English

Transformers need glasses! Information over-squashing in language tasks

Computation and Language 2024-10-28 v2 Machine Learning

Abstract

We study how information propagates in decoder-only Transformers, which are the architectural backbone of most existing frontier large language models (LLMs). We rely on a theoretical signal propagation analysis -- specifically, we analyse the representations of the last token in the final layer of the Transformer, as this is the representation used for next-token prediction. Our analysis reveals a representational collapse phenomenon: we prove that certain distinct sequences of inputs to the Transformer can yield arbitrarily close representations in the final token. This effect is exacerbated by the low-precision floating-point formats frequently used in modern LLMs. As a result, the model is provably unable to respond to these sequences in different ways -- leading to errors in, e.g., tasks involving counting or copying. Further, we show that decoder-only Transformer language models can lose sensitivity to specific tokens in the input, which relates to the well-known phenomenon of over-squashing in graph neural networks. We provide empirical evidence supporting our claims on contemporary LLMs. Our theory also points to simple solutions towards ameliorating these issues.

Keywords

Cite

@article{arxiv.2406.04267,
  title  = {Transformers need glasses! Information over-squashing in language tasks},
  author = {Federico Barbero and Andrea Banino and Steven Kapturowski and Dharshan Kumaran and João G. M. Araújo and Alex Vitvitskyi and Razvan Pascanu and Petar Veličković},
  journal= {arXiv preprint arXiv:2406.04267},
  year   = {2024}
}
R2 v1 2026-06-28T16:56:12.364Z