Search papers, labs, and topics across Lattice.
This paper investigates the layer-wise contributions of internal attention signals in Large Language Models (LLMs) for zero-shot document re-ranking, comparing them to traditional generative and likelihood-based mechanisms. They discover a "bell-curve" distribution of relevance signals across transformer layers, with mid-layers exhibiting the strongest signal. Based on this finding, they propose Selective-ICR, a method that selectively aggregates attention signals from the most relevant layers, achieving significant latency reductions (30%-50%) without sacrificing effectiveness.
Forget scaling and RLHF: carefully selecting internal attention signals from the right layers lets a zero-shot 8B model match a 14B reinforcement-learned re-ranker in complex reasoning tasks.
Zero-shot document re-ranking with Large Language Models (LLMs) has evolved from Pointwise methods to Listwise and Setwise approaches that optimize computational efficiency. Despite their success, these methods predominantly rely on generative scoring or output logits, which face bottlenecks in inference latency and result consistency. In-Context Re-ranking (ICR) has recently been proposed as an $O(1)$ alternative method. ICR extracts internal attention signals directly, avoiding the overhead of text generation. However, existing ICR methods simply aggregate signals across all layers; layer-wise contributions and their consistency across architectures have been left unexplored. Furthermore, no unified study has compared internal attention with traditional generative and likelihood-based mechanisms across diverse ranking frameworks under consistent conditions. In this paper, we conduct an orthogonal evaluation of generation, likelihood, and internal attention mechanisms across multiple ranking frameworks. We further identify a universal"bell-curve"distribution of relevance signals across transformer layers, which motivates the proposed Selective-ICR strategy that reduces inference latency by 30%-50% without compromising effectiveness. Finally, evaluation on the reasoning-intensive BRIGHT benchmark shows that precisely capturing high-quality in-context attention signals fundamentally reduces the need for model scaling and reinforcement learning: a zero-shot 8B model matches the performance of 14B reinforcement-learned re-rankers, while even a 0.6B model outperforms state-of-the-art generation-based approaches. These findings redefine the efficiency-effectiveness frontier for LLM-based re-ranking and highlight the latent potential of internal signals for complex reasoning ranking tasks. Our code and results are publicly available at https://github.com/ielab/Selective-ICR.
