InfoFlow KV: Information-Flow-Aware KV Recomputation for Long Context

• URL: http://arxiv.org/abs/2603.05353v1
• Date: Thu, 05 Mar 2026 16:33:20 GMT
• Title: InfoFlow KV: Information-Flow-Aware KV Recomputation for Long Context
• Abstract summary: We show that a simple attention-norm signal from the query reliably identifies tokens that are semantically relevant and structurally positioned to propagate information.<n> Experiments on LLM and VLM benchmarks demonstrate consistent gains over prior methods under comparable efficiency budgets.
• Score: 23.355776408834473
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Retrieval-augmented generation (RAG) for long-context question answering is bottlenecked by inference-time prefilling over large retrieved contexts. A common strategy is to precompute key-value (KV) caches for individual documents and selectively recompute a small subset of tokens to restore global causal dependencies, but existing methods rely on heuristics or representation discrepancies without modeling whether selected tokens can effectively influence generation. We cast selective KV recomputation as an information flow problem and show that a simple attention-norm signal from the query reliably identifies tokens that are both semantically relevant and structurally positioned to propagate information, when computed under an inference-consistent RoPE geometry. We therefore reconstruct global positional assignments for retrieved chunks and introduce an information-flow-guided chunk reordering strategy. Experiments on LLM and VLM benchmarks demonstrate consistent gains over prior methods under comparable efficiency budgets.

Related papers

• Where to Place the Query? Unveiling and Mitigating Positional Bias in In-Context Learning for Diffusion LLMs via Decoding Dynamics [5.345984783628041]
  This paper presents a comprehensive analysis unveiling that query position is actually a first-order variable in dLLMs.<n>We propose Average Confidence ($overlineC$), a novel metric tracking the iterative decoding process.<n>By establishing the foundational spatial ICL baselines, we introduce Auto-ICL, a training-free adaptive routing strategy.
  arXiv  Detail & Related papers  (2026-04-26T20:22:47Z)
• StructKV: Preserving the Structural Skeleton for Scalable Long-Context Inference [38.75965915663233]
  We propose StructKV, a structure-aware KV cache compression framework.<n>First, Global In-Degree Centrality aggregates attention patterns across the network depth to identify global information hubs.<n>Second, Dynamic Pivot Detection utilizes information-theoretic metrics to adaptively locate the optimal layer for compression.
  arXiv  Detail & Related papers  (2026-04-08T07:10:35Z)
• AnnoRetrieve: Efficient Structured Retrieval for Unstructured Document Analysis [11.689256498133446]
  We propose AnnoRetrieve, a novel retrieval paradigm that shifts from embeddings to structured annotations.<n>Our system replaces expensive vector comparisons with lightweight structured queries over automatically induced schemas.<n>AnoRetrieve establishes a new paradigm for cost-effective, precise, and scalable document analysis through intelligent structuring.
  arXiv  Detail & Related papers  (2026-04-03T03:34:19Z)
• Stem: Rethinking Causal Information Flow in Sparse Attention [28.46760179998361]
  We propose Stem, a novel, plug-and-play sparsity module aligned with information flow.<n>First, Stem employs the Token Position-Decay strategy, applying position-dependent top-k within each layer to retain initial tokens.<n>Second, to preserve information-rich tokens, Stem utilizes the Output-Aware Metric.
  arXiv  Detail & Related papers  (2026-03-06T13:33:29Z)
• SmartChunk Retrieval: Query-Aware Chunk Compression with Planning for Efficient Document RAG [41.16937860730275]
  We present SmartChunk, a query-adaptive framework for efficient and robust long-document question answering (QA)<n>SmartChunk uses a planner that predicts the optimal chunk abstraction level for each query, and a lightweight compression module that produces high-level chunk embeddings without repeated summarization.<n>To reflect real-world applications, where users face diverse document types and query styles, we evaluate SmartChunk on five QA benchmarks plus one out-of-domain dataset.
  arXiv  Detail & Related papers  (2025-12-17T01:21:44Z)
• CoT Referring: Improving Referring Expression Tasks with Grounded Reasoning [67.18702329644526]
  CoT Referring enhances model reasoning across modalities through a structured, chain-of-thought training data structure.<n>We restructure the training data to enforce a new output form, providing new annotations for existing datasets.<n>We also integrate detection and segmentation capabilities into a unified MLLM framework, training it with a novel adaptive weighted loss to optimize performance.
  arXiv  Detail & Related papers  (2025-10-03T08:50:21Z)
• Judge Q: Trainable Queries for Optimized Information Retention in KV Cache Eviction [53.83828564664595]
  Large language models (LLMs) utilize key-value ( KV) cache to store historical information during sequence processing.<n>Current methods for KV cache eviction typically utilize the last window from the pre-filling phase as queries to compute the KV importance scores for eviction.<n>We propose Judge Q, a novel training method which incorporates a soft token list.
  arXiv  Detail & Related papers  (2025-09-13T03:34:12Z)
• Topic Identification in LLM Input-Output Pairs through the Lens of Information Bottleneck [0.0]
  We develop a principled topic identification method grounded in the Deterministic Information Bottleneck (DIB) for geometric clustering.<n>Our key contribution is to transform the DIB method into a practical algorithm for high-dimensional data by substituting its intractable KL divergence term with a computationally efficient upper bound.
  arXiv  Detail & Related papers  (2025-08-26T20:00:51Z)
• PICASO: Permutation-Invariant Context Composition with State Space Models [98.91198288025117]
  State Space Models (SSMs) offer a promising solution by allowing a database of contexts to be mapped onto fixed-dimensional states.<n>We propose a simple mathematical relation derived from SSM dynamics to compose multiple states into one that efficiently approximates the effect of concatenating raw context tokens.<n>We evaluate our resulting method on WikiText and MSMARCO in both zero-shot and fine-tuned settings, and show that we can match the strongest performing baseline while enjoying on average 5.4x speedup.
  arXiv  Detail & Related papers  (2025-02-24T19:48:00Z)
• Activation-aware Probe-Query: Effective Key-Value Retrieval for Long-Context LLMs Inference [56.71209737306054]
  We propose textbfActQKV, a training-free, textbfActivation-aware approach that dynamically determines probe-textbfQuery and leverages it to retrieve the relevant textbfKV pairs for inference.<n>Experiments on the Long-Bench and $infty$ Benchmarks demonstrate its state-of-the-art performance with competitive inference quality and resource efficiency.
  arXiv  Detail & Related papers  (2025-02-19T08:50:44Z)
• RAGEval: Scenario Specific RAG Evaluation Dataset Generation Framework [66.93260816493553]
  This paper introduces RAGEval, a framework designed to assess RAG systems across diverse scenarios.<n>With a focus on factual accuracy, we propose three novel metrics: Completeness, Hallucination, and Irrelevance.<n> Experimental results show that RAGEval outperforms zero-shot and one-shot methods in terms of clarity, safety, conformity, and richness of generated samples.
  arXiv  Detail & Related papers  (2024-08-02T13:35:11Z)
• Optimization of Retrieval-Augmented Generation Context with Outlier Detection [0.0]
  We focus on methods to reduce the size and improve the quality of the prompt context required for question-answering systems. Our goal is to select the most semantically relevant documents, treating the discarded ones as outliers. It was found that the greatest improvements were achieved with increasing complexity of the questions and answers.
  arXiv  Detail & Related papers  (2024-07-01T15:53:29Z)
• Large-scale Fully-Unsupervised Re-Identification [78.47108158030213]
  We propose two strategies to learn from large-scale unlabeled data. The first strategy performs a local neighborhood sampling to reduce the dataset size in each without violating neighborhood relationships. A second strategy leverages a novel Re-Ranking technique, which has a lower time upper bound complexity and reduces the memory complexity from O(n2) to O(kn) with k n.
  arXiv  Detail & Related papers  (2023-07-26T16:19:19Z)
• New advances in enumerative biclustering algorithms with online partitioning [80.22629846165306]
  This paper further extends RIn-Close_CVC, a biclustering algorithm capable of performing an efficient, complete, correct and non-redundant enumeration of maximal biclusters with constant values on columns in numerical datasets. The improved algorithm is called RIn-Close_CVC3, keeps those attractive properties of RIn-Close_CVC, and is characterized by: a drastic reduction in memory usage; a consistent gain in runtime.
  arXiv  Detail & Related papers  (2020-03-07T14:54:26Z)

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.