Long-Context Language Modeling

Follow this topic Ask about this topic

Momentum

12 papers in the last four weeks, up 71% on the four weeks before. 0.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 112

May 7, 2026cs.CL

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

Language Model Pretraining Self-Attention Long-Context Language Modeling Efficient Attention Attention Mechanisms

May 7, 2026cs.LG

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

\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

d

-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

\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 Modeling Autoregressive Language Modeling

May 7, 2026cs.LG

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 Modeling KV-Cache Compression

Apr 30, 2026cs.LG

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 Modeling Autoregressive Language Modeling LLM Inference Acceleration

Apr 27, 2026cs.CL

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 \emph{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\times

through efficient post-training and reduces KV-cache memory by more than

90\%

, enabling up to 2M-token prefill and decoding in our \texttt{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 Inference Long-Context Language Modeling Sequence Modeling

Apr 27, 2026cs.CL

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

k

}. This

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-Attention KV Caching Long-Context Language Model Inference Long-Context Language Modeling Transformer Attention Attention Mechanisms+1

Apr 26, 2026cs.CL

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 Models Long-Context Language Model Inference Long-Context Language Modeling LLM Inference Acceleration Sparse Attention Attention Mechanisms+1

Apr 22, 2026cs.LG

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 Modeling Recurrent Neural Networks Recurrent Transformers Linear Attention Delta-Rule Linear Attention Linear RNNs

Apr 20, 2026cs.LG

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(\ell^{-β})

for

0 < 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

Self-Attention Long-Context Language Modeling Persistent Memory for Language Models Recurrent Transformers Attention Mechanisms State Space Models

Apr 19, 2026cs.CL

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 Modeling On-Policy Self-Distillation Language Model Distillation Language Model Post-Training

Apr 1, 2026cs.LG

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

1

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-Attention Long-Context Language Modeling Transformer Attention LLM Inference Acceleration Attention Mechanisms

Mar 18, 2026cs.CL

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

\sim

80% of tokens, outperforming prior baselines. We also design custom Triton kernels to efficiently implement this token-wise conditional Attention on GPUs, achieving up to

\sim

2

\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-Attention Long-Context Language Modeling LLM Inference Acceleration Sparse Attention Attention Mechanisms

Feb 12, 2026cs.LG

RAM-Net: Linear-Time Sequence Modeling with Sparsely Addressable State

Linear attention offers an efficient alternative to full attention with a fixed-size recurrent state. However, this state is shared by all tokens, so information from distinct tokens becomes superposed within it and produces inter-token interference that degrades long-range fine-grained recall. To address this issue, we propose RAM-Net, which replaces dense access to a shared state with sparse address-based access. RAM-Net organizes the recurrent state as a fixed-size array of independent slots and uses an Address Decoder that maps each key or query into a sparse address, selecting a small subset of slots to write to or read from at each step. This design directs tokens with non-overlapping addresses to disjoint slots, suppressing inter-token interference, while keeping per-step state access dependent only on the number of selected slots rather than the total state size. Empirically, RAM-Net outperforms strong recurrent baselines on fine-grained long-range retrieval and achieves the lowest perplexity with competitive commonsense reasoning. It does so while accessing fewer state elements per step than all baselines, e.g.,

8\times

fewer than Mamba2.

Kaicheng Xiao, Haotian Li, Liran Dong +1

The Chinese University of Hong Kong

Memory-Augmented Neural Networks Long-Context Language Modeling Linear Attention Associative Memory

Jan 17, 2026cs.LG

Threshold Differential Attention for Sink-Free, Ultra-Sparse, and Non-Dispersive Language Modeling

Softmax attention struggles with long contexts due to structural limitations: the strict sum-to-one constraint forces attention sinks on irrelevant tokens, and probability mass disperses as sequence lengths increase. We tackle these problems with Threshold Differential Attention (TDA), a sink-free attention mechanism that achieves ultra-sparsity and improved robustness at longer sequence lengths without the computational overhead of projection methods or the performance degradation caused by noise accumulation of standard rectified attention. TDA applies row-wise extreme-value thresholding with a length-dependent gate, retaining only exceedances. Inspired by the differential transformer, TDA also subtracts an inhibitory view to enhance expressivity. Theoretically, we prove that TDA controls the expected number of spurious survivors per row to

O(1)

and that consensus spurious matches across independent views vanish as context grows. Empirically, TDA produces

>99\%

exact zeros and eliminates attention sinks while maintaining competitive performance on standard and long-context benchmarks.

Xingyue Huang, Xueying Ding, Mingxuan Ju +3

University of Oxford · Carnegie Mellon University · Snap Inc.

Self-Attention Long-Context Language Modeling Sparse Attention Attention Sinks Attention Mechanisms

Dec 28, 2025cs.CL

TabiBERT: A Large-Scale ModernBERT Foundation Model and A Unified Benchmark for Turkish

The introduction of BERT established encoder-only transformer models as a foundational paradigm in natural language processing. Encoder-only models remain the standard tool for classification, tagging and retrieval, where contextual representations and low inference cost matter more than text generation, yet Turkish lacks a monolingual encoder trained from scratch with the advances consolidated in ModernBERT (rotary positional embeddings, FlashAttention, refined normalization). We introduce TabiBERT, a monolingual Turkish encoder based on the ModernBERT architecture, pretrained from scratch for one trillion tokens sampled from an 86.58B-token multi-domain corpus of web text (72%), scientific publications (19%), source code (6%) and mathematical content (0.3%). The model supports a context length of 8,192 tokens, sixteen times that of existing Turkish BERT models, and inherits the ModernBERT architecture's efficiency at long context. For rigorous and reproducible evaluation we introduce TabiBench, a benchmark of 27 datasets across eight task categories with standardized splits and evaluation protocols, summarized as a GLUE-style macro-average on a 0-100 scale. TabiBERT leads the Turkish models in five of eight categories and BERTurk, the previous best, in six of eight; the gains concentrate on question answering (+9.55 F1) and code retrieval (+2.41 NDCG@10), while the four short-text categories are near saturation. Its average of 77.28 exceeds BERTurk's 75.66; the multilingual mmBERT reaches 78.98 with twice the parameters and three times the training tokens, at 41% more tokens per Turkish input. We release model weights, training configurations and evaluation code as a transparent and reproducible foundation for future Turkish encoder research.

Melikşah Türker, A. Ebrar Kızıloğlu, Onur Güngör +1

Department of Computer Engineering, Boğaziçi University, Türkiye · VNGRS, Istanbul, Türkiye · Technical University of Munich (TUM), Munich, Germany

Language Model Pretraining Long-Context Language Modeling Language Modeling

Nov 7, 2025cs.LG

Controllably Efficient Language Models

The substantial inference costs of attention in transformers motivated the development of efficient sequence mixers: namely sparse and sliding window attention, convolutions and linear attention. Although these approaches result in impressive reductions in inference costs, they often trade-off with quality, specifically in-context recall. Apriori fixing this quality-cost tradeoff at training time means being suboptimal from the get-go: some downstream applications might fundamentally require more memory for in-context recall, while other tasks may require lower latency and memory. We propose a conceptually simple meta-sequence mixer with inference-cost controllability: the Compress & Attend Transformer (CAT). CAT decodes chunks of tokens by attending to compressed chunks of the sequence so far. Both compression and decoding can use any existing sequence mixer. Decoding from the compressed sequence yields compute and memory savings, with chunk size setting the operating point on the quality-cost trade-off. Importantly, training CAT across multiple chunk sizes at once unlocks test-time control of this trade-off without any retraining, all in a single model. Instantiated with the most basic choice, dense attention as the mixer, CAT surprisingly suffices to match 10 popular and diverse efficient models (linear, hybrids, sparse) on real-world long-context recall at comparable inference costs, all from a single trained model. CAT further performs competitively on long-context understanding benchmarks while providing 1.4-3.7x higher generation throughput than a dense transformer. Code is at: https://github.com/rajesh-lab/cat-transformer

Jatin Prakash, Aahlad Puli, Rajesh Ranganath

New York University

Efficient Transformer Inference Long-Context Language Modeling Transformer Attention Attention Mechanisms

Sep 25, 2025cs.DC

Concertina: Data-Centric Adaptive Pipeline Parallelism for Efficient Heterogeneous Long-Context LLM Training

Long context training is crucial for extending LLM context windows. Existing schemes, such as sequence parallelism, incur substantial communication overhead. Pipeline parallelism (PP) reduces this cost, but its effectiveness hinges on partitioning granularity. Batch-level PP employing sequence packing exhibits high memory consumption in long-context scenarios, whereas token-level PP splitting sequences into slices alleviates memory overhead but may introduce performance degradation. Moreover, the skewed sequence-length distribution in real-world datasets defeats any monolithic, static choice of PP granularity. In this paper, we propose \textit{Dynamic Pipeline Parallelism} (DPP), which transforms PP granularity from a static design choice into a workload-adaptive optimization space over packed, split, and hybrid chunks. DPP further introduces a new coupling between heterogeneous pipeline scheduling and gradient checkpointing. To solve this coupling, \name co-optimizes dynamic chunk scheduling with \textit{Stage-Aware Chunk-Level Adaptive Checkpointing}. Comprehensive experiments demonstrate that \name achieves up to 1.69\texttimes\ speedup over FlexSP and up to 1.40\texttimes\ over MEPipe. The source code is available at https://github.com/wsjdsg/InfiniPipe-code.

Shiju Wang, Yujie Wang, Fangcheng Fu +6

Tsinghua University · Peking University · Shanghai Jiao Tong University +1

Long-Context Language Modeling Efficient Language Model Training Pipeline Parallelism

Sep 12, 2025cs.LG

Multipole Semantic Attention: A Fast Approximation of Softmax Attention for Pretraining

Pretraining transformers on long sequences (entire code repositories, collections of related documents) is bottlenecked by quadratic attention costs. We present Multipole Semantic Attention (MuSe), which accelerates 64k-context pretraining by 36% while matching baseline loss, requiring no architectural changes. MuSe clusters queries and keys separately in representation space. This yields query-specific summaries that substantially outperform spatial blocking at matched sparsity, while also enabling drop-in compatibility with existing pretrained models; we validate on Llama 3.1-8B and 3.2-1B without retraining. We pretrain language models up to 1B parameters at 64k context on code and scientific documents, confirming that MuSe preserves quality and long-context utilization during training.

Rupert Mitchell, Kristian Kersting

Department of Computer Science, TU Darmstadt, Darmstadt, Germany · Hessian Center for AI (hessian.AI), Darmstadt, Germany · German Research Center for Artificial Intelligence (DFKI), Darmstadt, Germany +1

Language Model Pretraining Softmax Attention Self-Attention Long-Context Language Modeling Transformer Attention Sparse Attention+1

Sep 4, 2025cs.LG

Robust Filter Attention: Self-Attention as Precision-Weighted State Estimation

We introduce Robust Filter Attention (RFA), a formulation of self-attention as a robust state estimator. Each token is treated as a noisy observation of a latent trajectory governed by a linear stochastic differential equation (SDE), and attention weights are determined by consistency under this model rather than static feature similarity. Under isotropic noise and decay assumptions, RFA matches the computational complexity of standard attention. On language modeling benchmarks, RFA achieves lower perplexity than RoPE within the training window while remaining stable under zero-shot extrapolation to longer contexts. The framework also provides a dynamical interpretation of standard positional mechanisms, connecting rotational embeddings and recency biases to transport and uncertainty propagation induced by stochastic dynamics.

Peter Racioppo

Independent Researcher, Los Angeles, CA, USA.

Self-Attention Long-Context Language Modeling Autoregressive Language Modeling Transformer Attention Attention Mechanisms

Apr 24, 2025cs.CL

The Sparse Frontier: Sparse Attention Trade-offs in Transformer LLMs

Sparse attention offers a promising strategy to extend long-context capabilities in Transformer LLMs, yet its efficiency-accuracy trade-offs remain unclear due to the lack of comprehensive evaluation. We address this gap with the largest-scale empirical analysis to date of training-free sparse attention, evaluating six methods across multiple model families and sizes, sequences up to 128K tokens, and sparsity levels up to 0.95 (i.e.,

1/20

attention budget) on nine diverse tasks. We first organise the rapidly evolving landscape of sparse attention methods into a taxonomy along four design axes. Our analysis then yields actionable insights: 1) sparse attention is effective: larger sparse models outperform smaller dense ones at equivalent cost, improving the Pareto frontier; 2) for the training-free methods we study, fine-grained per-query importance estimation during prefilling remains impractical-due to both the cost of estimation and the lack of sparse kernels that translate fine-grained sparsity into wall-clock gains-forcing a task-dependent choice between global-to-token and block-to-block selection. Instead, during decoding, token-to-page selection becomes feasible, enabling better generalisation and higher sparsity tolerance; 3) longer sequences tolerate higher sparsity, suggesting that fixed-budget methods in production are suboptimal. Together, these findings provide practical guidance for deploying sparse attention and methodological recommendations for future evaluations. Our code is available at https://github.com/PiotrNawrot/sparse-frontier.

Piotr Nawrot, Robert Li, Renjie Huang +3

University of Edinburgh · Cohere · Meta

Transformer Inference LLM Evaluation Self-Attention Long-Context Language Modeling LLM Inference Acceleration Sparse Attention+1

Feb 20, 2025cs.CL

LLM-Microscope: Uncovering the Hidden Role of Punctuation in Context Memory of Transformers

We introduce methods to quantify how Large Language Models (LLMs) encode and store contextual information, revealing that tokens often seen as minor (e.g., determiners, punctuation) carry surprisingly high context. Notably, removing these tokens -- especially stopwords, articles, and commas -- consistently degrades performance on MMLU and BABILong-4k, even if removing only irrelevant tokens. Our analysis also shows a strong correlation between contextualization and linearity, where linearity measures how closely the transformation from one layer's embeddings to the next can be approximated by a single linear mapping. These findings underscore the hidden importance of filler tokens in maintaining context. For further exploration, we present LLM-Microscope, an open-source toolkit that assesses token-level nonlinearity, evaluates contextual memory, visualizes intermediate layer contributions (via an adapted Logit Lens), and measures the intrinsic dimensionality of representations. This toolkit illuminates how seemingly trivial tokens can be critical for long-range understanding.

Anton Razzhigaev, Matvey Mikhalchuk, Temurbek Rahmatullaev +4

AIRI · Skoltech · Lomonosov Moscow State University +1

Transformer Interpretability Long-Context Language Modeling LLM Interpretability Language Modeling

Date pendingcs.AI

AdaRoPE: Not All Attention Heads Should Rotate and Scale Equally

Rotary Position Embedding (RoPE) is widely adopted in Transformers to encode positional information, yet standard implementations enforce a uniform frequency schedule and scaling across all attention heads. Using simplified retrieval tasks and length generalization scenarios, we show -- both empirically and theoretically -- that heads with different functional roles require distinct frequency ranges and attention scaling factors to operate effectively. Ignoring this structure leads to suboptimal utilization of embedding dimensions and degraded performance, particularly under long-context settings. To address these limitations, we propose AdaRoPE, which equips each attention head with learnable rotation frequencies and attention scaling factors. Pretrained LLMs with AdaRoPE consistently outperform existing RoPE variants, including partial RoPE and NoPE baselines. For context extension, we further show that uniform frequency and attention scaling, used in methods such as YaRN, are suboptimal. By applying head-specific scaling, AdaRoPE enables better context extension while better preserving short-context performance in both the extrapolation setting and the long-context continued pretraining setting. These results highlight the importance of optimizing rotary position embedding at the level of individual attention heads.

Shaowen Wang, Yuke Zheng, Tansheng Zhu +4

Tsinghua University · Hunyuan Team, Tencent · Xiongan AI Institute +1

Rotary Positional Embeddings Self-Attention Long-Context Language Modeling Attention Mechanisms Attention Head Specialization