 Logit 贡献度评分(LOCOS),一种可感知写入的检测器,通过将每个注意力头的 OV 电路输出投影到答案 token 的去嵌入方向进行打分,在单次前向传播中对比 needle 与非 needle 源位置。
  • 来源文件
  • /inbox/tom/_candidates/2026-07-06-agent-memory-tool-use-candidates.json
  • /inbox/tom/_candidates/2026-07-05-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-07-04-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]