Long-Context Language Modeling

Ask about this topic

Overview

The field asks how models can retrieve relevant evidence and maintain useful context as sequences grow, for tasks such as document understanding, code analysis, and long-range recall. Approaches include sparse and hierarchical attention, positional-encoding extensions, recurrent and state-space memory, hybrid architectures, and compression of context or key-value caches. Training methods range from long-sequence continued pretraining and curricula to distillation and test-time adaptation, with evaluation often using retrieval and reasoning benchmarks such as RULER, LongBench, and Needle-in-a-Haystack. The central tension is reducing compute and memory without losing reliable access to distant information; performance also depends on information density and may not extrapolate from short training contexts.

Latest papers

CardsList

189 papers38 papers in the last four weeks, up 90% on the four weeks before. 0.2% of all new papers.

Apr 11, 2026

SinkTrack: Attention Sink based Context Anchoring for Large Language Models

Large language models (LLMs) suffer from hallucination and context forgetting. Prior studies suggest that attention drift is a primary cause of these problems, where LLMs' focus shifts towards newly generated tokens and away from the initial input context. To counteract this, we make use of a related, intrinsic characteristic of LLMs: attention sink -- the tendency to consistently allocate high attention to the very first token (i.e., <BOS>) of a sequence. Concretely, we propose an advanced context anchoring method, SinkTrack, which treats <BOS> as an information anchor and injects key contextual features (such as those derived from the input image or instruction) into its representation. As such, LLM remains anchored to the initial input context throughout the entire generation process. SinkTrack is training-free, plug-and-play, and introduces negligible inference overhead. Experiments demonstrate that SinkTrack mitigates hallucination and context forgetting across both textual (e.g., +21.6% on SQuAD2.0 with Llama3.1-8B-Instruct) and multi-modal (e.g., +22.8% on M3CoT with Qwen2.5-VL-7B-Instruct) tasks. Its consistent gains across different architectures and scales underscore the robustness and generalizability. We also analyze its underlying working mechanism from the perspective of information delivery. Our source code is available at https://github.com/67L1/SinkTrack.

Xu Liu, Guikun Chen, Wenguan Wang

The State Key Lab of Brain-Machine Intelligence, Zhejiang University

LLM Context ManagementLLM Hallucination MitigationLong-Context Language Modeling
AskPDF
Apr 9, 2026

PolicyLong: Towards On-Policy Context Extension

Extending LLM context windows is hindered by scarce high-quality long-context data. Recent methods synthesize data with genuine long-range dependencies via information-theoretic verification, selecting contexts that reduce a base model's predictive entropy. However, their single-pass offline construction with a fixed model creates a fundamental off-policy gap: the static screening landscape misaligns with the model's evolving capabilities, causing the training distribution to drift. We propose PolicyLong, shifting data construction towards a dynamic on-policy paradigm. By iteratively re-executing data screening (entropy computation, retrieval, and verification) using the current model, PolicyLong ensures the training distribution tracks evolving capabilities, yielding an emergent self-curriculum. Crucially, both positive and hard negative contexts derive from the current model's entropy landscape, co-evolving what the model learns to exploit and resist. Experiments on RULER, HELMET, and LongBench-v2 (Qwen2.5-3B) show PolicyLong consistently outperforms EntropyLong and NExtLong, with gains growing at longer contexts (e.g., +2.54 at 128K on RULER), confirming the value of on-policy data evolution.

Junlong Jia, Jiang Zhou, Ziyang Chen +5

Hunyuan Team, Tencent · Chinese Academy of Sciences

Long-Context Language ModelingSelf-Training for Language Models
AskPDF
Apr 8, 2026

Optimal Decay Spectra for Linear Recurrences

Linear recurrent models offer linear-time sequence processing but often suffer from suboptimal long-range memory. We trace this to the decay spectrum: for NN channels, random initialization collapses the minimum spectral gap to O(N−2)O(N^{-2}), yielding sub-exponential error exp⁡(−Ω(N/log⁡N))\exp(-Ω(N/\log N)); linear spacing avoids collapse but degrades to exp⁡(−O(N/T))\exp(-O(N/\sqrt{T})), practically algebraic over long contexts. We introduce Position-Adaptive Spectral Tapering (PoST), an architecture-agnostic framework combining two mechanisms: (1) Spectral Reparameterization, which structurally enforces geometrically spaced log-decay rates, proven minimax optimal at rate O(exp⁡(−cN/log⁡T))O(\exp(-cN/\log T)); and (2) Position-Adaptive Scaling, the provably unique mechanism that eliminates the scale mismatch of static spectra (where only Nlog⁡t/log⁡TN\log t/\log T of NN channels are effective at position tt) by stretching the spectrum to the actual dependency range, sharpening the rate to O(exp⁡(−cN/log⁡t))O(\exp(-cN/\log t)). This scaling natively induces fractional invariance: the impulse response becomes scale-free, with channels interpolating between relative and absolute temporal coordinates. PoST integrates into any diagonal linear recurrence without overhead. We instantiate it across Mamba-2, RWKV-7, Gated DeltaNet, Gated Linear Attention, and RetNet. Pre-training at 180M-440M scales shows consistent zero-shot language modeling improvements, significant long-context retrieval gains for Mamba-2 (MQAR and NIAH), and competitive or improved performance across other architectures. Code: https://github.com/SiLifen/PoST.

Yang Cao

Linear RNNsLong-Range Dependency ModelingLong-Context Language ModelingAutoregressive Language Modeling
AskPDF
Apr 7, 2026

Short Data, Long Context: Distilling Positional Knowledge in Transformers

Extending the context window of language models typically requires expensive long-context pre-training, posing significant challenges for both training efficiency and data collection. In this paper, we present evidence that long-context retrieval capabilities can be transferred to student models through logit-based knowledge distillation, even when training exclusively on packed short-context samples within a long-context window. We provide comprehensive insights through the lens of Rotary Position Embedding (RoPE) and establish three key findings. First, consistent with prior work, we show that phase-wise RoPE scaling, which maximizes rotational spectrum utilization at each training stage, also achieves the best long-context performance in knowledge distillation setups. Second, we demonstrate that logit-based knowledge distillation can directly enable positional information transfer. Using an experimental setup with packed repeated token sequences, we trace the propagation of positional perturbations from query and key vectors through successive transformer layers to output logits, revealing that positional information systematically influences the teacher's output distribution and, in turn, the distillation signal received by the student model. Third, our analysis uncovers structured update patterns in the query state during long-context extension, with distinct parameter spans exhibiting strong sensitivity to long-context training.

Patrick Huber, Ernie Chang, Chinnadhurai Sankar +4

Meta Reality Labs

Long-Context Language ModelingLLM Context ManagementLanguage Model DistillationRotary Positional Embeddings
AskPDF
Apr 7, 2026

Residual-Mass Accounting for Partial-KV Decoding

We study a controlled partial-KV decoding setting in which exact unnormalized softmax contributions are computed for sink/tail anchors and a retrieved token set, while the remaining prefill tokens are represented by a residual estimate. We focus on the accounting rule after the query-dependent exact support has been selected, and use exhaustive Top-K only as an oracle selector, not as a deployable retrieval system. The proposed rule leaves the backbone language model and the exact-branch KV tensors unchanged. It builds fixed-size summary states (S,u)(S,u) from learned positive feature maps φφ, subtracts retrieved-token feature contributions to keep the exact and residual sets non-overlapping, and merges the estimated residual numerator and denominator with the exact branch under one normalization. At a 1% exact-support budget, our residual-completion method improves over the selection-only Top-K baseline on RULER and BABILong across frozen 1B and 3B Llama-3.2-Instruct backbones at all reported context lengths. In the 0.5-4% exact-support budget sweeps, this trend largely persists. On LongBench, summarization results are mostly favorable, while multi-document QA is mixed. Attention-output diagnostics support retrieved-token subtraction as the partition-consistent accounting rule, while indicating that the main remaining error is imperfect learned-φφ approximation of the unretrieved residual mass.

Yasuto Hoshi, Daisuke Miyashita, Jun Deguchi

Kioxia Corporation

Softmax AttentionLong-Context Language ModelingLLM Inference Acceleration
AskPDF
Apr 5, 2026

CAWN: Continuous Acoustic Wave Networks for Autoregressive Language Modeling

Modern Large Language Models (LLMs) rely on Transformer self-attention, which scales quadratically with sequence length. Recent linear-time alternatives, like State Space Models (SSMs), often suffer from signal degradation over extended contexts. We introduce the Continuous Acoustic Wave Network (CAWN), a fully continuous sequence-mixing architecture. Instead of discrete matrix-based attention, CAWN projects hidden states into multi-headed complex-domain phasors, achieving sequence mixing through a causal, O(L)O(L) Phase Accumulation mechanism. To prevent signal degradation over ultra-long contexts, we introduce a dual-gated Selective Phase Resonance mechanism incorporating Frequency-Dependent Retention, Hard-Threshold Gating via Straight-Through Estimation, and a Temporal Syntax Cache to capture short-term local dependencies. We also replace standard dense linear projections with Depth-wise Harmonic Convolutions for optimal spatial frequency mixing, augmented by Block Attention Residuals for depth-wise state routing. Scaled to a 150M-parameter model, CAWN utilizes custom Triton kernels for hardware-efficient, true-complex phase accumulation in float32. Trained via a continuous streaming loop on a 100-Billion-token corpus, the prototype is evaluated at a 5-Billion-token milestone. Empirical evaluations via a Targeted Semantic Retrieval protocol demonstrate robust vocabulary acquisition and extended explicitly learned contextual denoising. By leveraging O(1)O(1) state-passing via chunked prefill, the model retrieves targeted information across 2,000,000 tokens while strictly plateauing at 8.72 GB of Peak VRAM, empirically overcoming the O(L2)O(L^2) context memory wall.

Dejan Čugalj, Aleksandar Jevremovic

Singidunum University

Long-Context Language ModelingSequence ModelingAutoregressive Language ModelingComplex-Valued Neural Networks
AskPDF
Apr 1, 2026

Screening Is Enough

We call query--key relevance absolute when its values lie on a fixed bounded scale, depend on neither competing keys nor sequence length, require no sequence-length-dependent calibration, and can all be zero. To realize this notion, we introduce screening, whose explicit threshold transforms bounded query--key similarities into relevance values, enabling exact rejection, empty selection, and direct inspection on a common scale. In a controlled comparison of 12 attention mechanisms on a matched Transformer backbone, only screening maintains both low long-context perplexity and robust retrieval beyond the training context; notably, it does so without inference-time scaling. Building on screening, we introduce Multiscreen, a language-model architecture composed of parallel gated screening tiles. Multiscreen retains these long-context gains while achieving greater parameter efficiency, stronger general zero-shot downstream performance, lower training cost at larger scales, and lower model-side time to first token than Transformer baselines. We further develop a normalization design that keeps Multiscreen training stable even at a learning rate of 11 and show that an adapted version likewise stabilizes Transformer at the same learning rate.

Ken M. Nakanishi

Center for Emergent Matter Science (CEMS), RIKEN · Graduate School of Science, The University of Tokyo

Self-AttentionLong-Context Language ModelingTransformer AttentionLLM Inference AccelerationAttention Mechanisms
AskPDF
Mar 27, 2026

Switch Attention: Towards Dynamic and Fine-grained Hybrid Transformers

The attention mechanism has been the core component in modern transformer architectures. However, the computation of standard full attention scales quadratically with the sequence length, serving as a major bottleneck in long-context language modeling. Sliding window attention restricts the context length for better efficiency at the cost of narrower receptive fields. While existing efforts attempt to take the benefits from both sides by building hybrid models, they often resort to static, heuristically designed alternating patterns that limit efficient allocation of computation in various scenarios. In this paper, we propose Switch Attention (SwiAttn), a novel hybrid transformer that enables dynamic and fine-grained routing between full attention and sliding window attention. For each token at each transformer layer, SwiAttn dynamically routes the computation to either a full-attention branch for global information aggregation or a sliding-window branch for efficient local pattern matching. An adaptive regularization objective is designed to encourage the model towards efficiency. Moreover, we adopt continual pretraining to optimize the model, transferring the full attention architecture to the hybrid one. Extensive experiments are conducted on twenty-three benchmark datasets across both regular (4K) and long (32K) context lengths, demonstrating the effectiveness of the proposed method.

Yusheng Zhao, Hourun Li, Bohan Wu +5

State Key Laboratory for Multimedia Information Processing, School of Computer Science, PKU-Anker LLM Lab, Peking University, Beijing, China · Contribution during internship at Huawei Technologies Co., Ltd. · Huawei Technologies Co., Ltd.

Hybrid Transformer ArchitecturesHybrid AttentionLong-Context Language ModelingConditional ComputationSliding-Window Attention
AskPDF
Mar 26, 2026

Density-aware Soft Context Compression with Semi-Dynamic Compression Ratio

Soft context compression reduces the computational workload of processing long contexts in LLMs by encoding long context into a smaller number of latent tokens. However, existing frameworks apply uniform compression ratios, failing to account for the extreme variance in natural language information density. While adopting a density-aware dynamic compression ratio seems intuitive, empirical investigations reveal that models struggle intrinsically with operations parameterized by input dependent, continuous structural hyperparameters. To resolve this pitfall, we introduce Semi-Dynamic Context Compression framework. Our approach features a Discrete Ratio Selector, which predicts a compression target based on intrinsic information density and quantizes it to a predefined set of discrete compression ratios. It is efficiently jointly trained with the compressor on synthetic data, with the summary lengths as a proxy to create labels for compression ratio prediction. Extensive evaluations confirm that our density-aware framework, utilizing mean pooling as the backbone, consistently outperforms static baselines, establishing a robust Pareto frontier for context compression techniques. Our code, data and model weights are available at https://github.com/yuyijiong/semi-dynamic-context-compress

Yijiong Yu, Shuai Yuan, Jie Zheng +2

Oregon State University · DeepSolution

LLM Context CompressionContext CompressionLLM CompressionLong-Context Language Modeling
AskPDF
Mar 26, 2026

Large Language Model as Token Compressor and Decompressor

In this paper, we study whether an off-the-shelf LLM can be adapted into a discrete, variable-length token compressor and decompressor for long-context processing. To this end, we design a self-expressive autoencoding framework that fine-tunes a pretrained LLM with lightweight LoRA adapters to map long texts into compact sequences of learned latent codes, termed Z-tokens, and to decode them back into natural language or task outputs. The resulting representation is content-adaptive: less predictable or information-dense segments can receive more Z-tokens, while redundant regions can be represented more compactly through a budget-aware length regularizer. Our method is evaluated on long-context datasets such as Wikipedia, CNN/DailyMail, HotpotQA, and QuALITY, showing that it preserves reconstruction quality and downstream performance while reducing effective context length, generation-stage memory usage, and end-to-end latency. This simple design supports both direct decoding from compressed contexts and autoregressive generation in the Z-token space, providing a practical interface for efficient long-context inference.

Wenbing Li, Yiran Wang, Zikai Song +4

LLM Context CompressionLLM CompressionLLM Inference AccelerationLong-Context Language ModelingLLM Fine-Tuning
AskPDF
Mar 23, 2026

MemDLM: Memory-Enhanced DLM Training

Diffusion Language Models (DLMs) offer attractive advantages over Auto-Regressive (AR) models, such as full-attention parallel decoding and flexible generation. However, standard DLM training uses a static, single-step masked prediction objective that never exposes the model to the progressive denoising dynamics of inference, and forces all contextual information to be maintained purely through token-space attention, which becomes increasingly diluted as context length grows. We propose MemDLM (Memory-Enhanced DLM), which introduces a second memory channel by embedding a simulated denoising trajectory into training via Bi-level Optimization. An inner loop updates a set of fast weights, forming a Parametric Memory that captures the local trajectory experience, while an outer loop updates the base model conditioned on this memory. By offloading part of the memorization burden from token-space attention to parameter space, MemDLM yields faster convergence, stronger long-context representations, and lower training loss, even when the fast weights are discarded at inference time. Re-enabling the inner loop at inference provides an additional prompt-specific adaptation effect, where the Parametric Memory acts as an emergent in-weight retrieval mechanism on challenging Needle-in-a-Haystack tasks. Code: https://github.com/JarvisPei/MemDLM.

Zehua Pei, Hui-Ling Zhen, Weizhe Lin +4

The Chinese University of Hong Kong · Huawei Technologies Co., Ltd

Diffusion Language Model TrainingDiffusion Language ModelsLong-Context Language Modeling
AskPDF
Mar 23, 2026

TAMTRL: Teacher-Aligned Reward Reshaping for Multi-Turn Reinforcement Learning in Long-Context Compression

The rapid progress of large language models (LLMs) has led to remarkable performance gains across a wide range of tasks. However, when handling long documents that exceed the model's context window limit, the entire context cannot be processed in a single pass, making chunk-wise processing necessary. This requires multiple turns to read different chunks and update memory. However, supervision is typically provided only by the final outcome, which makes it difficult to evaluate the quality of memory updates at each turn in the multi-turn training setting. This introduces a temporal credit assignment challenge. Existing approaches, such as LLM-as-a-judge or process reward models, incur substantial computational overhead and suffer from estimation noise. To better address the credit assignment problem in multi-turn memory training, we propose Teacher-Aligned Reward Reshaping for Multi-Turn Reinforcement Learning (TAMTRL). TAMTRL leverages relevant documents as teacher signals by aligning them with each turn of model input and assigns rewards through normalized probabilities in a self-supervised manner. This provides fine-grained learning signals for each memory update and improves long-context processing. Experiments with multiple models of varying scales across seven long-context benchmarks show that TAMTRL consistently outperforms strong baselines, demonstrating its effectiveness. Our code is available at https://anonymous.4open.science/r/TAMTRL-F1F8.

Li Wang, Yandong Wang, Xin Yu +3

School of Artificial Intelligence, Beihang University, Beijing, 100191, China · Hangzhou International Innovation Institute, Beihang University, Hangzhou, China · Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, Beihang University, Beijing, China

RL for Language ModelsLong-Context Language ModelingReward ShapingMemory-Augmented Language Models
AskPDF
Mar 21, 2026

HiCI: Hierarchical Construction-Integration for Long-Context Attention

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.

Xiangyu Zeng, Qi Xu, Yunke Wang +1

School of Computer Science, University of Sydney, Sydney, Australia · 2Westlake University, Hangzhou, China.

Long-Context Language ModelingLong-Context ModelingLong-Context Reasoning in Language ModelsAttention Mechanisms
AskPDF
Mar 21, 2026

Beyond Token Eviction: Mixed-Dimension Budget Allocation for Efficient KV Cache Compression

Key-value (KV) caching is widely used to accelerate transformer inference, but its memory cost grows linearly with input length, limiting long-context deployment. Existing token eviction methods reduce memory by discarding less important tokens, which can be viewed as a coarse form of dimensionality reduction that assigns each token either zero or full dimension. We propose MixedDimKV, a mixed-dimension KV cache compression method that allocates dimensions to tokens at a more granular level, and MixedDimKV-H, which further integrates head-level importance information. Experiments on long-context benchmarks show that MixedDimKV outperforms prior KV cache compression methods that do not rely on head-level importance profiling. When equipped with the same head-level importance information, MixedDimKV-H consistently outperforms HeadKV. Notably, our approach achieves comparable performance to full attention on LongBench with only 6.25% of the KV cache. Furthermore, in the Needle-in-a-Haystack test, our solution maintains 100% accuracy at a 50K context length while using as little as 0.26% of the cache.

Ruijie Miao, Zhiming Wang, Wang Li +4

Peking University · ByteDance Seed

KV-Cache ManagementKV-Cache CompressionLLM Inference AccelerationLong-Context Language Modeling
AskPDF
Mar 21, 2026

MKA: Memory-Keyed Attention for Efficient Long-Context Reasoning

As long-context language modeling becomes increasingly important, the cost of maintaining and attending to large Key/Value (KV) caches grows rapidly, becoming a major bottleneck in both training and inference. While prior works such as Multi-Query Attention (MQA) and Multi-Latent Attention (MLA) reduce memory by sharing or compressing KV features, they often trade off representation quality or incur runtime overhead. We propose Memory-Keyed Attention (MKA), a hierarchical attention mechanism that integrates multi-level KV caches (local, session, and long-term) and learns to route attention across them dynamically. We further introduce Route-Fused MKA (FastMKA), a broadcast-routed variant that fuses memory sources before attention computation for improved efficiency. Experiments on different sequence lengths show that FastMKA achieves a favorable accuracy-efficiency trade-off: comparable perplexity to MLA while achieving up to 5x faster training throughput and 1.8x lower evaluation latency. These results highlight MKA as a practical and extensible framework for efficient long-context attention.

Dong Liu, Yanxuan Yu, Ben Lengerich +1

University of California, Los Angeles, Los Angeles, California, USA · Columbia University, New York, New York, USA · University of Wisconsin-Madison, Madison, Wisconsin, USA

Attention MechanismsMemory-Efficient AttentionLong-Context Language ModelingKV-Cache ManagementLLM Inference AccelerationMemory-Augmented Language Models
AskPDF
Mar 20, 2026

Graph Signal Processing Meets Mamba2: Adaptive Filter Bank via Delta Modulation

State-space models (SSMs) offer efficient alternatives to attention with linear-time recurrence. Mamba2, a recent SSM-based language model, uses selective input gating and a multi-head structure, enabling parallel computation and strong benchmark performance. However, its multi-head recurrence operates independently without structured utilization or analysis. In this work, we propose a novel method called Hierarchical ADaptive filter bank for Efficient SSMs (HADES), a Graph Signal Processing (GSP)-inspired framework that reinterprets Mamba2 as an adaptive filter bank on a line graph. Our hierarchical architecture introduces two filter types: shared filters for global low-pass behavior and expert filters for local high-pass behavior, achieved through structured bias on the parameter Δ. HADES achieves comparable performance to baseline models including Mamba2 across various benchmarks in language modeling, commonsense reasoning, and long-context retrieval, while using only 58.9% of the original parameters. In this regard, HADES bridges GSP and neural sequence modeling, enabling efficient, hierarchical, and interpretable filtering within state-space models.

Yehjin Shin, Seojin Kim, Noseong Park

KAIST

Selective SSMsLong-Context Language Modeling
AskPDF
Mar 18, 2026

Learning When to Attend: Conditional Memory Access for Long-Context LLMs

Language models struggle to generalize beyond pretraining context lengths, limiting long-horizon reasoning and retrieval. Continued pretraining on long-context data can help but is expensive due to the quadratic scaling of Attention. We observe that most tokens do not require (Global) Attention over the entire sequence and can rely on local context. Based on this, we propose L2A (Learning To Attend), a layer that enables conditional (token-wise) long-range memory access by deciding when to invoke global attention. We evaluate L2A on Qwen 2.5 and Qwen 3 models, extending their effective context length from 32K to 128K tokens. L2A matches the performance of standard long-context training to within 3% while skipping Global Attention for ∼\sim80% of tokens, outperforming prior baselines. We also design custom Triton kernels to efficiently implement this token-wise conditional Attention on GPUs, achieving up to ∼\sim2×\times improvements in training throughput and time-to-first-token over FlashAttention. Moreover, L2A enables post-training pruning of highly sparse Global Attention layers, reducing KV cache memory by up to 50% with negligible performance loss. Our code is released under Apache 2.0 at https://github.com/awslabs/hybrid-model-factory/tree/main/examples/research/L2A.

Sakshi Choudhary, Aditya Chattopadhyay, Luca Zancato +4

Department of Electrical and Computer Engineering, Purdue University, USA · AWS Agentic AI, USA

Self-AttentionLong-Context Language ModelingLLM Inference AccelerationSparse AttentionAttention MechanismsLong-Context Reasoning in Language Models
AskPDF
Mar 15, 2026

M2^2RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling

Transformers are highly parallel but are limited to computations in the TC0^0 complexity class, excluding tasks such as entity tracking and code execution that provably require greater expressive power. Motivated by this limitation, we revisit non-linear Recurrent Neural Networks (RNNs) for language modeling and introduce Matrix-to-Matrix RNN (M2^2RNN): an architecture with matrix-valued hidden states and expressive non-linear state transitions. We demonstrate that the language modeling performance of non-linear RNNs is limited by their state size, and show how the state size expansion mechanism enables efficient use of tensor cores. Empirically, M2^2RNN achieves perfect state tracking generalization at sequence lengths not seen during training. These benefits also translate to large-scale language modeling. In hybrid settings that interleave recurrent layers with attention, Hybrid M2^2RNN outperforms equivalent Gated DeltaNet hybrids by 0.40.4-0.50.5 perplexity points on a 7B MoE model, while using 3×3\times smaller state sizes for the recurrent layers. Notably, replacing even a single recurrent layer with M2^2RNN in an existing hybrid architecture yields accuracy gains comparable to Hybrid M2^2RNN with minimal impact on training throughput. Further, the Hybrid Gated DeltaNet models with a single M2^2RNN layer also achieve superior long-context generalization, outperforming state-of-the-art hybrid linear attention architectures by up to 88 points on LongBench. Together, these results establish non-linear RNN layers as a compelling building block for efficient and scalable language models.

Mayank Mishra, Shawn Tan, Ion Stoica +2

UC Berkeley · Learning Machine · MIT-IBM Watson Lab +2

Recurrent Neural NetworksLanguage ModelingAutoregressive Language ModelingLong-Context Language Modeling
AskPDF
Mar 12, 2026

LPC-SM: Local Predictive Coding and Sparse Memory for Long-Context Language Modeling

Most current long-context language models still rely on attention to handle both local interaction and long-range state, which leaves relatively little room to test alternative decompositions of sequence modeling. We propose LPC-SM, a hybrid autoregressive architecture that separates local attention, persistent memory, predictive correction, and run-time control within the same block, and we use Orthogonal Novelty Transport (ONT) to govern slow-memory writes. We evaluate a 158M-parameter model in three stages spanning base language modeling, mathematical continuation, and 4096-token continuation. Removing mHC raises the Stage-A final LM loss from 12.630 to 15.127, while adaptive sparse control improves the Stage-B final LM loss from 12.137 to 10.787 relative to a matched fixed-ratio continuation. The full route remains stable at sequence length 4096, where Stage C ends with final LM loss 11.582 and improves the delayed-identifier diagnostic from 14.396 to 12.031 in key cross-entropy. Taken together, these results show that long-context autoregressive modeling can be organized around a broader division of labor than attention alone.

Keqin Xie

Independent Researcher, Suzhou, China

Long-Context Language ModelingAutoregressive Language ModelingLong-Range Dependency ModelingMemory-Augmented Language ModelsPersistent Memory for Language ModelsPredictive Coding
AskPDF
Mar 12, 2026

Long-Context Encoder Models for Polish Language Understanding

While decoder-only Large Language Models (LLMs) have recently dominated the NLP landscape, encoder-only architectures remain a cost-effective and parameter-efficient standard for discriminative tasks. However, classic encoders like BERT are limited by a short context window, which is insufficient for processing long documents. In this paper, we address this limitation for the Polish by introducing a high-quality Polish model capable of processing sequences of up to 8192 tokens. The model was developed by employing a two-stage training procedure that involves positional embedding adaptation and full parameter continuous pre-training. Furthermore, we propose compressed model variants trained via knowledge distillation. The models were evaluated on 25 tasks, including the KLEJ benchmark, a newly introduced financial task suite (FinBench), and other classification and regression tasks, specifically those requiring long-document understanding. The results demonstrate that our model achieves the best average performance among Polish and multilingual models, significantly outperforming competitive solutions in long-context tasks while maintaining comparable quality on short texts.

Sławomir Dadas, Rafał Poświata, Marek Kozłowski +4

National Information Processing Institute, Warsaw, Poland · PKO Bank Polski, Warsaw, Poland

Long-Context Language ModelingLong-Context ModelingLanguage Model PretrainingLanguage ModelingLong-Context Language Model Evaluation
AskPDF
Mar 9, 2026

LinearARD: Linear-Memory Attention Distillation for RoPE Restoration

The extension of context windows in Large Language Models is typically facilitated by scaling positional encodings followed by Continued Pre-Training (CPT). While effective, this paradigm is notoriously data-hungry and computationally expensive, requiring massive long-text corpora to recalibrate the model to the shifted positional distribution. We propose LinearARD, a restoration distillation method that restores Rotary Position Embedding (RoPE)-scaled students through attention-structure consistency with a frozen native-RoPE teacher. Rather than next-token prediction or opaque hidden-state matching, LinearARD aligns row-wise distributions of dense Q/QQ/Q, K/KK/K, and V/VV/V self-relation matrices from the final attention layer to directly supervise attention dynamics. To remove the quadratic memory bottleneck of n×nn \times n relation maps, we introduce a linear-activation-memory kernel that stores only per-token log-sum-exp statistics and recomputes logits in the backward pass to obtain exact Kullback-Leibler divergence gradients. Across LLaMA2-7B, LLaMA3-8B, and Mistral-7B-v0.1 extended to 32K context, LinearARD recovers 93.1%/94.2%/94.3% of native short-context performance and achieves strong long-context robustness on RULER using only 4.25M tokens---amounting to just 1.6% (a ∼\sim60×\times reduction) of the 256M-token budget required by state-of-the-art baselines. Under this severely constrained budget, CPT and LongReD remain near zero on RULER, demonstrating that relation-level restoration provides a substantially better efficiency-quality tradeoff. Our code is available at https://github.com/gracefulning/LinearARD.

Ning Yang, Hengyu Zhong, Wentao Wang +4

Institute of Automation, Chinese Academy of Sciences, Beijing, China · Southwest University, Chongqing, China · Dalian University of Technology, Dalian, Liaoning, China +3

Attention-Guided DistillationMemory-Efficient AttentionLong-Context Language ModelingLanguage Model DistillationKnowledge DistillationEfficient Attention+1
AskPDF
Mar 6, 2026

MSA: Memory Sparse Attention for Efficient End-to-End Memory Model Scaling to 100M Tokens

Long-term memory is a cornerstone of human intelligence. Enabling AI to process lifetime-scale information remains a long-standing pursuit in the field. Due to the constraints of full-attention architectures, the effective context length of large language models (LLMs) is typically limited to 1M tokens. Existing approaches, such as hybrid linear attention, fixed-size memory states (e.g., RNNs), and external storage methods like RAG or agent systems, attempt to extend this limit. However, they often suffer from severe precision degradation and rapidly increasing latency as context length grows, an inability to dynamically modify memory content, or a lack of end-to-end optimization. These bottlenecks impede complex scenarios like large-corpus summarization, Digital Twins, and long-history agent reasoning, while limiting memory capacity and slowing inference. We present Memory Sparse Attention (MSA), an end-to-end trainable, efficient, and massively scalable memory model framework. Through core innovations including scalable sparse attention and document-wise RoPE, MSA achieves linear complexity in both training and inference while maintaining exceptional stability, exhibiting less than 9% degradation when scaling from 16K to 100M tokens. Furthermore, KV cache compression, combined with Memory Parallel, enables 100M-token inference on 2xA800 GPUs. We also propose Memory Interleaving to facilitate complex multi-hop reasoning across scattered memory segments. MSA significantly surpasses frontier LLMs, state-of-the-art RAG systems, and leading memory agents in long-context benchmarks. These results demonstrate that by decoupling memory capacity from reasoning, MSA provides a scalable foundation to endow general-purpose models with intrinsic, lifetime-scale memory.

Yu Chen, Runkai Chen, Sheng Yi +9

Evermind · Shanda Group · Peking University

Memory-Augmented Language ModelsLong-Context Language ModelingSparse AttentionLLM Inference Acceleration
AskPDF
Mar 5, 2026

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

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~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, \modelnameeffectively generalizes to inputs exceeding 128K tokens. Across a comprehensive suite of long-context modeling and understanding benchmarks, \modelnameachieves 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 (2×2\times over streaming and 3×3\times over encoder-decoder architectures).

Wei Han, Pan Zhou, Soujanya Poria +1

Singapore University of Technology and Design (SUTD), Singapore · Singapore Management University (SMU), Singapore · Nanyang Technological University (NTU), Singapore +1

Long-Context Language Model InferenceLong-Context Language ModelingLLM Context ManagementContext CompressionLLM Inference Acceleration
AskPDF
Mar 3, 2026

MaBERT:A Padding Safe Interleaved Transformer Mamba Hybrid Encoder for Efficient Extended Context Masked Language Modeling

Self attention encoders such as Bidirectional Encoder Representations from Transformers(BERT) scale quadratically with sequence length, making long context modeling expensive. Linear time state space models, such as Mamba, are efficient; however, they show limitations in modeling global interactions and can suffer from padding induced state contamination. We propose MaBERT, a hybrid encoder that interleaves Transformer layers for global dependency modeling with Mamba layers for linear time state updates. This design alternates global contextual integration with fast state accumulation, enabling efficient training and inference on long inputs. To stabilize variable length batching, we introduce paddingsafe masking, which blocks state propagation through padded positions, and mask aware attention pooling, which aggregates information only from valid tokens. On GLUE, MaBERT achieves the best mean score on five of the eight tasks, with strong performance on the CoLA and sentence pair inference tasks. When extending the context from 512 to 4,096 tokens, MaBERT reduces training time and inference latency by 2.36x and 2.43x, respectively, relative to the average of encoder baselines, demonstrating a practical long context efficient encoder.

Jinwoong Kim, Sangjin Park

Graduate School of Industrial Data Engineering, Hanyang University, Seoul, Republic of Korea

Hybrid Transformer-State Space Model ArchitecturesLong-Context Language ModelingHybrid Mamba-Transformer ArchitecturesSequence ModelingLong-Context Language Model InferenceMasked Language Modeling+1
AskPDF
Mar 2, 2026

Understanding the Physics of Key-Value Cache Compression for LLMs through Attention Dynamics

As context windows in LLMs scale to 100K+ tokens, the key-value (KV) cache becomes the dominant memory bottleneck, with recent methods claiming 80-90% savings and minimal benchmark degradation. We argue these evaluations miss a structural issue: attention is not just storage but routing, and retaining KV pairs does not guarantee semantic accessibility. We propose a physics-inspired view of KV compression as a controlled perturbation of token-level routing, distinguishing retention, accessibility, and utilization. Using synthetic tasks probing multi-entity tracking, disambiguation, coreference, and multi-hop reasoning, we find that moderate compression degrades internal representations with little accuracy loss, revealing redundancy; all models exhibit a sharp hallucination safety cliff near 90% compression, correlated with spikes in Global Eviction Ratio (GER), suggesting a phase transition in semantic reachability; and architectures differ in routing dynamics, with LLaMA showing early consensus and late diversification, and Qwen showing funnel-like late convergence, leading to distinct resilience profiles. Beyond erasure, we identify representational rigidity, where excessive head-level consensus collapses routing flexibility despite token survival. These results suggest sparse token-route structures govern compression tolerance, reframing KV compression as a structural probe of attention geometry and linking long-context scalability to sparsity and the lottery ticket hypothesis in self-attention.

Samhruth Ananthanarayanan, Ayan Sengupta, Tanmoy Chakraborty

IIT Delhi, India

Attention RoutingKV-Cache CompressionLong-Context Language ModelingRepresentation Geometry in Language Models
AskPDF
Feb 27, 2026

Memory Caching: RNNs with Growing Memory

Transformers have been established as the de-facto backbones for most recent advances in sequence modeling, mainly due to their growing memory capacity that scales with the context length. While plausible for retrieval tasks, it causes quadratic complexity and so has motivated recent studies to explore viable subquadratic recurrent alternatives. Despite showing promising preliminary results in diverse domains, such recurrent architectures underperform Transformers in recall-intensive tasks, often attributed to their fixed-size memory. In this paper, we introduce Memory Caching (MC), a simple yet effective technique that enhances recurrent models by caching checkpoints of their memory states (a.k.a. hidden states). Memory Caching allows the effective memory capacity of RNNs to grow with sequence length, offering a flexible trade-off that interpolates between the fixed memory (i.e., O(L)O(L) complexity) of RNNs and the growing memory (i.e., O(L2)O(L^2) complexity) of Transformers. We propose four variants of MC, including gated aggregation and sparse selective mechanisms, and discuss their implications on both linear and deep memory modules. Our experimental results on language modeling, and long-context understanding tasks show that MC enhances the performance of recurrent models, supporting its effectiveness. The results of in-context recall tasks indicate that while Transformers achieve the best accuracy, our MC variants show competitive performance, close the gap with Transformers, and performs better than state-of-the-art recurrent models.

Ali Behrouz, Zeman Li, Yuan Deng +3

Recurrent Neural NetworksMemory-Augmented Language ModelsLong-Context Language Modeling
AskPDF
Feb 24, 2026

Frayed RoPE and Long Inputs: A Geometric Perspective

Rotary Positional Embedding (RoPE) is a widely adopted technique for encoding position in language models, which, while effective, causes performance breakdown when input length exceeds training length. Prior analyses assert (rightly) that long inputs cause channels to rotate ``out of distribution,'' but it is not clear how extra rotation relates to or causes pathological behavior. Through empirical and theoretical analysis we advance a unified geometric understanding of attention behavior with RoPE. We find that attention induces tight clustering of separated key and query latent point clouds, allowing for creation of sink tokens: placeholders that allow attention heads to avoid token mixing when not required. RoPE applied to longer inputs damages this key/query cluster separation, producing pathological behavior by inhibiting sink token functionality. From this geometric perspective, we propose RoPE-ID (In Distribution), a straightforward modification that allows attention layers to generalize to longer inputs out of the box: apply RoPE with high frequency to a subset of channels. We demonstrate the effectiveness of RoPE-ID for extended inputs using 1B and 3B parameter Transformers on the LongBench and RULER information retrieval benchmarks.

Davis Wertheimer, Aozhong Zhang, Derrick Liu +2

IBM Research · University at Albany, SUNY

Rotary Positional EmbeddingsLong-Context Language ModelingLong-Context ModelingRepresentation Geometry in Language ModelsAttention Sinks
AskPDF
Feb 23, 2026

Whole-Piece Training for Symbolic Music Language Models via Full-Horizon Compressed Recurrence

For computational efficiency, modern language models are typically trained on independently sampled fixed-length sequences. Symbolic music language models largely inherit this paradigm, despite musical structure naturally unfolding over complete compositions rather than isolated excerpts. Fragmenting compositions into independent training instances therefore prevents continuous conditioning over the complete work. We present a practical framework for whole-piece training of symbolic music language models via Full-Horizon Compressed Recurrence (FHCR). FHCR preserves the full temporal horizon of recurrent memory while reducing the dimensionality of its key-value (KV) representation, making continuous whole-piece training practical under limited GPU memory. To directly assess functional long-range dependence, we introduce KV-Reset Context Utilization (KRCU), an evaluation-time diagnostic. On the MAESTRO symbolic piano dataset, KRCU shows that full-horizon models utilize context far beyond the local segment window, whereas reducing the temporal extent of recurrent memory substantially weakens this measurable long-range dependence. FHCR preserves long-range context utilization while substantially reducing recurrent memory cost. These findings show that preserving the temporal extent of recurrent history is important for efficient whole-piece modeling, and that memory cost can instead be reduced through KV representation compression. The project demos and generated music samples are available at https://wholemusic.github.io.

Yungang Yi, Weihua Li, Matthew Kuo +2

Auckland University of Technology, Auckland, New Zealand · University of Tasmania, Tasmania, Australia

Long-Context Language ModelingSymbolic Music GenerationMusic GenerationKV-Cache Compression
AskPDF
Feb 18, 2026

Reinforced Fast Weights via Next-Sequence Prediction

Fast weight architectures offer a promising alternative to standard transformers for long-context modeling by replacing the KV-cache with a recurrently updated fixed-size memory. Despite this architectural shift, they are typically trained with the same next-token prediction (NTP) objective as standard transformers. This creates a mismatch: fast weight models rely on an evolving memory state to support future predictions, while NTP provides only token-level supervision for the immediate next token. We address this mismatch with next-sequence prediction (NSP), a sequence-level extension of NTP that trains models to produce coherent multi-token continuations from their recurrent memory state. To optimize this sequence-level objective, we propose ReFINE (Reinforced Fast weIghts via Next sEquence prediction). ReFINE selects informative token positions based on prediction entropy, generates multi-token rollouts, assigns sequence-level rewards, and optimizes the model with Group Relative Policy Optimization (GRPO). ReFINE is applicable throughout the training lifecycle of pre-trained language models: mid-training, post-training, and test-time training. Experiments on LaCT-760M and DeltaNet-1.3B show that ReFINE consistently outperforms NTP-based supervised fine-tuning in needle-in-a-haystack retrieval, long-context question answering, and diverse tasks in LongBench. ReFINE provides an effective and versatile framework for improving long-context modeling in fast weight architectures.

Hee Seung Hwang, Xindi Wu, Sanghyuk Chun +2

Princeton University

Fast WeightsLong-Context Language ModelingNext-Token PredictionRL for Language Models
AskPDF
Feb 18, 2026

Fast KV Compaction via Attention Matching

Scaling language models to long contexts is often bottlenecked by the size of the key-value (KV) cache. In deployed settings, long contexts are typically managed through compaction in token space via summarization. However, summarization can be highly lossy, substantially harming downstream performance. Recent work on Cartridges has shown that it is possible to train highly compact KV caches in latent space that closely match full-context performance, but at the cost of slow and expensive end-to-end optimization. This work describes an approach for fast context compaction in latent space through Attention Matching, which constructs compact keys and values to reproduce attention outputs and preserve attention mass at a per-KV-head level. We show that this formulation naturally decomposes into simple subproblems, some of which admit efficient closed-form solutions. Within this framework, we develop a family of methods that significantly push the Pareto frontier of compaction time versus quality, achieving up to 50x compaction in seconds on some datasets with little quality loss.

Adam Zweiger, Xinghong Fu, Han Guo +1

Massachusetts Institute of Technology.

KV-Cache ManagementLLM Context CompressionKV-Cache CompressionLong-Context Language ModelingLLM Inference Acceleration
AskPDF