Patches of Nonlinearity: Instruction Vectors in Large Language Models
For mechanistic interpretability researchers, this work challenges the linear representation hypothesis and provides a new method to study non-linear causal interactions in LLMs.
The paper investigates how instruction-tuned LLMs process instructions internally, identifying localized 'Instruction Vectors' (IVs) that exhibit both linear separability and non-linear causal interaction. A novel method is proposed to localize information processing without linear assumptions, revealing that IVs act as circuit selectors for task-specific pathways.
Despite the recent success of instruction-tuned language models and their ubiquitous usage, very little is known of how models process instructions internally. In this work, we address this gap from a mechanistic point of view by investigating how instruction-specific representations are constructed and utilized in different stages of post-training: Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO). Via causal mediation, we identify that instruction representation is fairly localized in models. These representations, which we call Instruction Vectors (IVs), demonstrate a curious juxtaposition of linear separability along with non-linear causal interaction, broadly questioning the scope of the linear representation hypothesis commonplace in mechanistic interpretability. To disentangle the non-linear causal interaction, we propose a novel method to localize information processing in language models that is free from the implicit linear assumptions of patching-based techniques. We find that, conditioned on the task representations formed in the early layers, different information pathways are selected in the later layers to solve that task, i.e., IVs act as circuit selectors.