Stacked from One: Multi-Scale Self-Injection for Context Window Extension

• URL: http://arxiv.org/abs/2603.04759v1
• Date: Thu, 05 Mar 2026 03:16:16 GMT
• Title: Stacked from One: Multi-Scale Self-Injection for Context Window Extension
• Authors: Wei Han, Pan Zhou, Shuicheng Yan,
• Abstract summary: modelname is a novel framework based on multi-grained context compression and query-aware information acquisition.<n>modelnameachieves performance superior or comparable to strong baselines.
• Score: 69.24689919827817
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: The limited context window of contemporary large language models (LLMs) remains a primary bottleneck for their broader application across diverse domains. Although continual pre-training on long-context data offers a straightforward solution, it incurs prohibitive data acquisition and computational costs. To address this challenge, we propose~\modelname, a novel framework based on multi-grained context compression and query-aware information acquisition. SharedLLM comprises two stacked short-context LLMs: a lower model serving as a compressor and an upper model acting as a decoder. The lower model compresses long inputs into compact, multi-grained representations, which are then forwarded to the upper model for context-aware processing. To maximize efficiency, this information transfer occurs exclusively at the lowest layers, bypassing lengthy forward passes and redundant cross-attention operations. This entire process, wherein the upper and lower models are derived from the same underlying LLM layers, is termed~\textit{self-injection}. To support this architecture, a specialized tree-based data structure enables the efficient encoding and query-aware retrieval of contextual information. Despite being trained on sequences of only 8K tokens, \modelname~effectively generalizes to inputs exceeding 128K tokens. Across a comprehensive suite of long-context modeling and understanding benchmarks, \modelname~achieves performance superior or comparable to strong baselines, striking an optimal balance between efficiency and accuracy. Furthermore, these design choices allow \modelname~to substantially reduce the memory footprint and yield notable inference speedups ($2\times$ over streaming and $3\times$ over encoder-decoder architectures).

Related papers

• Scaling LLM Speculative Decoding: Non-Autoregressive Forecasting in Large-Batch Scenarios [76.85739138203014]
  We present SpecFormer, a novel architecture that accelerates unidirectional and attention mechanisms.<n>We demonstrate that SpecFormer achieves lower training demands and reduced computational costs.
  arXiv  Detail & Related papers  (2025-11-25T14:20:08Z)
• Every Step Counts: Decoding Trajectories as Authorship Fingerprints of dLLMs [63.82840470917859]
  We show that the decoding mechanism of dLLMs can be used as a powerful tool for model attribution.<n>We propose a novel information extraction scheme called the Directed Decoding Map (DDM), which captures structural relationships between decoding steps and better reveals model-specific behaviors.
  arXiv  Detail & Related papers  (2025-10-02T06:25:10Z)
• SitEmb-v1.5: Improved Context-Aware Dense Retrieval for Semantic Association and Long Story Comprehension [77.93156509994994]
  We show how to represent short chunks in a way that is conditioned on a broader context window to enhance retrieval performance.<n>Existing embedding models are not well-equipped to encode such situated context effectively.<n>Our method substantially outperforms state-of-the-art embedding models.
  arXiv  Detail & Related papers  (2025-08-03T23:59:31Z)
• Two are better than one: Context window extension with multi-grained self-injection [111.1376461868317]
  SharedLLM is a novel approach grounded in the design philosophy of multi-grained context compression and query-aware information retrieval. We introduce a specialized tree-style data structure to efficiently encode, store and retrieve multi-grained contextual information for text chunks.
  arXiv  Detail & Related papers  (2024-10-25T06:08:59Z)
• LOCOST: State-Space Models for Long Document Abstractive Summarization [76.31514220737272]
  We propose LOCOST: an encoder-decoder architecture based on state-space models for conditional text generation with long context inputs. With a computational complexity of $O(L log L)$, this architecture can handle significantly longer sequences than state-of-the-art models that are based on sparse attention patterns.
  arXiv  Detail & Related papers  (2024-01-31T15:33:37Z)
• LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models [22.06402870816756]
  Large language models (LLMs) have been applied in various applications due to their astonishing capabilities. This paper presents LLMLingua, a coarse-to-fine prompt compression method that involves a budget controller to maintain semantic integrity. We show that the proposed approach yields state-of-the-art performance and allows for up to 20x compression with little performance loss.
  arXiv  Detail & Related papers  (2023-10-09T14:10:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.