Sink vs. diagonal patterns as mechanisms for attention switch and oversmoothing prevention
For researchers studying attention mechanisms in transformers, this work clarifies the conditions and trade-offs between sinks and diagonal patterns for oversmoothing prevention and attention switching.
This paper analyzes the role of sinks and diagonal patterns in attention mechanisms, showing that sinks require alignment between embeddings and can cause hard attention switches. It proves an equivalence between sinks and hard attention switches, and compares sinks vs. diagonal patterns, explaining why sinks are favored in pretrained transformers.
This paper studies the role of sinks and diagonal patterns as attention switch and anti-oversmoothing mechanisms. We analyze geometric conditions under which sinks can be represented, showing a necessary alignment between the embedding of the sink and all other embeddings. Next, we refine the current understanding of the role of sinks in oversmoothing prevention: we specify the conditions under which dense attention provably smooths more than sparse attention, and empirically verify that such conditions are often satisfied in practice. We further prove an equivalence between sinks and hard attention switch, in which the output of the attention is identically 0. Finally, we relax the hard attention switch by allowing token self-communication: we provide a quantitative comparison of the costs of representing sinks vs.\ diagonal patterns, showing why sinks are favored in pretrained transformers. The introduction and analysis of diagonal patterns and the generalization of the attention switch close the gap between what oversmoothing prevention requires and what sinks provide, while also establishing when and why attention layers act like MLPs if token communication is not necessary.