Why do LLMs attend to the first token?

Federico Barbero, Alvaro Arroyo, Xiangming Gu, Christos Perivolaropoulos, Petar Veličković, Razvan Pascanu, Michael M. Bronstein

Keywords: Large Language Models, Attention Sinks, Information Propagation, Pre-training

TL;DR: We study why it is useful for attention heads in LLMs to learn to "dump" most attention on the first token from an "over-mixing" perspective.

Abstract: Large Language Models (LLMs) tend to attend heavily to the first token in the sequence -- creating a so-called attention sink. Many works have studied this phenomenon in detail, proposing various ways to either leverage or alleviate it. Attention sinks have been connected to quantisation difficulties, security issues, and streaming attention. Yet, while many works have provided conditions in which they occur or not, a critical question remains shallowly answered: Why do LLMs learn such patterns and how are they being used? In this work, we argue theoretically and empirically that this mechanism provides a method for LLMs to avoid over-mixing, connecting this to existing lines of work that study mathematically how information propagates in Transformers. We run experiments to validate our theoretical intuitions and show how choices such as context length, depth, and data packing influence the sink behaviour. We hope that this study provides a new practical perspective on why attention sinks are useful in LLMs, leading to a better understanding of the attention patterns that form during training.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the COLM Code of Ethics on https://colmweb.org/CoE.html

Author Guide: I certify that this submission complies with the submission instructions as described on https://colmweb.org/AuthorGuide.html

Submission Number: 866