Why don't Graph Transformers attend to the same token?

arXiv:2610.19324v1 Announce Type: new Abstract: In large language models (LLMs), attention concentrates heavily on the first token. This 'attention sink' is believed to protect deep models against over-mixing. Even though graph transformers face a related pressure, we find that a SoTA graph transformer, GRIT, rarely forms sinks on molecules, and that it sinks less when trained on larger graphs, the opposite of the trend in LLMs. In principle, a graph can prevent sinks: we prove and show empirically that its symmetry and locality limit the attention any node can receive. On molecules, however, with distinguishable atoms and global attention, neither limit prevents sinks; what stops them lies in the model. We find that the attention scoring function is the main cause. Replacing GRIT's gated logit with a dot product raises its sink rate from 0.09 to 0.68 on identical graphs. In Graph Transformers using dot-product attention, removing the sink gives about the same error as removing a random node, while GRIT's most-attended nodes matter to its predictions. Attention sinks are therefore a property of dot-product attention rather than of Transformers. In our experiments, they form when dot-product scoring has a distinguishable node to concentrate on.

Title:Why don't Graph Transformers attend to the same token?

Abstract:In large language models (LLMs), attention concentrates heavily on the first token. This 'attention sink' is believed to protect deep models against over-mixing. Even though graph transformers face a related pressure, we find that a SoTA graph transformer, GRIT, rarely forms sinks on molecules, and that it sinks less when trained on larger graphs, the opposite of the trend in LLMs. In principle, a graph can prevent sinks: we prove and show empirically that its symmetry and locality limit the attention any node can receive. On molecules, however, with distinguishable atoms and global attention, neither limit prevents sinks; what stops them lies in the model. We find that the attention scoring function is the main cause. Replacing GRIT's gated logit with a dot product raises its sink rate from 0.09 to 0.68 on identical graphs. In Graph Transformers using dot-product attention, removing the sink gives about the same error as removing a random node, while GRIT's most-attended nodes matter to its predictions. Attention sinks are therefore a property of dot-product attention rather than of Transformers. In our experiments, they form when dot-product scoring has a distinguishable node to concentrate on.

Accepted at the Geometric Distributional Deep Learning (GDDL) workshop at NeurIPS 2026. 8 pages, 1 figure. Code: this https URL

Machine Learning (cs.LG)

arXiv:2610.19324 [cs.LG]

(or arXiv:2610.19324v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2610.19324 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

analiza aiqbench

Why don't Graph Transformers attend to the same token?

Subiecte conexe