Saved in:
Bibliographic Details
Main Authors: Zeng, Xiangyu, Xu, Qi, Wang, Yunke, Xu, Chang
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.20843
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910114912403456
author Zeng, Xiangyu
Xu, Qi
Wang, Yunke
Xu, Chang
author_facet Zeng, Xiangyu
Xu, Qi
Wang, Yunke
Xu, Chang
contents Long-context language modeling is commonly framed as a scalability challenge of token-level attention, yet local-to-global information structuring remains largely implicit in existing approaches. Drawing on cognitive theories of discourse comprehension, we propose HiCI (Hierarchical Construction--Integration), a hierarchical attention module that constructs segment-level representations, integrates them into a shared global context, and broadcasts both to condition segment-level attention. We validate HiCI through parameter-efficient adaptation of LLaMA-2 with only <5.5% additional parameters, extending context from 4K to 100K tokens (7B) and 64K tokens (13B). Across language modeling, retrieval, and instruction-following benchmarks, HiCI yields consistent improvements over strong baselines, including matching proprietary models on topic retrieval and surpassing GPT-3.5-Turbo-16K on code comprehension. These results demonstrate the effectiveness of explicit hierarchical structuring as an inductive bias for long-context modeling.
format Preprint
id arxiv_https___arxiv_org_abs_2603_20843
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle HiCI: Hierarchical Construction-Integration for Long-Context Attention
Zeng, Xiangyu
Xu, Qi
Wang, Yunke
Xu, Chang
Computation and Language
Artificial Intelligence
Machine Learning
Long-context language modeling is commonly framed as a scalability challenge of token-level attention, yet local-to-global information structuring remains largely implicit in existing approaches. Drawing on cognitive theories of discourse comprehension, we propose HiCI (Hierarchical Construction--Integration), a hierarchical attention module that constructs segment-level representations, integrates them into a shared global context, and broadcasts both to condition segment-level attention. We validate HiCI through parameter-efficient adaptation of LLaMA-2 with only <5.5% additional parameters, extending context from 4K to 100K tokens (7B) and 64K tokens (13B). Across language modeling, retrieval, and instruction-following benchmarks, HiCI yields consistent improvements over strong baselines, including matching proprietary models on topic retrieval and surpassing GPT-3.5-Turbo-16K on code comprehension. These results demonstrate the effectiveness of explicit hierarchical structuring as an inductive bias for long-context modeling.
title HiCI: Hierarchical Construction-Integration for Long-Context Attention
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2603.20843