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.

May 23, 2026

Interdomain Attention: Beyond Token-Level Key-Value Memory

Softmax attention-based transformers and recurrent memory-based models (e.g., state space models (SSMs)) form the foundation of modern sequence models. Hybrid architectures aim to combine the expressivity from attention and the computational efficiency from recurrent memory, yet so far common practices simply stack both attention and recurrent layers. We propose Interdomain Attention (IA) as a novel hybrid attention architecture that mixes attention and recurrent memory in the same layer. Motivated by a kernel regression view of softmax attention, IA compresses values and kernel features of keys into a recurrent memory parameterized by an SSM. Then IA approximates dot-product attention by implicitly reconstructing the attention matrix and values from memory states; importantly this construction maintains SSM's computational efficiency advantage. We demonstrate IA's efficacy by designing and pre-training IA network architectures, where ablation studies identify attention-like readout as the primary source of gain over SSMs. Across 125M-1.3B models pre-trained on text data, IA improves over an S4D baseline at matched memory size. It also outperforms softmax attention-based Transformer in perplexity and accuracy on common-sense reasoning tasks. Based on Caduceus backbone, IA matches or outperforms the original SSM-based architecture in both perplexity and downstream tasks on genomic data.

Naoki Kiyohara, Harrison Bo Hua Zhu, Riccardo El Hassanin +4

Imperial College London, UK · Canon Inc., Japan · University of Copenhagen, Denmark +2

Self-AttentionLong-Context Language ModelingEfficient AttentionLinear AttentionState Space Models
AskPDF
May 21, 2026

Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention

Linear attention replaces the unbounded cache of softmax attention with a fixed-size recurrent state, reducing sequence mixing to linear time and decoding to constant memory. The hard part is not just what to forget, but how to edit this compressed memory without scrambling existing associations. Delta-rule models subtract the current read before writing a new value, and Kimi Delta Attention (KDA) sharpens forgetting with channel-wise decay. But the active edit still uses a single scalar gate to control two different things: how much old content to erase on the key side and how much new content to commit on the value side. We introduce Gated DeltaNet-2, which generalizes both Gated DeltaNet and KDA by inheriting adaptive forgetting and channel-wise decay while addressing their shared limitation, the scalar tie between erasing and writing. Gated Delta Rule-2 separates these roles with a channel-wise erase gate b_t and a channel-wise write gate w_t, reducing to KDA when both gates collapse to the same scalar and to Gated DeltaNet when the decay also collapses. We derive a fast-weight update view, a chunkwise WY algorithm with channel-wise decay absorbed into asymmetric erase factors, and a gate-aware backward pass that preserves efficient parallel training. At 1.3B parameters trained on 100B FineWeb-Edu tokens, Gated DeltaNet-2 achieves the strongest overall results among Mamba-2, Gated DeltaNet, KDA, and Mamba-3 variants across language modeling, commonsense reasoning, and retrieval. Its advantage is most pronounced on long-context RULER needle-in-a-haystack benchmarks, where it improves the evaluated multi-key retrieval setting and remains strong in both recurrent and hybrid settings. Code is available at https://github.com/NVlabs/GatedDeltaNet-2.

Ali Hatamizadeh, Yejin Choi, Jan Kautz

NVIDIA

Long-Context Language ModelingRecurrent TransformersLinear AttentionDelta-Rule Linear AttentionGated Delta Networks
AskPDF
May 19, 2026

Efficient Long-Context Modeling in Diffusion Language Models via Block Approximate Sparse Attention

Diffusion Language Models (DLMs) enable globally coherent, bidirectional, and controllable text generation, offering advantages over traditional autoregressive LLMs, while scaling to ultra-long sequences remains costly. Many existing block-sparse attention methods select blocks by fixed sampling patterns over the high-resolution attention space, such as tail regions or anti-diagonal stripes. Such prior-driven sampling can miss salient tokens and introduce instability under distribution shifts. In this paper, we propose the Block Approximate Sparse Attention framework (BA-Att) with block-wise pre-downsampled operation, which identifies informative regions within a compact downsampled space, avoiding reliance on brittle positional priors. To analyze its theoretical behavior, we define an oracle post-downsample attention map and formalize the approximation error between pre- and post-downsample schemes. Based on this insight, we introduce a lightweight norm-sorting module and a covariance-compensated correction that approximates full covariance using diagonal QK variances, reducing computational complexity. Extensive experiments show that our operator achieves up to 6.95x acceleration over FlashAttention in attention computation, and maintains near full-attention performance at 50% sparsity across language models, multimodal language models, and video generation models, demonstrating strong efficiency and generalization.

Wenhu Zhang, Yiming Wu, Huanyu Wang +6

The Hong Kong University of Science and Technology · The University of Hong Kong · Zhejiang University +1

Self-AttentionLong-Context Language ModelingSparse AttentionDiffusion Language ModelsDiffusion Model Inference AccelerationBlock-Sparse Attention
AskPDF
May 18, 2026

Elastic-dLLM: Position Preserving Context Compression and Augmentation of Diffusion LLMs

Unlike autoregressive models, which generate one token at a time, dLLMs denoise a chunk of [MASK] tokens jointly and sample one or more tokens per step; despite enabling parallel decoding, this process incurs substantial computational cost due to the large chunk size of masked tokens. We observe that much of this cost is spent on repeatedly processing the preceding context and many [MASK] tokens with the same feature representations, indicating considerable computational redundancy. In this work, we revisit dLLM's redundancy from the perspective of [MASK] tokens. Through systematic analysis, we verify the redundancy of [MASK] tokens while revealing their critical role in providing structural information. Guided by these findings, we propose position-preserving [MASK] token compression and terminal-aware augmentation. By compressing redundant [MASK] computation, this approach accelerates decoding and further provides a natural extension toward context-folding-like long-context scaling under limited input-length constraints for full-sequence dLLMs such as LLaDA-8B-Instruct and LLaDA-1.5. Moreover, for block dLLMs such as LLaDA2.0-mini, it augments the context with a protected terminal [MASK] token to enhance generation quality with negligible overhead.

Junyi Wu, Tianchen Zhao, Shaoqiu Zhang +3

Tsinghua University · Shanghai Jiao Tong University · Infinigence AI

Masked Diffusion ModelsLong-Context Language ModelingDiffusion Language Model InferenceDiffusion Language ModelsDiffusion Model Inference Acceleration
AskPDF
May 15, 2026

RoPE Distinguishes Neither Positions Nor Tokens in Long Contexts, Provably

We identify intrinsic limitations of Rotary Positional Embeddings (RoPE) in Transformer-based long-context language models. Our theoretical analysis abstracts away from the specific content of the context and depends only on its length. We prove that as context length increases, RoPE-based attention becomes unpredictable and loses two properties that are central to its effectiveness. First, it loses its locality bias: RoPE is no more likely to favor nearer positions than substantially farther ones. Second, it loses consistency in token relevance: a key vector that receives a higher attention score than an alternative at one position may receive a lower score at another. In both cases, the probability of failure approaches 0.5, no better than random guessing. We further prove that the attention score can remain unchanged when a key token is moved to a different position, or even replaced by a different token, indicating a failure to distinguish positions or tokens. Adjusting the RoPE base trades off distinguishing positions against distinguishing tokens but cannot preserve both at the same time. Increasing the RoPE base hyperparameter, a common practice in today's long-context models, helps distinguish different tokens, but inevitably sacrifices the ability to distinguish positions. Our empirical analysis shows that multi-head, multi-layer architectures are insufficient to overcome these limitations. Our findings suggest that fundamentally new mechanisms for encoding position and token order may be needed in future Transformer long-context language models.

Yufeng Du, Phillip Harris, Minyang Tian +5

University of Illinois at Urbana-Champaign · USA · University of Bonn +3

Rotary Positional EmbeddingsSelf-AttentionLong-Context Language ModelingTransformer AttentionPositional Encoding
AskPDF
May 14, 2026

EndPrompt: Efficient Long-Context Extension via Terminal Anchoring

Extending the context window of large language models typically requires training on sequences at the target length, incurring quadratic memory and computational costs that make long-context adaptation expensive and difficult to reproduce. We propose EndPrompt, a method that achieves effective context extension using only short training sequences. The core insight is that exposing a model to long-range relative positional distances does not require constructing full-length inputs: we preserve the original short context as an intact first segment and append a brief terminal prompt as a second segment, assigning it positional indices near the target context length. This two-segment construction introduces both local and long-range relative distances within a short physical sequence while maintaining the semantic continuity of the training text--a property absent in chunk-based simulation approaches that split contiguous context. We provide a theoretical analysis grounded in Rotary Position Embedding and the Bernstein inequality, showing that position interpolation induces a rigorous smoothness constraint over the attention function, with shared Transformer parameters further suppressing unstable extrapolation to unobserved intermediate distances. Applied to LLaMA-family models extending the context window from 8K to 64K, EndPrompt achieves an average RULER score of 76.03 and the highest average on LongBench, surpassing LCEG (72.24), LongLoRA (72.95), and full-length fine-tuning (69.23) while requiring substantially less computation. These results demonstrate that long-context generalization can be induced from sparse positional supervision, challenging the prevailing assumption that dense long-sequence training is necessary for reliable context-window extension. The code is available at https://github.com/clx1415926/EndPrompt.

Han Tian, Luxuan Chen, Xinran Chen +10

1Nankai University · 2Baidu Inc. · 3Shanghai Jiao Tong University +1

Long-Context Language ModelingLLM Fine-TuningLLM Context Management
AskPDF
May 13, 2026

A Hierarchical Language Model with Predictable Scaling Laws and Provable Benefits of Reasoning

We introduce a family of synthetic languages with hierarchical structure -- generated by a broadcast process on trees -- for which the role of context length and reasoning in autoregressive generation can be analyzed precisely. At the heart of our analytic approach is an \emph{exact kk-gram ansatz} in place of transformers with context length kk, a substitution we then validate empirically. Using this ansatz we derive explicit asymptotic predictions for distributional statistics of the sequences produced by a trained model, instantiated in two settings. For the Ising broadcast process (a soft-constrained language), we prove that the variance of the generated sum scales log-linearly in the context depth and its kurtosis converges to that of a Gaussian -- both deviating from the true language for any sublinear context. For the coloring broadcast process (a hard-constrained language) in the freezing regime, bounded-context autoregression produces sequences that, with high probability, are inconsistent with any valid coloring of the underlying tree. Together these results imply an Ω(n)Ω(n) lower bound on the context length required to faithfully sample length-nn sequences. In contrast, we prove that an autoregressive reasoning model with only Θ(log⁡n)Θ(\log n) working memory can sample exactly from the true language -- an exponential improvement. We confirm both the lower-bound predictions and the reasoning-based upper bound empirically with transformers trained on the synthetic language; the trained models track our asymptotic predictions quantitatively across a wide range of context sizes.

Jason Gaitonde, Frederic Koehler, Elchanan Mossel +2

Duke University · University of Chicago · Massachusetts Institute of Technology +1

Long-Context Language ModelingLanguage Model Scaling LawsAutoregressive Language ModelingLanguage Modeling
AskPDF
May 13, 2026

Phasor Memory Networks: Stable Backpropagation Through Time for Scalable Explicit Memory

For over a decade, explicit memory architectures like the Neural Turing Machine have remained theoretically appealing yet practically intractable for language modeling due to catastrophic gradient instability during Backpropagation Through Time. In this work, we break this stalemate with Phasor Memory Network (PMNet), a novel architecture that structurally resolves memory volatility through Unitary Phasor Dynamics and Hierarchical Learnable Anchors. Rather than relying on brute-force scaling, we present a mechanistic proof-of-concept in a controlled byte-level setting. By constraining recurrent state updates to phase rotations on a complex unit circle, PMNet preserves gradient norms and inherently prevents divergence without the need for specialized initialization. We empirically demonstrate the active actuation of the memory module through a synthetic Copy-Paste task, where PMNet utilizes an expansive 85-slot hierarchical memory tree (=∑h=144h−1=\sum^{4}_{h=1}4^{h-1}) to achieve near 100% exact retrieval across temporal distances that completely exceed the local sliding window attention's receptive field. Furthermore, despite being a compact 119M parameter model trained on 18.8B tokens, PMNet matches the zero-shot long-context robustness of a Mamba model that is three times larger. Our ablation studies and gradient analyses confirm that the historical failure of explicit memory was a structural alignment problem, which PMNet effectively overcomes, providing a theoretically grounded foundation for scalable sequence modeling.

Sungwoo Goo, Hwi-yeol Yun, Sangkeun Jung

College of Pharmacy Chungnam National University Daejeon, South Korea · Department of Computer Science & Engineering Chungnam National University Daejeon, South Korea

Memory-Augmented Neural NetworksMemory-Augmented Language ModelsLong-Context Language ModelingPersistent Memory for Language ModelsRecurrent Neural NetworksMemorization in Generative Models
AskPDF
May 12, 2026

Training-Inference Consistent Segmented Execution for Long-Context LLMs

Transformer-based large language models face severe scalability challenges in long-context generation due to the computational and memory costs of full-context attention. Under practical computation and memory constraints, many inference-efficient long-context methods improve efficiency by adopting bounded-context or segment-level execution only during inference, while continuing to train models under full-context attention, resulting in a mismatch between training and inference execution and state-transition semantics. Based on this insight, we propose a training-inference consistent segment-level generation framework, in which training and inference follow the same segment-level forward execution semantics. During training, consistency with inference is enforced by restricting gradient propagation to KV states carried over from the immediately preceding segment, while permitting head-specific access to past KV states during the forward pass without involving them in gradient propagation. Across long-context benchmarks, our approach achieves performance comparable to full-context attention, while achieving competitive latency-memory trade-offs against strong inference-efficient baselines, and substantially improving scalability at very long context lengths (e.g., approximately 6x lower peak prefill memory at 128K compared to full-context attention with FlashAttention).

Xianpeng Shang, Jiang Li, Zehua Duo +2

College of Computer Science, Inner Mongolia University, Hohhot 010021, China · National & Local Joint Engineering Research Center of Intelligent Information Processing Technology for Mongolian, Hohhot 010021, China · Inner Mongolia Key Laboratory of Multilingual Artificial Intelligence Technology, Hohhot 010021, China +1

Self-AttentionLong-Context Language ModelingLong-Context InferenceEfficient Language Model Inference
AskPDF
May 11, 2026

Where Does Long-Context Supervision Actually Go? Effective-Context Exposure Balancing

Long-context adaptation is often viewed as window scaling, but this misses a token-level supervision mismatch: in packed training with document masking, each target token's effective context remains short. We introduce EXACT, a supervision-allocation objective that assigns extra weight to long effective-context targets by inverse frequency within the long tail. Across seven Qwen/LLaMA CPT configurations, EXACT improves all 28 trained/extrapolated NoLiMa and RULER comparisons. On Qwen2.5-0.5B, NoLiMa improves by +10.09 (trained) and +5.34 (extrapolated); RULER by +10.69 and +5.55. On LLaMA-3.2-3B, RULER improves by +17.91 and +16.11. Standard QA/reasoning are preserved (+0.24 macro change across six benchmarks). A distance-resolved probe shows gains arise when evidence is thousands of tokens away, while short cases remain unchanged. Results support a supervision-centric thesis: long-context adaptation depends on how strongly training supervises long-context predictions.

Jinchang Zhu, Jindong Li, Chengyu Zou +4

1The Hong Kong University of Science and Technology (Guangzhou) · Institute of Artificial Intelligence (TeleAI), China Telecom

Long-Context Language ModelingAdaptive Loss WeightingContinued Pretraining
AskPDF
May 11, 2026

Mela: Test-Time Memory Consolidation based on Transformation Hypothesis

Memory consolidation, the process by which transient experiences are transformed into stable, structured representations, is a foundational organizing principle in the human brain, yet it remains largely unexplored as a design principle for modern sequence models. In this work, we leverage established neuroscientific theories of memory consolidation and cross-frequency coupling to propose the Hierarchical Memory Module (HMM), a neural memory architecture composed of two functionally distinct sub-modules that operate at different update frequencies. Inspired by the transformation hypothesis, the low-frequency sub-module produces high-level representations that capture abstract, gist-level knowledge, while the high-frequency sub-module produces fine-grained representations that preserve richer episodic detail. The final memory output is dynamically reconstructed as a context-dependent combination of both representations, analogous to the reconstructive nature of human memory retrieval. We integrate HMM into a Transformer-based language decoder to form Mela, a family of memory-augmented language models that perform online memory consolidation at test time. To further exploit the multi-granularity memory representations produced by HMM, we introduce MemStack, a method that distributes different levels of memory features across the early layers of the decoder without introducing additional tokens. Experiments on language modeling demonstrate that Mela outperforms Transformer baselines across all the model sizes. Moreover, with the pretrained context length fixed at 4K, Mela maintains performance on significantly longer contexts, whereas Transformer baselines degrade rapidly beyond their training length. Extensive ablation studies validate the contribution of each component and provide guidance for practical configuration.

Lungchuan Chen

MusubiAI

Memory-Augmented Neural NetworksMemory-Augmented Language ModelsLong-Context Language Modeling
AskPDF
May 11, 2026

Remember to Forget: Gated Adaptive Positional Encoding

Rotary Positional Encoding (RoPE) is widely used in modern large language models. However, when sequences are extended beyond the range seen during training, rotary phases can enter out-of-distribution regimes, leading to spurious long-range alignments, diffuse attention, and degraded retrieval. Existing remedies only partially address these failures, as they often trade local positional resolution for long-context stability. We propose GAPE (Gated Adaptive Positional Encoding), a drop-in augmentation for positional encodings that introduces a content-aware bias directly into the attention logits while preserving the rotary geometry. GAPE decouples distance-based suppression from token importance through a query-dependent gate that contracts irrelevant context and a key-dependent gate that preserves salient distant tokens. We show that weakly protected distant context is exponentially attenuated as a function of the query gate, while selected keys can remain accessible through landmark protection. We further show that GAPE can be implemented within standard scaled dot-product attention. Empirically, GAPE improves long-context robustness across controlled retrieval and language-modeling experiments, extrapolating up to 8x the training length. We further retrofit GAPE into a pretrained 7B model, maintaining performance on standard benchmarks and improving performance at the longest evaluated context. These results support adaptive context suppression as a complement to positional representation for long-context generalization.

Riccardo Ali, Alessio Borgi, Mario Severino +2

Department of Computer Science and Technology, University of Cambridge, United Kingdom · Department of Computer, Control and Management Engineering, Sapienza University, Italy · Department of Information Engineering, University of Padova, Italy

Rotary Positional EmbeddingsSelf-AttentionLong-Context Language ModelingTransformer AttentionPositional Encoding
AskPDF
May 11, 2026

Key-Value Means: Transformers with Expandable Block-Recurrent Compressed Memory

We present Key-Value Means ("KVM"), a novel block-recurrence for attention that can accommodate either fixed-size or growing state. Equipping a strong transformer baseline with fixed-size KVM attention layers yields a strong O(N)O(N) chunked RNN, while adding only an insignificant number of new parameters. We train a transformer with a growable KVM cache and show it performs competitively on long-context tests with only subquadratic prefill time and sublinear state growth. KVM is implementable with standard operations and without custom kernels, and supports chunk-wise parallelizable training and prefill. It provides many of the benefits of both traditional transformers (expandable context memory, chunk-wise parallelizable training and prefill) and linear RNNs in a single unified package. It can be used on every layer, saving KV-cache memory, and allowing a continuous range of choices of prefill time complexity between O(N)O(N) and O(N2)O(N^2). It can also be implemented in a hybrid solution in tandem with LRNN layers in place of traditional attention, to supplement the LRNN with improved sublinear memory growth context length usage and long context decoding. We release our code at https://github.com/featherless-ai/KVM-paper and trained models at https://huggingface.co/collections/featherless-ai/kvm-paper under the Apache 2.0 license.

Daniel Goldstein, Eugene Cheah

Featherless AI · Eleuther AI

Transformer InferenceSelf-AttentionKV CachingLong-Context Language ModelingRecurrent TransformersAttention Mechanisms
AskPDF
May 10, 2026

Positional LSH: Binary Block Matrix Approximation for Attention with Linear Biases

Positional encoding in transformers is commonly implemented through positional embeddings, attention masks, or bias terms, but formal connections between these mechanisms remain limited. We study attention with positional bias through the lens of locality-sensitive hashing (LSH), focusing on Attention with Linear Biases (ALiBi). We show that the ALiBi bias matrix is the expectation of contiguous block-diagonal binary masks induced by a ``positional LSH'' scheme. The empirical mean of masks sampled from this scheme yields spectral norm and max-norm approximation guarantees with bounded block sizes with high probability. This structural theorem implies a uniform approximation theorem for ALiBi-biased attention: with high probability over the sampled masks, the approximate attention output is accurate simultaneously for all query-key-value inputs and can be computed in near-linear time in the context length, reducing long-context ALiBi to a collection of randomized short-context regular (positionally unbiased) attention operations. Conceptually, this connects positional bias, masks, and positional embeddings in a single formal framework and suggests an approach to efficient ALiBi-biased attention. Experiments on large language models validate our theoretical findings.

Daniel Wolfson, Tal Wagner

Blavatnik School of Computer Science and AI, Tel Aviv University

Self-AttentionLong-Context Language ModelingEfficient AttentionPositional EncodingAttention Mechanisms
AskPDF
May 9, 2026

VORT: Adaptive Power-Law Memory for NLP Transformers

Standard Transformers impose near-exponential decay on the influence of distant tokens, conflicting with the power-law structure of long-range dependencies in natural language. We introduce the Variable-Order Retention Transformer (\VORT{}), a memory architecture in which each ingested token is assigned a learnable fractional order α_i\in[δ,1] that governs a Grünwald--Letnikov power-law retention kernel. Because the fractional weighted sum is non-Markovian, we approximate it through a sum-of-exponentials (SOE) decomposition computed by Gauss--Laguerre quadrature on a Laplace-type integral representation of the kernel weights. Each exponential component admits a one-step Markovian recurrence at O(Sd_v) per step, where S=O(\log(T/\varepsilon)) terms suffice for \varepsilon-uniform accuracy on horizon [1,T]. Retrieval is keyed and associative via a linear-attention accumulator with an exact O(KSd_φd_v) -per-step recurrence. Four results are established: (i) an SOE approximation theorem with geometric convergence rate from the analyticity of the integrand after a log-change of variables; (ii) a quantisation bound valid on [δ,1] with correct analysis near α=0; (iii) a direct L^2 energy argument (Proposition) showing that for α>1/2 any mixture with fixed minimum decay rate Λ>0 incurs L^2([1,T]) error at least N_α(T)-C(Λ)\to\infty, with the Λ-dependence made explicit; and (iv) linear convergence of a gradient plasticity rule under the Polyak--Łojasiewicz condition. Two synthetic experiments confirm the architectural advantage: a Zipf-distributed retrieval benchmark and an entity label-copy task with uniform lag distribution, the latter ruling out prior-matching as an explanation for the power-law kernel's advantage.

Nabil Mlaiki

Department of Mathematics and Sciences, Prince Sultan University, Riyadh, Saudi Arabia

Memory-Augmented Language ModelsTransformerLong-Context Language ModelingPersistent Memory for Language Models
AskPDF
May 9, 2026

Kaczmarz Linear Attention

Long-context language modeling remains central to modern sequence modeling, but the quadratic cost of Transformer attention makes scaling computationally prohibitive. Linear recurrent models address this bottleneck by compressing the context into a fixed-size state, making the rule that forgets, writes, and edits information a central design problem. To address state maintenance, Gated DeltaNet (GDN) combines gated state decay with delta-rule residual writes, using a learnable coefficient to balance forgetting and update magnitude. However, this coefficient is learned empirically rather than derived from the underlying objective, which can lead to suboptimal update magnitudes. We revisit the online-regression objective underlying GDN and, inspired by the Kaczmarz projection method, derive the key-norm-normalized dynamic step size βt=ηt/(∥kt∥22+ε)β_t = η_t / (\|k_t\|_2^2 + ε) for residual updates. We propose Kaczmarz Linear Attention (KLA), a one-scalar modification of GDN that preserves the state shape, gates, linear recurrence, and chunkwise parallel algorithm. At the 0.4B scale with a 1B-token budget, KLA achieves the lowest validation perplexity among evaluated linear-time baselines, 8.09 versus 8.50 for GDN, and remains stable up to 65K tokens. On controlled tasks, KLA reaches 100% on single-needle-in-a-haystack retrieval, improves 8x multi-query associative recall by 7.03 points over GDN, and delivers 2.1x higher decode throughput at 32K context. These results suggest that the key-norm-normalized Kaczmarz coefficient is a first-order design axis for delta-rule sequence models: it improves accuracy, extrapolation, and decoding efficiency without changing the recurrent state or hardware kernel.

Jiaxuan Zou, Ruifeng Ren, Yong Liu

School of Mathematics and Statistics Xi’an Jiaotong University Xi’an, China · Gaoling School of Artificial Intelligence Renmin University of China Beijing, China

Long-Context Language ModelingRecurrent TransformersLinear AttentionDelta-Rule Linear AttentionGated Delta NetworksLinear RNNs
AskPDF
May 8, 2026

Self-Consolidating Language Models: Continual Knowledge Incorporation from Context

Large language models (LLMs) increasingly receive information as streams of passages, conversations, and long-context workflows. While longer context windows expose more evidence, they do not ensure that useful information is preserved and reused. We study continual context consolidation: writing current context into model weights while limiting interference with previously consolidated information. We propose Self-Consolidating Language Models (SCoL), a post-training framework in which, given current context, an LLM learns to generate textual update instructions specifying which of its own Transformer layers should be updated. Because committed updates change the model that later generates future selections, we train SCoL with meta-reinforcement learning over an evolving model state. We instantiate SCoL with supervised QA rewards on SQuAD knowledge incorporation and intrinsic likelihood-based rewards for LongBench v2 long-context consolidation. Across both settings, SCoL improves acquisition and retention over prompting, summarization, batch test-time training, and sequential finetuning baselines. Analysis of learned selection patterns shows that SCoL encourages the LLM to generate sparse update locations that align with layers of high Fisher information, suggesting that the model learns to route plasticity toward loss-sensitive regions while limiting interference. Moreover, SCoL transfers from shorter meta-training streams to longer LongBench v2 streams at evaluation, suggesting that our framework supports scalable streaming consolidation.

Zekun Wang, Anant Gupta, Zihan Dong +1

Georgia Institute of Technology

Continual Learning for LLMsLong-Context Language ModelingIncremental LearningLLM Post-Training
AskPDF
May 7, 2026

Long Context Pre-Training with Lighthouse Attention

Training causal transformers at extreme sequence lengths is bottlenecked by the quadratic time and memory of scaled dot-product attention (SDPA). In this work, we propose Lighthouse Attention, a training-only symmetrical selection-based hierarchical attention algorithm that wraps around ordinary SDPA and can be easily removed towards the end of the training. Our hierarchical selection is also gradient-free, which exempts us from dealing with a complicated and potentially inefficient backward pass kernel. Our contribution is three-fold: (i) A subquadratic hierarchical pre- and post-processing step that does adaptive compression and decompression of the sequence. (ii) A symmetrical compression strategy that pools queries, keys and values at the same time, while preserving left-to-right causality, which greatly improves parallelism. (iii) A two stage training approach which we pre-train for the majority of the time with Lighthouse Attention and recover a full attention model at the end with a short training. We run preliminary small scale LLM pre-training experiments that show the effectiveness of our method compared to full attention training with all other settings matched, where we achieve a faster total training time and lower final loss after the recovery phase. Full code is available at: https://github.com/ighoshsubho/lighthouse-attention

Bowen Peng, Subho Ghosh, Jeffrey Quesnelle

Nous Research

Attention MechanismsSelf-AttentionLong-Context Language ModelingLanguage Model PretrainingEfficient AttentionCausal Attention+1
AskPDF
May 7, 2026

Matrix-Decoupled Concentration for Autoregressive Sequences: Dimension-Free Guarantees for Sparse Long-Context Rewards

Sequence-level evaluations in autoregressive Large Language Models (LLMs) rely on highly dependent token generation. Establishing tight concentration bounds for these processes remains a challenge due to two fundamental bottlenecks in existing frameworks: (i) classical inequalities typically separate dependency structures from target sensitivities, leading to a scalar collapse that inflates the variance proxy to a suboptimal O(N)\mathcal{O}(N) for sparse terminal rewards; (ii) conversely, while certain spatial methods achieve tighter bounds, they lack the strictly causal filtration required by sequential generation, rendering them inapplicable to the autoregressive setting. To resolve both bottlenecks, we establish a sharp McDiarmid-type inequality for dependent sequences, governed strictly by the exact matrix-vector multiplication of the causal dependency resolvent and the target sensitivity vector. This Matrix-Decoupled Concentration (MDC) framework natively recovers optimal constants for Markov chains and exploits directed dd-separation to yield order-optimal bounds for causal trees. Crucially, by exactly preserving the coordinate-wise sparsity of rewards within a strictly causal framework, MDC mathematically prevents scalar collapse, guaranteeing a dimension-free O(1)\mathcal{O}(1) variance proxy and providing a rigorous mathematical justification for the stability of long-context reasoning.

Pei-Sen Li

School of Mathematics and Statistics, Beijing Institute of Technology, Beijing 100872, China

Long-Context Language ModelingAutoregressive Language Modeling
AskPDF
May 7, 2026

Training Transformers for KV Cache Compressibility

Long-context language modeling is increasingly constrained by the Key-Value (KV) cache, whose memory and decode-time access costs scale linearly with the prefix length. This bottleneck has motivated a range of context-compression methods, from token-level summarization to recent optimization-based KV compression methods. These post-hoc methods operate on the KV cache of a fixed pretrained model, so their effectiveness is fundamentally limited by how well the model's internal representations can be compressed. In this work, we formalize the notion of KV compressibility and show that it is a property of the learned representations, rather than of the context alone. We prove that almost any sequence-to-vector function admits both highly compressible and inherently non-compressible transformer implementations, highlighting the need to guide transformers toward compressible representations during training. Motivated by this, we propose KV-Compression Aware Training (KV-CAT), a continued pretraining procedure that incentivizes the emergence of compressible representations. We introduce a train-time KV sparsification policy that masks KV slots during training. This forces the model to use fewer KV slots and encourages it to learn representations amenable to post-hoc compression. Empirically, we show that KV-CAT improves the quality-budget tradeoff of downstream compression methods across retrieval, long-context question answering, and perplexity-based evaluation of compressed-prefix continuation.

Yoav Gelberg, Yam Eitan, Michael Bronstein +2

University of Oxford · 2Technion – Israel Institute of Technology · 3AITHYRA +1

Long-Context Language ModelingKV-Cache Compression
AskPDF
Apr 30, 2026

Caracal: Causal Architecture via Spectral Mixing

The scalability of Large Language Models to long sequences is hindered by the quadratic cost of attention and the limitations of positional encodings. To address these, we introduce Caracal, a novel architecture that replaces attention with a parameter-efficient, O(L log(L)) Multi-Head Fourier (MHF) module. Our contributions are threefold: (1) We leverage the Fast Fourier Transform (FFT) for sequence mixing, inherently addressing both bottlenecks mentioned above. (2) We apply a frequency-domain causal masking technique that enforces autoregressive capabilities via asymmetric padding and truncation, overcoming a critical barrier for Fourier-based generative models. (3) Unlike efficient models relying on hardware-specific implementations (e.g., Mamba), we uses standard library operators. This ensures robust portability, eliminating common deployment barriers. Evaluations demonstrate that Caracal performs competitively with Transformer and SSM baselines, offering a scalable and simple pathway for efficient long-sequence modeling. Code is available in Appendix.

Bingzheng Gan, Tianyi Zhang, Yusu Li +4

Huawei Technologies Co., Ltd., China.

Long-Context Language ModelingAutoregressive Language ModelingLLM Inference Acceleration
AskPDF
Apr 27, 2026

Long-Context Aware Upcycling: A New Frontier for Hybrid LLM Scaling

Hybrid sequence models that combine efficient Transformer components with linear sequence modeling blocks are a promising alternative to pure Transformers, but most are still pretrained from scratch and therefore fail to reuse existing Transformer checkpoints. We study upcycling as a practical path to convert pretrained Transformer LLMs into hybrid architectures while preserving short-context quality and improving long-context capability. We call our solution HyLo (HYbrid LOng-context): a long-context upcycling recipe that combines architectural adaptation with efficient Transformer blocks, Multi-Head Latent Attention (MLA), and linear blocks (Mamba2 or Gated DeltaNet), together with staged long-context training and teacher-guided distillation for stable optimization. HyLo extends usable context length by up to 32×32\times through efficient post-training and reduces KV-cache memory by more than 90%90\%, enabling up to 2M-token prefill and decoding in our vLLM inference stack, while comparable Llama baselines run out of memory beyond 64K context. Across 1B- and 3B-scale settings (Llama- and Qwen-based variants), HyLo delivers consistently strong short- and long-context performance and significantly outperforms state-of-the-art upcycled hybrid baselines on long-context evaluations such as RULER. Notably, at similar scale, HyLo-Qwen-1.7B trained on only 10B tokens significantly outperforms JetNemotron (trained on 400B tokens) on GSM8K, Lm-Harness common sense reasoning and RULER-64K.

Parsa Ashrafi Fashi, Utkarsh Saxena, Mehdi Rezagholizadeh +7

1AMD

Long-Context Language Model InferenceLong-Context Language ModelingSequence Modeling
AskPDF
Apr 27, 2026

Kwai Summary Attention Technical Report

Long-context ability, has become one of the most important iteration direction of next-generation Large Language Models, particularly in semantic understanding/reasoning, code agentic intelligence and recommendation system. However, the standard softmax attention exhibits quadratic time complexity with respect to sequence length. As the sequence length increases, this incurs substantial overhead in long-context settings, leading the training and inference costs of extremely long sequences deteriorate rapidly. Existing solutions mitigate this issue through two technique routings: i) Reducing the KV cache per layer, such as from the head-level compression GQA, and the embedding dimension-level compression MLA, but the KV cache remains linearly dependent on the sequence length at a 1:1 ratio. ii) Interleaving with KV Cache friendly architecture, such as local attention SWA, linear kernel GDN, but often involve trade-offs among KV Cache and long-context modeling effectiveness. Besides the two technique routings, we argue that there exists an intermediate path not well explored: {Maintaining a linear relationship between the KV cache and sequence length, but performing semantic-level compression through a specific ratio kk}. This O(n/k)O(n/k) path does not pursue a ``minimum KV cache'', but rather trades acceptable memory costs for complete, referential, and interpretable retention of long distant dependency. Motivated by this, we propose Kwai Summary Attention (KSA), a novel attention mechanism that reduces sequence modeling cost by compressing historical contexts into learnable summary tokens.

Chenglong Chu, Guorui Zhou, Guowang Zhang +35

Kuaishou

Self-AttentionKV CachingLong-Context Language Model InferenceLong-Context Language ModelingTransformer AttentionAttention Mechanisms+1
AskPDF
Apr 26, 2026

DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence

We present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models -- DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) -- both supporting a context length of one million tokens. DeepSeek-V4 series incorporate several key upgrades in architecture and optimization: (1) a hybrid attention architecture that combines Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to improve long-context efficiency; (2) Manifold-Constrained Hyper-Connections (mHC) that enhance conventional residual connections; (3) and the Muon optimizer for faster convergence and greater training stability. We pre-train both models on more than 32T diverse and high-quality tokens, followed by a comprehensive post-training pipeline that unlocks and further enhances their capabilities. DeepSeek-V4-Pro-Max, the maximum reasoning effort mode of DeepSeek-V4-Pro, redefines the state-of-the-art for open models, outperforming its predecessors in core tasks. Meanwhile, DeepSeek-V4 series are highly efficient in long-context scenarios. In the one-million-token context setting, DeepSeek-V4-Pro requires only 27% of single-token inference FLOPs and 10% of KV cache compared with DeepSeek-V3.2. This enables us to routinely support one-million-token contexts, thereby making long-horizon tasks and further test-time scaling more feasible. The model checkpoints are available at https://huggingface.co/collections/deepseek-ai/deepseek-v4.

DeepSeek-AI, Anyi Xu, Bangcai Lin +316

Mixture-of-Experts Language ModelsLong-Context Language Model InferenceLong-Context Language ModelingLLM Inference AccelerationSparse AttentionAttention Mechanisms+1
AskPDF
Apr 22, 2026

Preconditioned DeltaNet: Curvature-aware Sequence Modeling for Linear Recurrences

To address the increasing long-context compute limitations of softmax attention, several subquadratic recurrent operators have been developed. This work includes models such as Mamba-2, DeltaNet, Gated DeltaNet (GDN), and Kimi Delta Attention (KDA). As the space of recurrences grows, a parallel line of work has arisen to taxonomize them. One compelling view is the test-time regression (TTR) framework, which interprets recurrences as performing online least squares updates that learn a linear map from the keys to values. Existing delta-rule recurrences can be seen as first-order approximations to this objective, but notably ignore the curvature of the least-squares loss during optimization. In this work, we address this by introducing preconditioning to these recurrences. Starting from the theory of online least squares, we derive equivalences between linear attention and the delta rule in the exactly preconditioned case. Next, we realize this theory in practice by proposing a diagonal approximation: this enables us to introduce preconditioned variants of DeltaNet, GDN, and KDA alongside efficient chunkwise parallel algorithms for computing them. Empirically, we find that our preconditioned delta-rule recurrences yield consistent performance improvements across synthetic recall benchmarks and language modeling at the 340M and 1B scale.

Neehal Tumma, Noel Loo, Daniela Rus

1MIT · 2Liquid AI

Long-Context Language ModelingRecurrent Neural NetworksRecurrent TransformersLinear AttentionDelta-Rule Linear AttentionLinear RNNs
AskPDF
Apr 20, 2026

Sessa: Selective State Space Attention

Modern sequence modeling is dominated by two families: Transformers, whose self-attention can access arbitrary elements of the visible sequence, and structured state-space models, which propagate information through an explicit recurrent state. These mechanisms face different limitations on long contexts: when attention is diffuse, the influence of individual tokens is diluted across the effective support, while recurrent state propagation can lose long-range sensitivity unless information is actively preserved. As a result, both mechanisms face challenges in preserving and selectively retrieving information over long contexts. We propose Sessa, a decoder that places attention inside a recurrent feedback path. This creates many attention-based paths through which past tokens can influence future states, rather than relying on a single attention read or a single recurrent chain. We prove that, under explicit assumptions and matched regimes, Sessa admits power-law memory tails O(ℓ−β)O(\ell^{-β}) for 0<β<10 < β< 1, with slower decay than in the corresponding Transformer and Mamba-style baselines. We further give an explicit construction that achieves this power-law rate. Under the same assumptions, Sessa is the only model class among those considered that realizes flexible selective retrieval, including profiles whose influence does not decay with distance. Consistent with this theoretical advantage, across matched experiments, Sessa achieves the strongest performance on long-context benchmarks while remaining competitive with Transformer and Mamba-style baselines on short-context language modeling.

Liubomyr Horbatko

Attention MechanismsSelf-AttentionLong-Context Language ModelingPersistent Memory for Language ModelsState Space ModelsRecurrent Transformers+3
AskPDF
Apr 19, 2026

OPSDL: On-Policy Self-Distillation for Long-Context Language Models

Extending the effective context length of large language models (LLMs) remains a central challenge for real-world applications. While recent post-training methods have made progress in long-context scaling, they either rely on high-quality supervision data or sparse sequence-level rewards, leading to unstable and inefficient optimization. We propose OPSDL, an On-Policy Self-Distillation method for enhancing the Long-context capabilities of LLMs. Unlike other recent self-distillation methods that inject privileged information and rely on the model's in-context learning ability to act as a teacher, OPSDL leverages the model's own inherently strong short-context capability as a self-teacher to supervise its own generation in long-context scenarios. The model first generates responses conditioned on the full long-context, then the self-teacher provides per-token supervision signals via point-wise reverse KL divergence under the relevant extracted short-context. This dense token-level signal encourages faithful use of relevant evidence and mitigates hallucinations induced by irrelevant context. We evaluate OPSDL on long-context benchmarks across a range of models from 7B to 32B parameters. Results show consistent and substantial improvements across varying context lengths, outperforming standard post-training approaches such as SFT and DPO with higher sample efficiency. Notably, these gains are achieved without degrading general short-context performance. These findings highlight the effectiveness of OPSDL as a scalable and stable approach for long-context learning.

Xinsen Zhang, Zhenkai Ding, Tianjun Pan +4

Baidu Inc.

Long-Context Language ModelingOn-Policy Self-DistillationLanguage Model DistillationLanguage Model Post-Training
AskPDF
Apr 15, 2026

Shuffle the Context: RoPE-Perturbed Self-Distillation for Long-Context Adaptation

Large language models (LLMs) increasingly operate in settings that require reliable long-context understanding, such as retrieval-augmented generation and multi-document reasoning. A common strategy is to fine-tune pretrained short-context models at the target sequence length. However, we find that standard long-context adaptation can remain brittle: model accuracy depends strongly on the absolute placement of relevant evidence, exhibiting high positional variance even when controlling for task format and difficulty. We propose RoPE-Perturbed Self-Distillation, a training regularizer that improves positional robustness. The core idea is to form alternative "views" of the same training sequence by perturbing its RoPE indices -- effectively moving parts of the context to different positions -- and to train the model to produce consistent predictions across views via self-distillation. This encourages reliance on semantic signals instead of brittle position dependencies. Experiments on long-context adaptation of Llama-3-8B and Qwen-3-4B demonstrate consistent gains on long-context benchmarks, including up to 12.04% improvement on RULER-64K for Llama-3-8B and 2.71% on RULER-256K for Qwen-3-4B after SFT, alongside improved length extrapolation beyond the training context window.

Zichong Li, Chen Liang, Liliang Ren +3

Work is done during internship at Microsoft. · Georgia Institute of Technology · Microsoft.

Long-Context Language ModelingLanguage Model RobustnessSelf-Distillation
AskPDF
Apr 15, 2026

SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention

While sparse attention mitigates the computational bottleneck of long-context LLM training, its distributed training process exhibits extreme heterogeneity in both 1) sequence length and 2) sparsity sensitivity, leading to a severe imbalance problem and sub-optimal model accuracy. Existing algorithms and training frameworks typically focus on single issue, failing to systematically co-optimize these two problems. Therefore, we propose SparseBalance, a novel algorithm-system co-design framework, which exploits the sparsity and sequence heterogeneity to optimize model accuracy and system efficiency jointly. First, we propose workload-aware dynamic sparsity tuning, which employs a bidirectional sparsity adjustment to eliminate stragglers and exploit inherent bubbles for free accuracy. Second, we propose a sparsity-aware batching strategy to achieve coarse-grained balance, which complements dynamic sparsity tuning. Experimental results demonstrate that SparseBalance achieves up to a 1.33×\times end-to-end speedup while still improving the long-context capability by 0.46% on the LongBench benchmark.

Hongtao Xu, Jianchao Tan, Yuxuan Hu +8

School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences, Beijing, China · Meituan, Beijing, China · University of Chinese Academy of Sciences, Beijing, China

Sparse AttentionDistributed TrainingLong-Context Language Modeling
AskPDF
Apr 14, 2026

Latent-Condensed Transformer for Efficient Long Context Modeling

Large language models (LLMs) face significant challenges in processing long contexts due to the linear growth of the key-value (KV) cache and quadratic complexity of self-attention. Existing approaches address these bottlenecks separately: Multi-head Latent Attention (MLA) reduces the KV cache by projecting tokens into a low-dimensional latent space, while sparse attention reduces computation. However, sparse methods cannot operate natively on MLA's compressed latent structure, missing opportunities for joint optimization. In this paper, we propose Latent-Condensed Attention (LCA), which directly condenses context within MLA's latent space, where the representation is disentangled into semantic latent vectors and positional keys. LCA separately aggregates semantic vectors via query-aware pooling and preserves positional keys via anchor selection. This approach jointly reduces both computational cost and KV cache without adding parameters. Beyond MLA, LCA's design is architecture-agnostic and readily extends to other attention mechanisms such as GQA. Theoretically, we prove a length-independent error bound. Experiments show LCA achieves up to 2.5×\times prefilling speedup and 90% KV cache reduction at 128K context while maintaining competitive performance.

Zeng You, Yaofo Chen, Qiuwu Chen +5

South China University of Technology · Pengcheng Laboratory · AIGCode +2

Long-Context Language ModelingLLM Inference AccelerationKV-Cache CompressionMulti-Head Latent Attention
AskPDF