机制化数据溯源：可解释性大模型单元的训源追踪

摘要

尽管机械可解释性研究已在大型语言模型中发现可解释电路，但其在训练数据中的因果起源仍不明确。我们提出机械数据归因（MDA）框架，该可扩展方法利用影响函数将可解释单元溯源至特定训练样本。通过对Pythia模型系列的广泛实验，我们因果验证了目标干预——移除或增强少量高影响力样本——能显著调控可解释注意力头的形成，而随机干预则无此效果。分析表明，重复性结构化数据（如LaTeX、XML）发挥着机械催化剂作用。此外，针对归纳头形成的干预会同步改变模型的上下文学习能力，这为归纳头与上下文学习功能关联的长期假说提供了直接因果证据。最后，我们提出一种机械数据增强流程，能持续加速不同规模模型的电路收敛，为引导大语言模型发展轨迹提供了原理性方法。

English

While Mechanistic Interpretability has identified interpretable circuits in LLMs, their causal origins in training data remain elusive. We introduce Mechanistic Data Attribution (MDA), a scalable framework that employs Influence Functions to trace interpretable units back to specific training samples. Through extensive experiments on the Pythia family, we causally validate that targeted intervention--removing or augmenting a small fraction of high-influence samples--significantly modulates the emergence of interpretable heads, whereas random interventions show no effect. Our analysis reveals that repetitive structural data (e.g., LaTeX, XML) acts as a mechanistic catalyst. Furthermore, we observe that interventions targeting induction head formation induce a concurrent change in the model's in-context learning (ICL) capability. This provides direct causal evidence for the long-standing hypothesis regarding the functional link between induction heads and ICL. Finally, we propose a mechanistic data augmentation pipeline that consistently accelerates circuit convergence across model scales, providing a principled methodology for steering the developmental trajectories of LLMs.

机制化数据溯源：可解释性大模型单元的训源追踪

Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units

摘要

Support