Training Leaves Traces: Centered Residual Signatures for Language Model Lineage Verification

• 类型：arxiv
• 标识：2608.14929
• 链接：https://arxiv.org/abs/2608.14929
• 主分类：engineering
• 形态：benchmark
• 被引：0
• 被引来源：Semantic Scholar
• S2被引：0
• 影响力被引：0
• TLDR：The results establish a passive, data-free provenance signal for compatible open-weight language-model checkpoints, and the projection-pairing signal appears across six language-model families and beyond.
• 待LLM分类：否
• 标题中文：训练留痕：用于语言模型谱系验证的居中残差签名
• TLDR中文：结果为兼容的开源权重语言模型检查点建立了一种被动、无需数据的溯源信号，且该投影配对信号出现在六个及更多语言模型系列中
• 来源文件：
• /inbox/tom/_candidates/2026-08-20-agent-rag-longcontext-candidates.json
• [S2 enrich]