Why do LLMs attend to the first token?
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915427745005568 |
|---|---|
| author | Barbero, Federico Arroyo, Álvaro Gu, Xiangming Perivolaropoulos, Christos Bronstein, Michael Veličković, Petar Pascanu, Razvan |
| author_facet | Barbero, Federico Arroyo, Álvaro Gu, Xiangming Perivolaropoulos, Christos Bronstein, Michael Veličković, Petar Pascanu, Razvan |
| contents | 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 conduct 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_02732 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Why do LLMs attend to the first token? Barbero, Federico Arroyo, Álvaro Gu, Xiangming Perivolaropoulos, Christos Bronstein, Michael Veličković, Petar Pascanu, Razvan Computation and Language 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 conduct 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. |
| title | Why do LLMs attend to the first token? |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2504.02732 |