English

Outlier Dimensions Encode Task-Specific Knowledge

Computation and Language 2024-01-24 v2 Artificial Intelligence

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.

Keywords

Cite

@article{arxiv.2310.17715,
  title  = {Outlier Dimensions Encode Task-Specific Knowledge},
  author = {William Rudman and Catherine Chen and Carsten Eickhoff},
  journal= {arXiv preprint arXiv:2310.17715},
  year   = {2024}
}

Comments

Camera-ready version for EMNLP 2023

R2 v1 2026-06-28T13:03:12.622Z