When Attention Sink Emerges in Language Models: An Empirical View

TL;DR

This paper investigates the universal emergence of attention sink in language models, revealing its causes, effects, and how alternative attention mechanisms can mitigate this phenomenon.

Contribution

It provides the first comprehensive empirical analysis of attention sink in LMs, exploring its origins, influencing factors, and potential solutions through alternative attention methods.

Findings

Attention sink exists universally across various LMs and inputs.

Attention sink emerges during pre-training after effective optimization.

Replacing softmax with sigmoid attention prevents sink emergence in models up to 1B parameters.

Abstract

Language Models (LMs) assign significant attention to the first token, even if it is not semantically important, which is known as attention sink. This phenomenon has been widely adopted in applications such as streaming/long context generation, KV cache optimization, inference acceleration, model quantization, and others. Despite its widespread use, a deep understanding of attention sink in LMs is still lacking. In this work, we first demonstrate that attention sinks exist universally in LMs with various inputs, even in small models. Furthermore, attention sink is observed to emerge during the LM pre-training, motivating us to investigate how optimization, data distribution, loss function, and model architecture in LM pre-training influence its emergence. We highlight that attention sink emerges after effective optimization on sufficient training data. The sink position is highly correlated with the loss function and data distribution. Most importantly, we find that attention sink acts more like key biases, storing extra attention scores, which could be non-informative and not contribute to the value computation. We also observe that this phenomenon (at least partially) stems from tokens' inner dependence on attention scores as a result of softmax normalization. After relaxing such dependence by replacing softmax attention with other attention operations, such as sigmoid attention without normalization, attention sinks do not emerge in LMs up to 1B parameters. The code is available at https://github.com/sail-sg/Attention-Sink.