Outlier Dimensions Encode Task Specific Knowledge

William Rudman, Catherine Chen, Carsten Eickhoff

Submission Type: Regular Short Paper

Submission Track: Interpretability, Interactivity, and Analysis of Models for NLP

Keywords: outlier dimensions, LLMs, fine-tuning, interpretability

TL;DR: Our results suggest that outlier dimensions can encode crucial task-specific knowledge and that the value of a representation in a single outlier dimension drives downstream model decisions.

Abstract: Representations from large language models (LLMs) are known to be dominated by a small subset of dimensions with exceedingly high variance. Previous works have argued that although ablating these outlier dimensions in LLM representations hurts downstream performance, outlier dimensions are detrimental to the representational quality of embeddings. In this study, we investigate how fine-tuning impacts outlier dimensions and show that 1) outlier dimensions that occur in pre-training persist in fine-tuned models and 2) a single outlier dimension can complete downstream tasks with a minimal error rate. Our results suggest that outlier dimensions can encode crucial task-specific knowledge and that the value of a representation in a single outlier dimension drives downstream model decisions.

Submission Number: 3242