While many advanced LLMs are designed to handle long sequence data, we can still observe notable quality degradation even within the sequence limit. In this work, we introduce a novel approach called Scaling to Emphasize Attention for Long-context retrieval (SEAL), which enhances the retrieval performance of large language models (LLMs) over long contexts. We observe that specific attention heads are closely tied to long-context retrieval, showing positive or negative correlation with retrieval scores, and adjusting the strength of these heads boosts the quality of LLMs in long context by a large margin. Built on this insight, we propose a learning-based mechanism that leverages generated data to emphasize these heads. By applying SEAL, we achieve significant improvements in long-context retrieval performance across various tasks and models. Additionally, when combined with existing training-free context extension techniques, SEAL extends the contextual limits of LLMs while maintaining highly reliable outputs.
@article{arxiv.2501.15225,
title = {SEAL: Scaling to Emphasize Attention for Long-Context Retrieval},
author = {Changhun Lee and Minsang Seok and Jun-gyu Jin and Younghyun Cho and Eunhyeok Park},
journal= {arXiv preprint arXiv:2501.15225},
year = {2025}
}