Garbage Attention in Large Language Models: BOS Sink Heads and Sink-aware Pruning

TL;DR

This paper identifies the BOS sink phenomenon as a key reason for redundancy in large language models, and proposes a pruning method targeting high-BOS sink heads to improve model compression without performance loss.

Contribution

It introduces the BOS sink concept to explain layer-wise redundancy and develops a pruning strategy based on this insight, outperforming traditional weight- or activation-based methods.

Findings

High-BOS sink heads are highly redundant and contribute little to performance.

Pruning high-BOS sink heads preserves accuracy even with aggressive compression.

Sink head behavior remains stable across different sequence lengths.

Abstract

Large Language Models (LLMs) are known to contain significant redundancy, yet a systematic explanation for why certain components, particularly in higher layers, are more redundant has remained elusive. In this work, we identify the BOS sink phenomenon as a key mechanism driving this layer-wise sensitivity. We show that attention heads with high BOS sink scores are strongly associated with functional redundancy: such heads, especially in deeper layers, contribute little to predictive performance and effectively serve as \emph{dumping grounds} for superfluous attention weights. This provides a concrete functional explanation for the structural redundancy reported in prior studies. Leveraging this insight, we introduce a simple pruning strategy that removes high-BOS sink heads. Experiments on Gemma-3, Llama-3.1, and Qwen3 demonstrate that this approach identifies redundant transformer components more reliably than weight- or activation-based criteria, while preserving performance close to dense baselines even under aggressive pruning. Moreover, we find that the behavior of sink heads remains stable across different sequence lengths. Overall, our results suggest that structural properties of attention offer a more intuitive and robust basis for model compression than magnitude-based methods.