SLASH the Sink: Sharpening Structural Attention Inside LLMs

In three linesLLMs spontaneously reconstruct graph topology via sawtooth attention patterns, but this structural understanding is diluted by attention sink. SLASH, a training-free solution, re-amplifies this understanding through plug-and-play attention redistribution, showing significant gains on graph tasks and molecular prediction across diverse LLMs.

Summary generated by Claude — human-verified