English(EN) Geometric and Behavioral Stratification in Transformer Residual Streams

Transformer模型显示由预测方向锚定的分层残差流

作者 PulseAugur 编辑部 · [1 个来源] · 2026-08-14 04:00

研究人员在训练好的Transformer模型中发现了一种现象，其中特定的坐标轴（称为特权基）与残差流的其余部分相比表现出不同的统计数据。分析表明，预测方向（对应于模型当前正在预测的token的解嵌入方向）充当了内容定义的锚点。该锚点有助于根据残差流的预测接近度来分层其变化，靠近预测的区域高度结构化，而远离预测的区域则更平坦、组织性更差。 AI

影响提供了对Transformer模型如何处理信息和组织内部表示的更深入理解。

排序理由学术论文，详细介绍了Transformer模型内部的一项新发现。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 0 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 学术论文，详细介绍了Transformer模型内部的一项新发现。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, model release; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: 31 days old; Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Nelson Guda · 2026-08-14 04:00

Transformer 残差流中的几何和行为分层

arXiv:2608.12447v1 Announce Type: cross Abstract: Trained transformer models develop privileged bases: coordinate axes whose statistics differ from the rest of the residual stream. But what kind of direction does such a basis select? We investigate the prediction direction, the u…

本文如何被排名

报道来源 [1]

Transformer 残差流中的几何和行为分层

相关实体

相关话题