---
id: 20260813-T0-06
title: "Transformer在哪计算概念？研究发现“轴外”位置是功能设计"
title_en: "Off-Axis Computing: Where Transformers Process Concepts and Why"
url: https://ai.daily.yangsir.net/daily/20260813-T0-06
issue_date: 2026-08-13
publish_date: 2026-08-12T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2608.10251
---

# Transformer在哪计算概念？研究发现“轴外”位置是功能设计

arXiv 新研究探讨了 Transformer 内部概念计算的物理位置。模型答案通常位于一个轴上（unembedding 读取的方向），而中间状态大多不在此轴上，这通常被视为解释障碍。研究发现这种“轴外”位置是功能性的：12 层模型在特定位置计算概念，而非随机分布。这为理解 Transformer 内部机制提供了新视角，可能对模型可解释性和调试有帮助。

## English Version

**Off-Axis Computing: Where Transformers Process Concepts and Why**

A new arXiv study explores where a transformer computes concepts internally. The answer lies on one axis (the unembedding direction), but intermediate states are off-axis, usually treated as an obstacle. The research shows this off-axis position is functional: a 12-layer model computes concepts at specific sites, not randomly. This offers a new lens for understanding transformer internals, potentially aiding interpretability and debugging.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2608.10251)

**详情页**：https://ai.daily.yangsir.net/daily/20260813-T0-06

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*