Datasets:

fineset-io
/

efficient-llm-papers

id string	sources list	title string	abstract string	authors list	categories list	fields_of_study list	published_date timestamp[s]	url string	pdf_url string	arxiv_id string	doi string	citation_count int64	has_code bool	code_url string	venue string	quality_score float64
219a22b230259ea97336b8090150dacc77948d9a35534711418df43c754d22f9	[ "arxiv" ]	Can I Buy Your KV Cache?	Right now, across the world, AI agents are repeating the same absurd act: to read one document, they each recompute it from scratch. Every agent re-runs prefill, the most compute-intensive step a large model takes, over identical text, only to rebuild a key-value (KV) cache identical to the one the agent before it just...	[ "Luoyuan Zhang" ]	[ "cs.AI", "cs.CE", "cs.MA" ]	[]	2026-06-11T00:00:00	https://arxiv.org/abs/2606.13361	https://arxiv.org/pdf/2606.13361v1	2606.13361	null	0	false	null	null	0.35
363441bf3b6e700abd0cf2940be987393677c46caa1cecd4c2121e14de11dfd9	[ "arxiv" ]	MiniPIC: Flexible Position-Independent Caching in <100LOC	Retrieval-augmented and agentic workloads repeatedly prefill recurring predictable structured inputs (which we call "spans") such as documents and code files. Yet, prefix caching in engines such as vLLM cannot reuse their KV entries unless they share identical prefixes with another request, while Position-Independent C...	[ "Nathan Ordonez", "Thomas Parnell" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[]	2026-06-11T00:00:00	https://arxiv.org/abs/2606.13126	https://arxiv.org/pdf/2606.13126v1	2606.13126	null	0	false	null	null	0.35
4fdecfa82b3d9e9bb030d3cf4a75de1007d4193de7d873f3dd881cdfb348a3eb	[ "arxiv", "semantic_scholar" ]	Multi-Rate Mixture of Experts for Accelerating Liquid Neural Network Training	Multivariate time-series data often exhibit complex temporal dependencies, irregular sampling, and heterogeneous dynamics across multiple time scales, making accurate sequence modeling particularly challenging. Traditional recurrent neural networks (RNNs), such as Long Short-Term Memory (LSTM) networks, operate in disc...	[ "Shilong Zong", "Almuatazbellah Boker", "Hoda Eldardiry" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-06-10T00:00:00	https://arxiv.org/abs/2606.12240	https://arxiv.org/pdf/2606.12240v1	2606.12240	null	0	false	null	null	0.35
e6285ce1e5e00dc6f200445d06208fa2cce60e6d585f8a2a9bfc8b9cdcd4087a	[ "arxiv", "semantic_scholar" ]	Redesign Mixture-of-Experts Routers with Manifold Power Iteration	Router is the cornerstone component to the Mixture-of-Experts models. Serving as expert proxies, the rows of the router matrix compute their similarity to the MoE inputs to determine which subset of experts is activated. Ideally, each router row is designed to encode the expert matrix into this representative vector, s...	[ "Songhao Wu", "Ang Lv", "Ruobing Xie", "Yankai Lin" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-06-10T00:00:00	https://arxiv.org/abs/2606.12397	https://arxiv.org/pdf/2606.12397v1	2606.12397	null	0	false	null	null	0.35
8f96f13b132484676858fd520f097975b3e493dbfb7ba7fbcb3656499f01dddc	[ "arxiv", "semantic_scholar" ]	From Observation to Intervention: A Causal Audit of Expert Importance in Mixture-of-Experts Models	Interpretability methods routinely use population-level summary statistics over observed model behaviour to license claims about the effects of targeted interventions on specific computations; in Pearl's terms, they treat rung-1 associational evidence as if it supported rung-2 interventional conclusions, a move whose v...	[ "Leonard Engmann", "Christian Medeiros Adriano", "Holger Giese" ]	[ "cs.LG", "cs.CL" ]	[ "Computer Science" ]	2026-06-09T00:00:00	https://arxiv.org/abs/2606.10703	https://arxiv.org/pdf/2606.10703v1	2606.10703	null	0	false	null	null	0.35
aeea47930489c3c29a15b5d032ebbd01137ca01018d6230a4d3ceabe6832ac05	[ "arxiv", "semantic_scholar" ]	Routing-Aware Expert Calibration for Machine Unlearning in Mixture-of-Experts Language Models	Machine unlearning is increasingly important for large language models, yet unlearning in Mixture-of-Experts (MoE) architectures remains underexplored. Unlike dense models, MoE architectures employ a router at each layer to assign each token to a sparse subset of experts. In this work, we observe that forget data often...	[ "Jingyi Xie", "Yijun Lin", "Yinjiang Xiong", "Zhikun Zhang", "Sai Li" ]	[ "cs.CL", "cs.AI" ]	[ "Computer Science" ]	2026-06-09T00:00:00	https://arxiv.org/abs/2606.10338	https://arxiv.org/pdf/2606.10338v1	2606.10338	null	0	false	null	null	0.35
4f7b301109d74f7f9d2385065d1af1c3a91b9ebb8cdccba0e0ccbdb224c63788	[ "arxiv", "semantic_scholar" ]	Enhancing Multilingual LLM-based ASR with Mixture of Experts and Dynamic Downsampling	The rapid progress of large language models (LLMs) has opened up a new frontier for automatic speech recognition (ASR), making their effective integration a critical and challenging research direction. To this end, this work proposes a projector-based LLM-ASR framework targeting the key challenges of multilingual gener...	[ "Guodong Lin", "Ziqi Chen", "Yuxiang Fu", "Ke Li", "Wei-Qiang Zhang" ]	[ "cs.SD", "cs.CL", "eess.AS" ]	[ "Computer Science", "Engineering" ]	2026-06-09T00:00:00	https://arxiv.org/abs/2606.10439	https://arxiv.org/pdf/2606.10439v1	2606.10439	10.1109/ICASSP55912.2026.11464266	0	false	null	IEEE International Conference on Acoustics, Speech, and Signal Processing	0.55
19f1030636588c82241484d9b6f7f0589343f2f874642fb66319b91465fda801	[ "arxiv", "semantic_scholar" ]	Which LoRA? An Empirical Study on the Effectiveness of LoRA Techniques During Multilingual Instruction Tuning	We investigate whether commonly available LoRA variants have an advantage over basic LoRA in multilingual instruction tuning. Experiments involving LoRA and four other variants on two datasets across diverse target languages show that there is no significant advantage in using more complex LoRA variants instead of basi...	[ "Thamali Wijewardhana", "Napoleon H. Reyes", "Surangika Ranathunga" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-06-09T00:00:00	https://arxiv.org/abs/2606.10428	https://arxiv.org/pdf/2606.10428v1	2606.10428	null	0	false	null	null	0.35
ad2764c1694461754b9d3f9bf0d1bd04a98208f3430237100c6a3193876a01d6	[ "arxiv", "semantic_scholar" ]	ReasonAlloc: Hierarchical Decoding-Time KV Cache Budget Allocation for Reasoning Models	Long chain-of-thought (CoT) trajectories in large language model (LLM) reasoning cause severe inference bottlenecks due to rapid key-value (KV) cache growth. Current decoding-time compression methods mitigate this issue via token eviction, but typically assume a uniform budget distribution across all layers and heads. ...	[ "Wenhao Liu", "Hao Shi", "Yunhe Li", "Weizhi Fei", "Xiangyuan Wang", "Mengzhe Ruan", "Hanxu Hou", "Peisong Wang", "Linqi Song", "Shuang Qiu" ]	[ "cs.AI" ]	[ "Computer Science" ]	2026-06-09T00:00:00	https://arxiv.org/abs/2606.11164	https://arxiv.org/pdf/2606.11164v1	2606.11164	null	0	false	null	null	0.35
63958be0f7d79f55b95e7ad84e19c28b0f011c5e5a7651835a5caed433a9ff80	[ "arxiv", "semantic_scholar" ]	Hasse Diagrams for Attention: A Partial Order Framework for Designing Transformer Masks	During the training of large Transformer models, attention masks regulate the scope and direction of information flow across a sequence. Numerous mask variants exist, and operators such as FlexAttention already support arbitrary attention masks. Nevertheless, a systematic formal analysis of the information-flow structu...	[ "Chentao Li", "Han Guo" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-06-08T00:00:00	https://arxiv.org/abs/2606.09951	https://arxiv.org/pdf/2606.09951v1	2606.09951	null	0	false	null	null	0.35
b67d706fa49ff7442ed69f855cce85b8f7a8f6a137daecd6262db5419ec6a586	[ "arxiv", "semantic_scholar" ]	FAME: Forecastability-Aware Mixture of Experts for Heterogeneous Time Series Forecasting	Large-scale retail and industrial forecasting systems contain many heterogeneous time series whose lifecycle, sparsity, volatility, seasonality, spectral patterns, and contextual sensitivity differ substantially. A single forecasting model rarely performs well across all regimes, while dense ensembles increase inferenc...	[ "Qianyang Li", "Xingjun Zhang", "Shaoxun Wang", "Tao Peng", "Jia Wei" ]	[ "cs.AI" ]	[ "Computer Science" ]	2026-06-08T00:00:00	https://arxiv.org/abs/2606.08896	https://arxiv.org/pdf/2606.08896v1	2606.08896	null	0	true	https://github.com/hit636/FAME	null	0.65
5b8b06a740c2f8125a9b833ea12dd15aaf50baff0b999dc6966696052160f236	[ "arxiv", "semantic_scholar" ]	STAR-KV: Low-Rank KV Cache Compression via Soft Thresholding for Adaptive Rank Control	Low-rank projection has emerged as a promising approach for compressing the KV cache by exploiting hidden-dimension redundancy. However, prior methods rely on fixed or heuristic rank selection and struggle to achieve aggressive compression with minimal accuracy degradation. We propose STAR-KV, an adaptive low-rank KV c...	[ "Priyansh Bhatnagar", "Ashkan Moradifirouzabadi", "Se-Hyun Yang", "SeungJae Lee", "Jungwook Choi", "Mingu Kang" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-06-07T00:00:00	https://arxiv.org/abs/2606.08382	https://arxiv.org/pdf/2606.08382v1	2606.08382	null	0	true	https://github.com/PriyanshBhatnagar/STAR-KV	null	0.65
efc8440b17f8aaa7bdb84eb6002ed6eaa67e37e7480cd8e091cc563d8e6e1da6	[ "arxiv", "semantic_scholar" ]	RKSC: Reasoning-Aware KV Cache Sharing and Confident Early Exit for Multi-Step LLM Inference	We introduce RKSC (Reasoning-Aware KV Cache Sharing), a training-free inference framework that eliminates two structural redundancies in multi-branch LLM reasoning pipelines. ASKS (Attention-Similarity KV Sharing) computes the prefix KV cache once and broadcasts it to all semantically similar branches via hidden-state ...	[ "Anirudh Sekar" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-06-07T00:00:00	https://arxiv.org/abs/2606.09937	https://arxiv.org/pdf/2606.09937v1	2606.09937	null	0	true	https://github.com/AnirudhSekar/RKSC	null	0.65
33fae3d60cba349da24ac57e1f7e475f6068bbf2641e46632a9820997a80dbcd	[ "arxiv", "semantic_scholar" ]	SpectrumKV: Per-Token Mixed-Precision KV Cache Transfer for Prefill-Decode Disaggregated LLM Serving	Prefill-decode (PD) disaggregation decouples prompt processing from token generation, but it also turns the key-value (KV) cache into a network payload. Existing PD-side KV reduction methods are mostly binary: selected tokens are transmitted at full precision and the rest are not transmitted. This paper argues that bin...	[ "Yang Pengju" ]	[ "cs.LG", "cs.DC" ]	[ "Computer Science" ]	2026-06-07T00:00:00	https://arxiv.org/abs/2606.08635	https://arxiv.org/pdf/2606.08635v1	2606.08635	null	0	false	null	null	0.35
b23081733a04a06845371ec11f9dd0f8d8f83c9830e641ce0dcfa2a03cd0a1a3	[ "arxiv", "semantic_scholar" ]	IntentKV: Cross-Turn Intent-Aware KV Cache Pruning for Agent Inference	Multi-turn LLM agents fan short queries into long trajectories of tool calls, search results, and intermediate reasoning. Both KV memory and KV read bandwidth grow by orders of magnitude across a single trajectory, making the key-value (KV) cache, not parameter compute, the dominant serving bottleneck for long-horizon ...	[ "Junjie Li", "Jiong Lou", "Jie Li" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-06-06T00:00:00	https://arxiv.org/abs/2606.09916	https://arxiv.org/pdf/2606.09916v1	2606.09916	null	0	false	null	null	0.35
dbc3790ce5ee01b2483c84c367f615f419cf3aab4616a9c2bf9c9005338becab	[ "arxiv", "semantic_scholar" ]	Joint Structural Pruning and Mixed-Precision Quantization for LLM Compression	Recently, the efficiency of Large Language Models (LLMs) deployment has become a critical concern in practical applications. While post-training quantization (PTQ) and structural pruning are established techniques for reducing memory footprint and inference latency, most existing PTQ approaches optimize quantization er...	[ "Hoang-Loc La", "Truong-Thanh Le", "Amir Taherkordi", "Phuong Hoai Ha" ]	[ "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2026-06-05T00:00:00	https://arxiv.org/abs/2606.07819	https://arxiv.org/pdf/2606.07819v1	2606.07819	null	0	false	null	null	0.35
fc94663e3a52b334f0bb7bf17e50e656f58afafbafa092487ad58b721e75ed1c	[ "arxiv", "semantic_scholar" ]	Still: Amortized KV Cache Compaction in a Single Forward Pass	The KV cache is the memory bottleneck of long-horizon language model deployment. Practically, a deployable compactor must be lightweight enough to call during inference, expressive enough to preserve context under constraint, and reusable across a trajectory. Existing compaction methods satisfy only part of this requir...	[ "Charles O'Neill", "Alex Sandomirsky", "Harry Partridge", "Mudith Jayasekara", "Max Kirkby" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-06-05T00:00:00	https://arxiv.org/abs/2606.07878	https://arxiv.org/pdf/2606.07878v1	2606.07878	null	0	false	null	null	0.35
372102df62ed14b3d57d05fe483fa9c3d00f97f737e9bc0c395a409e85d8ff70	[ "arxiv", "semantic_scholar" ]	Rethinking LoRA Memory Through the Lens of KV Cache Compression	Parametric retrieval augmentation encodes document information into lightweight, document-specific modules such as LoRA adapters, reducing the need to include all evidence as input context. However, it remains unclear how this parameter-side memory interacts with context-side memory stored in the KV cache. We study thi...	[ "Chunsheng Zuo", "Liaoyaqi Wang", "William Jurayj", "William Fleshman", "Benjamin Van Durme" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-06-04T00:00:00	https://arxiv.org/abs/2606.05698	https://arxiv.org/pdf/2606.05698v1	2606.05698	null	0	false	null	null	0.35
51c0fafe4237deb7e37e634c4b1eb03507a834d97ea6de053d7c8a30353a889a	[ "arxiv", "semantic_scholar" ]	Value-and-Structure Alignment for Routing-Consistent Quantization of Mixture-of-Experts Models	Mixture-of-Experts (MoE) models scale foundation models efficiently by activating only a subset of experts for each token, but their large number of expert parameters still makes quantization essential for practical deployment. Unlike dense models, however, MoE models are sensitive to routing instability: small quantiz...	[ "Hancheol Park", "Geonho Lee", "Tairen Piao", "Tae-Ho Kim" ]	[ "cs.CL", "cs.AI" ]	[ "Computer Science" ]	2026-06-04T00:00:00	https://arxiv.org/abs/2606.05688	https://arxiv.org/pdf/2606.05688v1	2606.05688	null	0	false	null	null	0.35
471c2374e70edf4b8dca0cb9a2abe3a82a2e50ca6d72249e1675752607e033d6	[ "arxiv", "semantic_scholar" ]	High-Dimensional Theory of LoRA Fine-Tuning in a Solvable Attention Model	We develop a high-dimensional statistical theory of low-rank adaptation (LoRA) in attention models, capturing the interplay between pre-training and fine-tuning. We introduce a solvable framework in which a single-head attention layer is first pre-trained on a data-abundant task and subsequently adapted via a rank-one ...	[ "O. Duranthon", "F. Boncoraglio", "L. Zdeborová" ]	[ "cs.LG", "cond-mat.dis-nn" ]	[ "Computer Science", "Physics" ]	2026-06-04T00:00:00	https://arxiv.org/abs/2606.05899	https://arxiv.org/pdf/2606.05899v1	2606.05899	null	0	false	null	null	0.35
c4b8ac494f3da3094edc9fca7613c163c350b710455dac1c9809d5594defdd47	[ "arxiv", "semantic_scholar" ]	Tangram: Unlocking Non-Uniform KV Cache for Efficient Multi-turn LLM Serving	Multi-turn Large Language Model (LLM) serving is critical for consistent user experiences, yet the linear growth of the Key-Value (KV) cache imposes significant pressure on GPU memory and bandwidth. Non-uniform KV compression effectively preserves more information by considering the individual importance of each KV cac...	[ "Hyungmin Kim", "Minsoo Kim", "Hongseok Kim", "Jungwook Choi" ]	[ "cs.LG", "cs.SE" ]	[ "Computer Science" ]	2026-06-04T00:00:00	https://arxiv.org/abs/2606.06302	https://arxiv.org/pdf/2606.06302v1	2606.06302	null	0	true	https://github.com/aiha-lab/TANGRAM	null	0.65
129136d5dd14aef1a6b02c7efcdccb266321e11616a9bdeefbdd807fd2d42058	[ "arxiv", "semantic_scholar" ]	RedKnot: Efficient Long-Context LLM Serving with Head-Aware KV Reuse and SegPagedAttention	As the input length of large language model (LLM) serving continues to grow, the KV cache has become a dominant bottleneck in AI infrastructure. It limits GPU memory capacity, serving concurrency, cache reuse, and distributed scalability. Several important problems, including position-independent KV cache, prefix KV ca...	[ "Yang Liu", "ZhaoKai Luo", "HuaYi Jin", "ZhiYong Wang", "RuoZhou He", "BoYu Wang", "Guanjie Chen", "Junhao Hu" ]	[ "cs.AI" ]	[ "Computer Science" ]	2026-06-04T00:00:00	https://arxiv.org/abs/2606.06256	https://arxiv.org/pdf/2606.06256v1	2606.06256	null	0	false	null	null	0.35
5e79679071a5fc73dbb9b13e6c745fa1e5037e94db9182a12319b137c46ffb3f	[ "arxiv", "semantic_scholar" ]	TENP: Trapezoidal Expert Neuron Pruning For Mixture-of-Experts	Mixture-of-Experts large language models (LLMs) scale efficiently through sparse activation, yet their deployment is fundamentally constrained by the large static parameter footprint of experts. Existing compression approaches either remove entire experts, disrupting routing topology and harming performance, or rely on...	[ "Jiangyang He", "Shaolin Zhu", "Deyi Xiong" ]	[ "cs.LG", "stat.ML" ]	[ "Computer Science", "Mathematics" ]	2026-06-03T00:00:00	https://arxiv.org/abs/2606.09885	https://arxiv.org/pdf/2606.09885v1	2606.09885	null	0	false	null	null	0.35
ff4d98f6b0b06cd072a826b0088f9a2be1ac76f474e701014d7d0e6caf8817ff	[ "arxiv", "semantic_scholar" ]	SHAPE: Coalition-Aware Expert Pruning for Sparse Mixture-of-Experts LLMs	Sparse Mixture-of-Experts (MoE) large language models achieve strong quality with low per-token compute, yet their deployment is often limited by the memory wall: the full expert pool must remain resident to support token-dependent routing. Expert pruning is a direct remedy, but prior criteria typically score experts i...	[ "Yuhao Zhang" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-06-03T00:00:00	https://arxiv.org/abs/2606.09886	https://arxiv.org/pdf/2606.09886v1	2606.09886	null	0	true	https://github.com/Alizen-1009/Shapley-Moe	null	0.65
537f9582cfd26567a2497952d288598b0a2b66512119b9c0727089ab5cade6a8	[ "arxiv", "semantic_scholar" ]	AlphaQ: Calibration-Free Bit Allocation for Mixture-of-Experts Quantization	Mixture-of-Experts (MoE) architectures scale model capacity through sparse expert activation, but their deployment remains memory-bound because all expert weights must reside in memory. Mixed-precision quantization can substantially reduce this footprint by assigning different bit-widths to different experts. Existing ...	[ "Wanqi Yang", "Yuexiao Ma", "Alexander Conzelmann", "Xiawu Zheng", "Michael W. Mahoney", "T. Konstantin Rusch", "Shiwei Liu" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-06-03T00:00:00	https://arxiv.org/abs/2606.04980	https://arxiv.org/pdf/2606.04980v1	2606.04980	null	0	true	https://github.com/Superone77/AlphaQ	null	0.65
a0f8e959719bbb512617d133507d6f0ef3ba6ad79a0bc204ce1a0e227db08f48	[ "arxiv", "semantic_scholar" ]	Cartridges at Scale: Training Modular KV Caches over Large Document Collections	Large Language Models can reason over long contexts, yet prefilling millions of tokens is wasteful as much of the content remains static across queries. Cartridges address this by distilling document collections into reusable key-value (KV) caches that eliminate prefilling while preserving accuracy. A critical limitati...	[ "Momchil Hardalov", "Gonzalo Iglesias", "Adrià de Gispert" ]	[ "cs.CL", "cs.IR", "cs.LG" ]	[ "Computer Science" ]	2026-06-03T00:00:00	https://arxiv.org/abs/2606.04557	https://arxiv.org/pdf/2606.04557v1	2606.04557	null	0	false	null	null	0.35
77729832691f6dcb2f801e0f5b838fbbc5d527cd03d27068d1e61d8f3715fd4e	[ "arxiv", "semantic_scholar" ]	KVarN: Variance-Normalized KV-Cache Quantization Mitigates Error Accumulation in Reasoning Tasks	Test-time scaling is a powerful approach to obtain better reasoning in large language models, but it becomes memory-bottlenecked during long-horizon decoding, as the KV-cache grows. KV-cache quantization can help improve this, but current methods are evaluated under prefill-like settings and errors behave differently u...	[ "Lorenz K. Muller", "Philippe Bich", "Chiara Boretti", "Hyun-Min Chang", "Jiawei Zhuang", "Lukas Cavigelli" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-06-02T00:00:00	https://arxiv.org/abs/2606.03458	https://arxiv.org/pdf/2606.03458v1	2606.03458	null	0	true	https://github.com/huawei-csl/KVarN	null	0.65
c68b97e1f51efc1fd997dbcd5b61cbb54762a477c10b83230598596ab9ced734	[ "arxiv", "semantic_scholar" ]	Value-Aware Stochastic KV Cache Eviction for Reasoning Models	Reasoning models improve accuracy through extended chains of thought, but their long outputs create a memory and compute bottleneck. KV cache eviction methods reduce this cost by evicting unimportant key-value pairs from the cache, yet they often yield worse accuracy than selection-based sparse attention alternatives, ...	[ "Ting-Yun Chang", "Harvey Yiyun Fu", "Deqing Fu", "Chenghao Yang", "Jesse Thomason", "Robin Jia" ]	[ "cs.LG", "cs.CL" ]	[ "Computer Science" ]	2026-06-02T00:00:00	https://arxiv.org/abs/2606.03928	https://arxiv.org/pdf/2606.03928v1	2606.03928	null	0	true	https://github.com/terarachang/VaSE	null	0.65
7681332cc50b38aa553bfe6aab5d226bdad32a0775c5903011864fe46027c399	[ "arxiv", "semantic_scholar" ]	When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics	Vision-Language Models (VLMs) have demonstrated remarkable capabilities but suffer from significant computational overhead during inference. While visual token pruning offers a promising solution, existing methods predominantly rely on initial attention scores. This single-metric paradigm presents a critical flaw: high...	[ "Jiahui Wang", "Kai Zhang", "Mai Han", "Huanghe Zhang" ]	[ "cs.CV", "cs.AI" ]	[ "Computer Science" ]	2026-06-02T00:00:00	https://arxiv.org/abs/2606.03569	https://arxiv.org/pdf/2606.03569v1	2606.03569	null	0	false	null	null	0.35
57fa458b7940c77d7244cde31c9cbc114944961a12c01886d993d23623d27554	[ "arxiv", "semantic_scholar" ]	Recover-LoRA for Aggressive Quantization: Reclaiming Accuracy in 2-Bit Language Models via Low-Rank Adaptation with Knowledge Distillation on Synthetic Data	Aggressive weight quantization to 2-bit precision offers substantial throughput and memory gains for large language model (LLM) inference, but typically incurs severe accuracy degradation. These gains are particularly relevant for edge and on-device deployment, where memory capacity and bandwidth are primary constraint...	[ "Devleena Das", "Rajeev Patwari", "Elliott Delaye", "Ashish Sirasao" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-06-02T00:00:00	https://arxiv.org/abs/2606.04238	https://arxiv.org/pdf/2606.04238v1	2606.04238	null	0	false	null	null	0.35
ffa7dd593613a170a8aefdf8639b244173ec134c9309d4131f852ba6e5d3818e	[ "arxiv", "semantic_scholar" ]	Sparse Mixture-of-Experts Reward Models Learn Interpretable and Specialized Experts for Personalized Preference Modeling	Preference modeling plays a central role in reinforcement learning from human feedback (RLHF), enabling large language models (LLMs) to align with human values. However, most existing approaches assume a universal reward function, neglecting the diversity and heterogeneity of human preferences. To address this limitati...	[ "Yifan Wang", "Jinyi Mu", "Mayank Jobanputra", "Yu Wang", "Ji-Ung Lee", "Soyoung Oh", "Isabel Valera", "Vera Demberg" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-06-02T00:00:00	https://arxiv.org/abs/2606.04284	https://arxiv.org/pdf/2606.04284v1	2606.04284	null	0	false	null	null	0.35
223efefd910aa3e306c10b25d31094236bc1e17c222b923eec57e564dd79e118	[ "arxiv", "semantic_scholar" ]	Multi-Segment Attention: Enabling Efficient KV-Cache Management for Faster Large Language Model Serving	Large Language Model (LLM) inference relies on key-value (KV) caches to avoid redundant attention computation. While approximate KV cache retention techniques reduce memory usage by sacrificing model accuracy, lossless approaches instead evict KV cache blocks from GPU memory and reconstruct them on demand to preserve e...	[ "Chunan Shi", "Yilei Chen", "Yilin Chen", "Xupeng Miao", "Bin Cui" ]	[ "cs.AR", "cs.CL", "cs.LG" ]	[ "Computer Science" ]	2026-06-01T00:00:00	https://arxiv.org/abs/2606.02964	https://arxiv.org/pdf/2606.02964v1	2606.02964	null	0	false	null	null	0.35
01cad6f7756b1cae7bee87682d79867eba6abae69871100607ffff9f192bfe9d	[ "arxiv", "semantic_scholar" ]	Alignment Collapse Under KV Cache Quantization: Diagnosis and Mitigation	Key-value (KV) cache quantization is widely used to reduce Large Language Model (LLM) inference memory, yet existing evaluations solely focus on measuring perplexity and accuracy without assessing the safety impact. In this study, we explore alignment preservation under KV cache quantization. Across eleven instruction-...	[ "Bruce Changlong Xu", "Adarsh Kumarappan", "Mu Zhou" ]	[ "cs.LG", "cs.AI", "cs.ET" ]	[ "Computer Science" ]	2026-06-01T00:00:00	https://arxiv.org/abs/2606.09864	https://arxiv.org/pdf/2606.09864v1	2606.09864	null	0	false	null	null	0.35
ebaadca24f30b7f91ac90e2f14a6d5f63864ec97f3eebf6db9f12d587fa98ca1	[ "arxiv", "semantic_scholar" ]	STaR-KV: Spatio-Temporal Adaptive Re-weighting for KV Cache Compression in GUI Vision-Language Models	Vision-language-model-based graphical user interface (GUI) agents have shown broad automation capabilities, yet deployment is bottlenecked by a key-value (KV) cache that grows linearly with interaction steps. For instance, UI-TARS-1.5-7B consumes 76 GB of GPU memory on merely five screenshots, approaching the capacity ...	[ "Yuhang Han", "Wenzheng Yang", "Yujie Chen", "Xiangqi Jin", "Yaojie Zhang", "Siteng Huang", "Linfeng Zhang" ]	[ "cs.CV", "cs.AI" ]	[ "Computer Science" ]	2026-06-01T00:00:00	https://arxiv.org/abs/2606.01790	https://arxiv.org/pdf/2606.01790v1	2606.01790	null	0	true	https://github.com/kawhiiiileo/STaR-KV	null	0.65
21cec4d20fc191f6dd2e40165103f4531030b51c2ccda7f55a6a2fc4e34f34a4	[ "arxiv", "semantic_scholar" ]	SparseX: Efficient Segment-Level KV Cache Sharing for Interleaved LLM Serving	In long-context LLM serving, the prefill stage often dominates time-to-first-token and computational cost. Although Prefix Cache in vLLM/PagedAttention has been widely used to reuse identical prompt prefixes, repeated content in practical applications frequently appears as non-prefix, cross-request, cross-turn, and cro...	[ "Quqing Zhang", "Kai Chen", "Ning Liao", "Zehao Lin", "Bo Tang", "Feiyu Xiong", "Zhiyu Li", "Xiaoxing Wang" ]	[ "cs.PF" ]	[ "Computer Science" ]	2026-06-01T00:00:00	https://arxiv.org/abs/2606.01751	https://arxiv.org/pdf/2606.01751v2	2606.01751	null	0	false	null	null	0.35
88daca95816060cd6b9d90ec55dda0248756ec9b3b2fd67349693fd80469acb0	[ "arxiv", "semantic_scholar" ]	MomentKV: Closing the Directional Gap in KV Cache Eviction for Long-Context Inference	Autoregressive decoding in Transformer-based language models relies on the KV cache, whose memory footprint grows linearly with sequence length and becomes the primary bottleneck for long-context inference. KV cache eviction addresses this by retaining a fixed-size subset of key-value pairs and discarding the rest. We ...	[ "Yu Li", "Binxu Li", "Tian Lan" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-06-01T00:00:00	https://arxiv.org/abs/2606.01563	https://arxiv.org/pdf/2606.01563v1	2606.01563	null	0	false	null	null	0.35
84a56ce3f3c4253499659b6510485f8b267ea5ca1eb3e8bd49c5ee207fb2c611	[ "arxiv", "semantic_scholar" ]	ProbMoE: Differentiable Probabilistic Routing for Mixture-of-Experts	Mixture-of-Experts (MoE) models scale by activating only a small subset of experts per token. However, training such models remains challenging because top-$k$ routing is discrete and non-differentiable, requiring gradient estimators for expert selection whose design remains a central open problem. We introduce ProbMoE...	[ "Heng Zhao", "Zilei Shao", "Guy Van den Broeck", "Zhe Zeng" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-06-01T00:00:00	https://arxiv.org/abs/2606.01509	https://arxiv.org/pdf/2606.01509v1	2606.01509	null	0	false	null	null	0.35
18b4c9e5c93899223d77e3e61292cf70ee4e7143cfbb027d6f3e662fbfada7d8	[ "arxiv", "semantic_scholar" ]	GC-MoE: Genomics-Guided Cell-Type-Specific Mixture of Experts for Histology-Based Single-Cell Spatial Transcriptomics	Histology-based single-cell spatial transcriptomics (ST) estimation aims to predict gene expression for individual cells from histopathological images and cell locations, reducing the need for costly single-cell ST measurements. Unlike existing histology-to-ST methods that mainly predict spot-level profiles for local r...	[ "Kaito Shiku", "Ahtisham Fazeel Abbasi", "Ryoma Bise", "Yuichiro Iwashita", "Kazuya Nishimura", "Andreas Dengel", "Muhammad Nabeel Asim" ]	[ "cs.CV", "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2026-06-01T00:00:00	https://arxiv.org/abs/2606.02424	https://arxiv.org/pdf/2606.02424v1	2606.02424	null	0	false	null	null	0.35
fed23812d85c65e9e76f2e00ab602a797fbd6cbc608e73c2d9fd3955f4c48089	[ "arxiv", "semantic_scholar" ]	Fail-Closed Lowering of Resident KV Claims onto LLM Serving Runtimes	LLM serving runtimes increasingly expose KV-cache primitives that resemble future-reuse controls: retention priority, TTL-like duration, host or storage offload, block events, active no-evict scheduling, and KV-aware routing. This paper argues that such primitives are weaker than accepted future-KV obligations. A runti...	[ "Lukas Stepanek" ]	[ "cs.DC" ]	[ "Computer Science" ]	2026-05-31T00:00:00	https://arxiv.org/abs/2606.01387	https://arxiv.org/pdf/2606.01387v1	2606.01387	null	0	true	https://github.com/gustavgauge/resident-kv-lowering-artifact	null	0.65
5e08d6e964fd07b595469996c0e5cc94a098d4e5f52a5f5576a071f08bb36269	[ "arxiv", "semantic_scholar" ]	Leyline: KV Cache Directives for Agentic Inference	Modern KV cache management assumes the chatbot workload: prompts arrive once and the cache grows append-only, so prefix caching and forward-only eviction are correct by construction. Agentic LLMs break this assumption. Their conversations evolve through policy-driven editing: failed tool calls are retried, stale output...	[ "Bole Ma", "Jan Eitzinger", "Harald Koestler" ]	[ "cs.DC", "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2026-05-31T00:00:00	https://arxiv.org/abs/2606.01065	https://arxiv.org/pdf/2606.01065v1	2606.01065	null	0	false	null	null	0.35
4566c8ebabf44446d1e7dcb104792c586e106c406342110f37bd8775152326b2	[ "arxiv", "semantic_scholar" ]	Move the Query, Not the Cache: Characterizing Cross-Instance Latent Attention Redistribution Across GPU Fabrics	Frontier LLMs increasingly decide what a query attends to with a sparse-attention indexer that picks a few KV-cache blocks per query: attention's unit is now a small, reusable chunk. Agentic workloads hammer it: many sub-agents query one large codebase, reusing the same blocks. When that corpus outgrows one GPU it is p...	[ "Bole Ma", "Jan Eitzinger", "Harald Köstler", "Gerhard Wellein" ]	[ "cs.DC", "cs.AI", "cs.NI" ]	[ "Computer Science" ]	2026-05-31T00:00:00	https://arxiv.org/abs/2606.01502	https://arxiv.org/pdf/2606.01502v1	2606.01502	null	0	false	null	null	0.35
eb64acd20eec9ae04601a7c5b6c9f96227749847698233b4f7fbde795831ab90	[ "arxiv", "semantic_scholar" ]	GPTQ-intrinsic LoRA: A Near-optimal Algorithm for Low-precision Quantization with Low-rank Adaptation	Post-training quantization is widely used for compressing large neural networks, but aggressive low-bit quantization can significantly degrade model quality. A common remedy is to augment the quantized weights with a low-rank correction, leading to approximations of the form $W\approx Q+LR$. In this paper, we study thi...	[ "Shihao Zhang", "Rayan Saab" ]	[ "cs.LG", "cs.IT" ]	[ "Computer Science", "Mathematics" ]	2026-05-31T00:00:00	https://arxiv.org/abs/2606.01412	https://arxiv.org/pdf/2606.01412v1	2606.01412	null	0	false	null	null	0.35
165ca59c1fddca3100ab19b859e0e396026aa5e16487e06b79ebc5551157fbab	[ "arxiv", "semantic_scholar" ]	Deft Scheduling of Dynamic Cloud Workflows with Varying Deadlines via Mixture-of-Experts	Workflow scheduling in cloud computing demands the intelligent allocation of dynamically arriving, graph-structured workflows with varying deadlines onto ever-changing virtual machine resources. However, existing deep reinforcement learning (DRL) schedulers remain limited by rigid, single-path inference architectures t...	[ "Ya Shen", "Gang Chen", "Hui Ma", "Mengjie Zhang" ]	[ "cs.AI" ]	[ "Computer Science" ]	2026-05-31T00:00:00	https://arxiv.org/abs/2606.01162	https://arxiv.org/pdf/2606.01162v2	2606.01162	null	0	false	null	null	0.35
27ad4d392659179ee22e669923b194321238275ce5ff05d1c3f0ecddffe1e993	[ "arxiv", "semantic_scholar" ]	DAG-MoE: From Simple Mixture to Structural Aggregation in Mixture-of-Experts	Mixture-of-Experts (MoE) models have become a leading approach for decoupling parameter count from computational cost in large language models, yet effectively scaling MoE performance remains a challenge. Prior work shows that fine-grained experts enlarge the space of expert combinations and improve flexibility, but th...	[ "Jiarui Feng", "Hanqing Zeng", "Karish Grover", "Ruizhong Qiu", "Yinglong Xia", "Qiang Zhang", "Qifan Wang", "Ren Chen", "Dongqi Fu", "Jiayi Liu", "Zhoukai Zhao", "Xiangjun Fan", "Benyu Zhang", "Yixin Chen" ]	[ "cs.AI" ]	[ "Computer Science" ]	2026-05-31T00:00:00	https://arxiv.org/abs/2606.01062	https://arxiv.org/pdf/2606.01062v1	2606.01062	null	1	false	null	null	0.35
d2c04403620606df3da04a69bcb42bb0e5c62701b1c8a8b15bc70e399f95d134	[ "arxiv", "semantic_scholar" ]	WaveFilter: Enhancing the Long-Context Capability of Diffusion LLMs via Wavelet-Guided KV Cache Filtering	Diffusion Large Language Models (DLMs) have demonstrated significant advantages across various tasks. However, constrained by their multi-step iterative inference mechanism, their computational overhead and inference latency in long-context tasks have become core bottlenecks restricting their large-scale deployment. Wh...	[ "Jinnan Yang", "Yan Wang", "Zhen Bi", "Kehao Wu", "Xiaojie Li", "Jungang Lou", "Zechao Li", "Jing Liu" ]	[ "cs.CL", "cs.AI" ]	[ "Computer Science" ]	2026-05-30T00:00:00	https://arxiv.org/abs/2606.00724	https://arxiv.org/pdf/2606.00724v1	2606.00724	null	0	false	null	null	0.35
606626c8a61150db7ec47a49c8898a828dc66bda9b788898ead0a2ede2743c3f	[ "arxiv", "semantic_scholar" ]	GRKV: Global Regression for Training-Free KV Cache Compression in Long-Context LLMs	Large language models (LLMs) with extended context lengths rely on the key-value (KV) cache to support attention over prior tokens. However, maintaining the KV cache incurs substantial memory overhead, motivating KV-cache compression methods that enforce a fixed budget through eviction and merging. Modern eviction meth...	[ "Junjie Peng", "You Wu", "Haoyi Wu", "Jialong Han", "Xiaohua Xie", "Kewei Tu", "Jianhuang Lai" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-05-29T00:00:00	https://arxiv.org/abs/2605.31105	https://arxiv.org/pdf/2605.31105v1	2605.31105	null	0	false	null	null	0.35
cbb76144a4bd9f99ccc8170bc2fc3b0b971a9977413a76d0b9f2dbf5712474db	[ "arxiv", "semantic_scholar" ]	MoG: Mixture of Experts for Graph-based Retrieval-Augmented Generation	Retrieval-augmented generation is intensively studied to ground large language models on external evidence. However, retrieving from a unified knowledge base could inevitably introduce irrelevant information that may mislead generation for complex reasoning. Inspired by the conditional computation of mixture of experts...	[ "Zheng Yuan", "Chuang Zhou", "Linhao Luo", "Siyu An", "Di Yin", "Xing Sun", "Xiao Huang" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-05-29T00:00:00	https://arxiv.org/abs/2605.31010	https://arxiv.org/pdf/2605.31010v1	2605.31010	null	0	true	https://github.com/DEEP-PolyU/MoG	null	0.65
480b2444ea2621b8205612a32ebaf89754a2c92b624a19df5ece5cf3830de48e	[ "arxiv", "semantic_scholar" ]	Moment-KV: Momentum-Based Decode-Time KV Cache Compression for Long Generation	Key-Value (KV) cache remains a major bottleneck for deploying Large Language Models (LLMs) in long-generation tasks. Prior work often applies uniform compression across both prefill and decoding caches, but compressing the prefill cache degrades performance by corrupting critical context. While preserving the prefill c...	[ "Soumyadeep Jana", "Sagar Nishad", "Sanasam Ranbir Singh" ]	[ "cs.AI" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.29873	https://arxiv.org/pdf/2605.29873v1	2605.29873	null	0	false	null	null	0.35
d849f156bd6459011ea800f1917f617cd5a83f9c30c555d2e3af929ad27431bf	[ "arxiv", "semantic_scholar" ]	Probing the Prompt KV Cache: Where It Becomes Dispensable	Prior KV cache compression schemes empirically demonstrate that the prompt cache is partially redundant during decoding, dropping or summarising entries with little accuracy loss. We ask when and what kind of redundancy: at which layers, after how many decoding steps, and in what form can the prompt span KV cache be re...	[ "Vinayshekhar Bannihatti Kumar", "Manoj Ghuhan Arivazhagan", "Disha Makhija", "Rashmi Gangadharaiah" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.30574	https://arxiv.org/pdf/2605.30574v1	2605.30574	null	0	false	null	null	0.35
18ae5baff67ad341235cfead0b2c46642ff6bdc0a5c93d428495a3465b0951b2	[ "arxiv", "semantic_scholar" ]	Token-Level Generalization in LoRA Adapter Backdoors: Attack Characterization and Behavioral Detection	We show that LoRA adapters, the dominant distribution format for fine-tuned LLMs, can be reliably backdoored through training data poisoning while preserving baseline task performance. On a Qwen 2.5 1.5B prompt-injection classifier, a small fraction of poisoned examples drives a clean-accuracy-preserving backdoor to sa...	[ "Travis Lelle" ]	[ "cs.CR", "cs.AI", "cs.CL", "cs.LG" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.30189	https://arxiv.org/pdf/2605.30189v1	2605.30189	null	0	true	https://github.com/Travis-ML/lora-backdoors	null	0.65
8fbc8848c46d26f4c106e48ab59488ac9f15c3d1d73c98fc1f42964f404622ed	[ "arxiv", "semantic_scholar" ]	Graph-Conditioned Mixture of Graph Neural Network Experts for Traffic Forecasting	Spatio-temporal forecasting on sensor graphs is commonly tackled with a single backbone architecture applied uniformly across all nodes, although graph regions can exhibit different dynamics. Road segments differ in functional class, structure, and traffic behavior, suggesting that node-wise expert specialization can b...	[ "Amirhossein Ghaffari", "Saeid Sheikhi", "Ekaterina Gilman" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.30486	https://arxiv.org/pdf/2605.30486v1	2605.30486	null	0	true	https://github.com/Ahghaffari/gc_moe	null	0.65
d9d3a45758265d7004f68f7b5644d57b939c64ac32ab6065d7d84d6e9a95053e	[ "arxiv", "semantic_scholar" ]	Future Forcing: Future-aware Training-free KV Cache Policy for Autoregressive Video Generation	Autoregressive (AR) video generation has emerged as a promising paradigm for long-horizon video synthesis, where each frame is generated conditioned on previously generated tokens. To accelerate inference, the KV cache is used to avoid redundant recomputation across generation steps. Nevertheless, its growth with gener...	[ "Jiayi Luo", "Qiyan Liu", "Tengyang Wang", "JunHao Liu", "Jiayu Chen", "Cong Wang", "Hanxin Zhu", "Chen Gao", "Xiaobin Hu", "Qingyun Sun", "Zhibo Chen" ]	[ "cs.CV" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.30083	https://arxiv.org/pdf/2605.30083v1	2605.30083	null	0	false	null	null	0.35
57134b7ec601baa633ea4fb7453e121b6e78cb924efcfe82136ff4649fac06b6	[ "arxiv", "semantic_scholar" ]	SAFE-Pruner: Semantic Attention-Guided Future-Aware Token Pruning for Efficient Vision-Language-Action Manipulation	Real-time inference of vision-language-action (VLA) models is essential for robotic control. While visual token pruning has shown strong potential for accelerating inference, most existing methods mainly base pruning decisions on shallow-layer cues and risk discarding visual information required by deep layers. To addr...	[ "Shilin Ma", "Chubin Zhang", "Changyuan Wang", "Yuji Wang", "Yue Wu", "Zixuan Wang", "Jingqi Tian", "Zheng Zhu", "Yansong Tang" ]	[ "cs.CV", "cs.RO" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.29662	https://arxiv.org/pdf/2605.29662v1	2605.29662	null	0	false	null	null	0.35
05f97c39813203b864b3adbdc470c0df905d404ef2f6075e18a4117bfe066f28	[ "arxiv", "semantic_scholar" ]	VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion	Long-rollout causal video diffusion has converged on a fixed-size sliding-window KV cache, with recent progress innovating within this layout by changing which tokens occupy the window or how their positions are encoded. The per-head KV layout itself, a dominant contributor to streaming memory and latency, has been mos...	[ "Hidir Yesiltepe", "Jiazhen Hu", "Tuna Han Salih Meral", "Adil Kaan Akan", "Kaan Oktay", "Hoda Eldardiry", "Pinar Yanardag" ]	[ "cs.CV", "cs.AI" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.30351	https://arxiv.org/pdf/2605.30351v1	2605.30351	null	0	false	null	null	0.35
a8d3312b5c7c7d574e91e4e921cbaea3ebbcfb099210631933c18bf8e7f457ab	[ "arxiv", "semantic_scholar" ]	Understanding Safety-Sensitive Expert Behavior in Mixture-of-Experts LLMs	Mixture-of-Experts (MoE) LLMs rely on sparse, router-driven expert activation, yet how safety alignment interacts with routed expert specialization remains underexplored. A common intuition is that safety behavior may be controlled by routing harmful requests to distinct refusal-oriented experts. In this work, we provi...	[ "Zhibo Zhang", "Yuxi Li", "Zhen Ouyang", "Ling Shi", "Kailong Wang" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.29708	https://arxiv.org/pdf/2605.29708v1	2605.29708	null	0	false	null	null	0.35
b3cb9bc4db16463797c29f7aeb159da90a71881165b7cf0408eb0ab78e576a67	[ "arxiv", "semantic_scholar" ]	LoRA-Key: User-Centric LoRA Watermarking for Text-to-Image Diffusion Models	Low-Rank Adaptation (LoRA) has become a widely used mechanism for customizing text-to-image diffusion models, enabling lightweight modules that are shared, reused, and commercialized as independent assets. This LoRA-centric ecosystem shifts copyright protection from foundation models to distributed LoRA modules, which ...	[ "Yaopeng Wang", "Qingliang Wang", "Zhibo Wang", "Huiyu Xu", "Jiacheng Du", "Qiu Wang", "Jia-Li Yin", "Kui Ren" ]	[ "cs.CR" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.29569	https://arxiv.org/pdf/2605.29569v2	2605.29569	null	0	false	null	null	0.35
d5819cb62e548d24b148ce1fb26508010e4dccd0321b6e2833c65191cd07bf57	[ "arxiv", "semantic_scholar" ]	Analyzing Quality-Latency-Resource Trade-offs in a Technical Documentation RAG Assistant Using LoRA Adaptation	We study quality-latency-resource trade-offs in a documentation-grounded retrieval-augmented generation (RAG) system that uses Low-Rank Adaptation (LoRA) of the generator. We build a manually verified benchmark of 5,144 question-answer pairs over the official Kubernetes documentation and combine it with a fixed hybrid-...	[ "Evgenii Palnikov", "Elizaveta Gavrilova" ]	[ "cs.CL", "cs.IR", "cs.LG" ]	[ "Computer Science" ]	2026-05-27T00:00:00	https://arxiv.org/abs/2605.28222	https://arxiv.org/pdf/2605.28222v1	2605.28222	null	0	true	https://github.com/EugPal/rag-lora-tradeoffs	null	0.65
b072a449dec1c46642779094f317b82e93465738f3000f82df42b3ee9594c5da	[ "arxiv", "semantic_scholar" ]	Pruning and Distilling Mixture-of-Experts into Dense Language Models	Mixture-of-Experts (MoE) is now the dominant architecture for frontier language models, yet it requires all expert parameters to be loaded in memory, making it less preferable for memory-constrained deployment. Existing compression methods reduce the number of experts but the output remains an MoE model with the same f...	[ "Junhyuck Kim", "Jihun Yun", "Haechan Kim", "Gyeongman Kim", "Joonghyun Bae", "Jaewoong Cho" ]	[ "cs.CL", "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2026-05-27T00:00:00	https://arxiv.org/abs/2605.28207	https://arxiv.org/pdf/2605.28207v2	2605.28207	null	0	false	null	null	0.35
d6fbfe91c025e87a885ee867b695f85a57a46df4c6731fc5a274d8c1a8823ef6	[ "arxiv", "semantic_scholar" ]	Extracting Small Translation Specialists from LLMs by Aggressively Pruning Experts	Modern large language models (LLMs) achieve state-of-the-art machine translation performance, but they do so as broad generalists largely trained for many tasks and capabilities unrelated to translation. Thus, they are heavily overparameterized for this task, resulting in excessive memory and compute requirements. In t...	[ "Liu O. Martin", "Lucas Bandarkar", "Nanyun Peng" ]	[ "cs.CL", "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2026-05-27T00:00:00	https://arxiv.org/abs/2605.28042	https://arxiv.org/pdf/2605.28042v1	2605.28042	null	0	false	null	null	0.35
72cdce781e1992da82f551a21468663f4825f84ddd6f5f12ca6d9c842b129a98	[ "arxiv", "semantic_scholar" ]	Augmenting Attention with Exponentially Decaying Memory Improves Query-Aware KV Sparsity	Efficient inference is critical for long-context language models, where attention computation and KV-cache access dominate the cost. Recent work RAT+, introduces a recurrence-augmented attention backbone that enables flexible dilated attention at inference time. In this paper, we investigate whether this exponentially ...	[ "Xiuying Wei", "Caglar Gulcehre" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-05-27T00:00:00	https://arxiv.org/abs/2605.28640	https://arxiv.org/pdf/2605.28640v1	2605.28640	null	0	false	null	null	0.35
f40b39281efe4da642a26e1f62501aeb33b8e8b1556525844cb5fe30cf72a9be	[ "arxiv", "semantic_scholar" ]	VidPrism: Heterogeneous Mixture of Experts for Image-to-Video Transfer	With the rapid development of pre-training technologies, adapting large-scale Vision-Language Models (VLMs) for video understanding \emph{\ie} image-to-video transfer learning has become a dominant paradigm. To achieve superior performance, it raises as an effective strategy among recent advances to employ Mixture-of-E...	[ "Rui Lin", "Chuanming Wang", "Huadong Ma" ]	[ "cs.CV", "cs.AI" ]	[ "Computer Science" ]	2026-05-27T00:00:00	https://arxiv.org/abs/2605.28229	https://arxiv.org/pdf/2605.28229v1	2605.28229	null	0	true	https://github.com/Lrrrr549/VidPrism.git}{https://github.com/Lrrrr549/VidPrism.git}	null	0.65
619782071139f2881c62308f33d2304efce763456816654f39ae2d84631f464b	[ "arxiv", "semantic_scholar" ]	Hurwitz Quaternion Multiplicative Quantization for KV Cache Compression	We propose \textbf{Hurwitz Quaternion Multiplicative Quantization (HQMQ)}, a \textbf{calibration-free} method for KV cache compression of large language models. HQMQ treats each 4-element chunk of K or V as a quaternion and quantizes its unit direction to the \emph{product} $q_p \cdot q_s$, where $q_p$ ranges over the ...	[ "Kabir Swain", "Sijie Han", "Daniel Karl I. Weidele", "Mauro Martino", "David Cox", "Antonio Torralba" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-26T00:00:00	https://arxiv.org/abs/2605.27646	https://arxiv.org/pdf/2605.27646v1	2605.27646	null	0	false	null	null	0.35
06131d1f2dfad5b268fe0742e54b50b455356a5f0e4b8efefd331ee7850b2e04	[ "arxiv", "semantic_scholar" ]	Enabling KV Caching of Shared Prefix for Diffusion Language Models	Key-value (KV) caching for shared prefixes is essential for high-throughput large language model (LLM) serving, but it faces critical challenges in emerging diffusion language models (DLMs). In DLMs, bidirectional attention means that updating any token dynamically alters the entire context and its corresponding KVs. T...	[ "Younghun Go", "Jaehoon Han", "Changyong Shin", "Chuk Yoo", "Gyeongsik Yang" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-26T00:00:00	https://arxiv.org/abs/2606.07571	https://arxiv.org/pdf/2606.07571v1	2606.07571	null	0	false	null	null	0.35
dd22e611ba08ed40d42c79aefc6759615a0666ae6a2764c04ba3a401fd3541de	[ "arxiv", "semantic_scholar" ]	NestedKV: Nested Memory Routing for Long-Context KV Cache Compression	Long-context language models are limited by the memory footprint of the key-value (KV) cache. Existing training-free KV compression methods usually rank tokens by one importance signal -- attention, recency, layer-wise allocation, or key distinctiveness -- which becomes brittle when useful context is globally distincti...	[ "Hong Chen", "Xiang Liu", "Yubo Gao", "Yuxuan Fan", "Bo Wang", "Yuanlin Chu", "Yuanguo Lin", "Xuming Hu" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-05-26T00:00:00	https://arxiv.org/abs/2605.26678	https://arxiv.org/pdf/2605.26678v1	2605.26678	null	0	false	null	null	0.35
05e450fd9df6d4ac973de0e490ef76323eb52505ec6ba032dd89f249b5f433c8	[ "arxiv", "semantic_scholar" ]	Grounded Cache Routing for Retrieval-Augmented Generation: When Is It Safe to Reuse an Answer?	Modern retrieval-augmented generation(RAG) deployments increasingly rely on caching to reduce token cost and time-to-first-token(TTFT). Prefix-level KV reuse is now standard in serving stacks such as vLLM, and chunk-level and position-independent reuse have been pushed further by recent systems(RAGCache, TurboRAG, Cach...	[ "Syed Huma Shah" ]	[ "cs.CR", "cs.AI", "cs.CL", "cs.IR", "cs.LG" ]	[ "Computer Science" ]	2026-05-26T00:00:00	https://arxiv.org/abs/2605.27494	https://arxiv.org/pdf/2605.27494v1	2605.27494	null	0	true	https://github.com/syedhumarahim/grounded-cache-router	null	0.65
20ac5783ef8efbcb98b419adfff1c1b46f91eec929ee0fdb4aaf4d9a1b66e2da	[ "arxiv", "semantic_scholar" ]	MobileMoE: Scaling On-Device Mixture of Experts	Mixture-of-Experts (MoE) has become the de facto architecture for hundred-billion-parameter language models, yet its advantages at sub-billion scales for on-device deployment remain largely unexplored. To close this gap, we present MobileMoE, a family of on-device MoE language models with sub-billion active parameters ...	[ "Yanbei Chen", "Hanxian Huang", "Ernie Chang", "Jacob Szwejbka", "Digant Desai", "Zechun Liu", "Vikas Chandra", "Raghuraman Krishnamoorthi" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-05-26T00:00:00	https://arxiv.org/abs/2605.27358	https://arxiv.org/pdf/2605.27358v1	2605.27358	null	0	true	null	null	0.65
41c56abf6d36eb412de9024993e72a1d4e23306a8231eedacdb1a438f5e6091c	[ "arxiv", "semantic_scholar" ]	Dense2MoE: Pushing the Pareto Frontier of On-Device LLMs via Unified Pruning and Upcycling	The Mixture of Experts MoE architecture is highly promising for resource constrained on device deployments yet training these models from scratch incurs prohibitive costs Current methods attempt to alleviate this by upcycling dense models into MoEs however they often introduce parameter redundancy that degrades inferen...	[ "Fengfa Li", "Hongjin Ji", "Yifeng Ding", "Lei Ren", "Chen Wei" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-26T00:00:00	https://arxiv.org/abs/2605.26496	https://arxiv.org/pdf/2605.26496v1	2605.26496	null	0	false	null	null	0.35
41b9164ffddb8ff8c255329a5f8f781d50bb682758f6835b0cd681f9a11e8099	[ "arxiv", "semantic_scholar" ]	Quantized Keys Steal Attention: Bias Correction for KV-Cache Compression in Video Diffusion	Chunk-wise autoregressive video diffusion models rely on a KV cache of previously generated chunks to avoid redundant computation, but this cache quickly becomes a memory bottleneck as videos grow longer. Methods that quantize the KV cache to low bitwidths reduce memory pressure but degrade video quality. We show that ...	[ "Tuna Tuncer", "Felix Becker", "Thomas Pfeil" ]	[ "cs.LG", "cs.AI", "cs.CV", "cs.GR", "eess.IV" ]	[ "Computer Science", "Engineering" ]	2026-05-25T00:00:00	https://arxiv.org/abs/2605.26266	https://arxiv.org/pdf/2605.26266v1	2605.26266	null	0	false	null	null	0.35
01766860289bf7462da1adc8c4096eca2fb38bf947e91d55fa8beee387e6a605	[ "arxiv", "semantic_scholar" ]	RotMoLE: Enhancing Mixture of Low-Rank Experts through Rotational Gating Mechanism	While Large Language Models (LLMs) are commonly fine-tuned to handle domain-specific tasks before being applied to vertical applications, adapting them to complex scenarios with diverse specialized knowledge remains challenging. Meanwhile, Mixture-of-Experts (MoE) architecture has risen as a crucial paradigm for traini...	[ "Mengyang Sun", "Maochuan Dou", "Tao Feng", "Dan Zhang", "Yihao Wang", "Junpeng Liu", "Yifan Zhu", "Jie Tang" ]	[ "cs.LG", "cs.CL" ]	[ "Computer Science" ]	2026-05-25T00:00:00	https://arxiv.org/abs/2605.25565	https://arxiv.org/pdf/2605.25565v1	2605.25565	null	0	false	null	null	0.35
c6ea193b8da4a99e252d5debcd134afef4993bb4cb1a1b43d6d649a25070cb88	[ "arxiv", "semantic_scholar" ]	IndexMem: Learned KV-Cache Eviction with Latent Memory for Long-Context LLM Inference	Large Language Models (LLMs) are increasingly expected to operate over long contexts, yet standard softmax attention incurs a KV cache that grows linearly with sequence length, quickly becoming the bottleneck for long context inference. A practical remedy is to evict less important KV entries; however, existing evictio...	[ "Xintong Yang", "Hao Gu", "Binxing Xu", "Lujun Li", "Bei Liu", "Jiacheng Liu", "Qiyuan Zhu", "Sirui Han", "Yike Guo" ]	[ "cs.CL", "cs.AI" ]	[ "Computer Science" ]	2026-05-25T00:00:00	https://arxiv.org/abs/2605.25475	https://arxiv.org/pdf/2605.25475v1	2605.25475	null	0	false	null	null	0.35
739f51731bbda29e4c51999da60649438c66bab2b35169378fb4ee0bfe790450	[ "arxiv", "semantic_scholar" ]	Energy-Gated Attention and Wavelet Positional Encoding: Complementary Inductive Biases for Transformer Attention	Standard transformer attention computes pairwise token similarity but treats all tokens as equally salient and all positions as equally local, regardless of the informational structure of the input. We identify two complementary inductive biases that standard attention lacks: energy salience (which tokens concentrate i...	[ "Athanasios Zeris" ]	[ "cs.LG", "cs.CL", "eess.SP" ]	[ "Computer Science", "Engineering" ]	2026-05-25T00:00:00	https://arxiv.org/abs/2605.26355	https://arxiv.org/pdf/2605.26355v1	2605.26355	null	2	true	https://github.com/AthanasiosZeris/energy-gated-attention	null	0.65
f41ad2dc1999c909f7bfa1a3a4865f72a7ec3c65630ae546a1d0f849e10891b9	[ "arxiv", "semantic_scholar" ]	Cross-Stage Attention Multi-Expert Network for Radiologist-Inspired Breast Ultrasound Diagnosis	Breast ultrasound imaging is an important noninvasive method for early breast cancer diagnosis, but automatic benign/malignant classification remains challenging due to tumor heterogeneity, blurred boundaries, and data imbalance. To improve feature representation and classification accuracy, this paper proposes the Cro...	[ "Xinyang Zhai", "Chong Yang", "Ruizhi Zhang" ]	[ "cs.CV", "cs.AI" ]	[ "Computer Science" ]	2026-05-25T00:00:00	https://arxiv.org/abs/2605.25518	https://arxiv.org/pdf/2605.25518v1	2605.25518	null	0	false	null	null	0.35
7f7d46f06b88ff67f5cbf59fa71d1ec67699ecd5d03d7d6ec773b5d6d4e3c17b	[ "arxiv", "semantic_scholar" ]	SP-MoMamba: Superpixel-driven Mixture of State Space Experts for Efficient Image Super-Resolution	State space models (SSMs) have emerged as a powerful paradigm for efficient single-image super-resolution (SR) due to their linear complexity and long-range modeling capabilities. However, existing Mamba-based methods typically rely on data-agnostic rigid scanning, which reshapes 2D images into 1D sequences over a fixe...	[ "Wenbin Zou", "Yawen Cui", "Yi Wang", "Lap-Pui Chau", "Liang Chen", "Jinshan Pan", "Huiping Zhuang", "Guanbin Li" ]	[ "cs.CV" ]	[ "Computer Science" ]	2026-05-25T00:00:00	https://arxiv.org/abs/2605.25892	https://arxiv.org/pdf/2605.25892v1	2605.25892	null	0	false	null	null	0.35
677939281c8d7a54a5d7ede97dcf98e88e9b19f9790230c0376118c6cbb2b7ce	[ "arxiv", "semantic_scholar" ]	CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM	Long-horizon LLM inference turns the key--value (KV) cache into the dominant GPU memory consumer and makes per-token attention increasingly expensive. Many common eviction policies use static recency windows or historical attention, leaving unused a signal computed on every decoding step: the model's current uncertaint...	[ "Yubo Li", "Yidi Miao" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-24T00:00:00	https://arxiv.org/abs/2605.24786	https://arxiv.org/pdf/2605.24786v1	2605.24786	null	0	false	null	null	0.35
15527e54056fd1526d557abcb1cb7e4bc87a7e7ac85ac015ad83083607eec68c	[ "arxiv", "semantic_scholar" ]	CachePrune: Privacy-Aware and Fine-Grained KV Cache Sharing for Efficient LLM Inference	Large Language Models (LLMs) rely on Key-Value (KV) caching to accelerate inference, and many serving systems further share the KV cache across users' requests to reduce redundant computation. While widely adopted, unrestricted cross-user sharing introduces side-channel vulnerabilities, allowing an adversary to infer u...	[ "Guanlong Wu", "Zhaohan li", "Yao Zhang", "Zheng Zhang", "Jianyu Niu", "Ye Wu", "Yinqian Zhang" ]	[ "cs.CR" ]	[ "Computer Science" ]	2026-05-22T00:00:00	https://arxiv.org/abs/2605.23640	https://arxiv.org/pdf/2605.23640v1	2605.23640	null	0	false	null	null	0.35
0ee3d65a28bbec50d7f02d6a0331e203a828b9d12dbc954728bd7462f7d85300	[ "arxiv", "semantic_scholar" ]	GMENet: Generative Mixture of Experts Network for Multi-Center Glioma Diagnosis with Incomplete Imaging Sequences	Contemporary glioma diagnosis integrates molecular features with histopathology to guide clinical decision-making. However, in clinical settings, divergent imaging protocols result in incomplete MRI sequences, leading to two primary challenges: forcing existing frameworks to discard a large portion of clinical data dur...	[ "Pengfei Song", "Fangjin Liu", "Wenwen Zeng", "Yonghuang Wu", "Chengqian Zhao", "Feiyu Yin", "Xuan Xie", "Jinhua Yu" ]	[ "eess.IV", "cs.CV" ]	[ "Engineering", "Computer Science" ]	2026-05-22T00:00:00	https://arxiv.org/abs/2605.23183	https://arxiv.org/pdf/2605.23183v1	2605.23183	null	0	false	null	null	0.35
eea15cc3dec66849c9ed3729c80fe6b6093bef69d68ca97e01866e546c54c9fd	[ "arxiv", "semantic_scholar" ]	A Simple Plug-in for Improving Eviction-Based KV Cache Compression	KV cache growth is a major bottleneck for long-context inference in large language models. Existing methods are often dominated by binary eviction or representation approximation, which may underutilize tokens that are not critical for exact retention but are still reconstructable. We present VECTOR, a plug-and-play au...	[ "Yuping Lin", "Jiayuan Ding", "Yue Xing", "Pengfei He", "Jiliang Tang", "Subhabrata Mukherjee" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-05-22T00:00:00	https://arxiv.org/abs/2605.23258	https://arxiv.org/pdf/2605.23258v1	2605.23258	null	0	false	null	null	0.35
4be72da8658a586d7a1cf706beab00e6c1c2393b733bab2842ca241619319f22	[ "arxiv", "semantic_scholar" ]	ECo-MoE: Embodiment-Conditioned Mixture of Experts Increases the Evolvability of Robots	In this paper, we introduce a model of evolution and learning in robots that co-optimizes a distribution of latent design vectors (genotypes) and a mixture of control experts (neural modules), which are gated by the latent coordinates of each decoded design (phenotype). This provides a scalable alternative to co-design...	[ "Yibin Wang", "Muhan Li", "Zihan Guo", "Sam Kriegman" ]	[ "cs.RO" ]	[ "Computer Science" ]	2026-05-22T00:00:00	https://arxiv.org/abs/2605.24225	https://arxiv.org/pdf/2605.24225v1	2605.24225	null	0	false	null	null	0.35
8ab9207fcd09a2932a8334a1592b81a1bf4ad91367a1c356ba0d690747088db9	[ "arxiv", "semantic_scholar" ]	Semantically Structured Mixture-of-Experts for Compositional Robotic Manipulation	Diffusion-based policies have established a new standard for precise robotic manipulation but face a critical scalability bottleneck: high-performance models are computationally expensive, while lightweight alternatives often fail to generalize across diverse multi-task environments. Mixture-of-Experts (MoE) architectu...	[ "Chengyu Deng", "Guanqi Chen", "Yizhou Chen", "Zejia Liu", "Zhiwen Ruan", "Guanhua Chen", "Jia Pan" ]	[ "cs.RO" ]	[ "Computer Science" ]	2026-05-22T00:00:00	https://arxiv.org/abs/2605.23477	https://arxiv.org/pdf/2605.23477v1	2605.23477	null	0	false	null	null	0.35
55784f072a01b76fd56a1e72f980d9421307899a17c96f5f03ea42f2cb48e6c6	[ "arxiv", "semantic_scholar" ]	Resident KV Claims: A Conformance Contract for Future Reuse under Active KV Pressure	KV-cache reuse mechanisms increasingly expose priority, duration, offload, routing hints, scheduler modes, and event streams. These mechanisms help preserve reusable prefixes, but they do not by themselves define a portable contract for accepted future-reuse state when resident KV and active live KV cannot both fit. We...	[ "Lukas Stepanek" ]	[ "cs.DC" ]	[ "Computer Science" ]	2026-05-22T00:00:00	https://arxiv.org/abs/2605.24259	https://arxiv.org/pdf/2605.24259v1	2605.24259	null	0	false	null	null	0.35
b5bed29f77996fba3a0a93d149771ab189355540fdd1b16122b3ddd7d8c6b96c	[ "arxiv", "semantic_scholar" ]	MuKV: Multi-Grained KV Cache Compression for Long Streaming Video Question-Answering	Long streaming video QA remains challenging due to growing visual tokens and limited reasoning length of large language models (LLMs). KV-caching stores the Key-Value (KV) of the historical tokens via LLM prefill and enables more efficient streaming QA. However, existing methods cache every one or two frames, causing r...	[ "Junbin Xiao", "Jiajun Chen", "Tianxiang Sun", "Xun Yang", "Angela Yao" ]	[ "cs.CV", "cs.AI", "cs.MM" ]	[ "Computer Science" ]	2026-05-21T00:00:00	https://arxiv.org/abs/2605.22269	https://arxiv.org/pdf/2605.22269v1	2605.22269	null	0	true	https://github.com/IMBALDY/MuKV	null	0.65
5c1fa6f5cd729e4de740c84a6528488c846d8af0e5aa39e6696bd1094bc2b8e6	[ "arxiv", "semantic_scholar" ]	GEMQ: Global Expert-Level Mixed-Precision Quantization for MoE LLMs	Mixture-of-Experts Large Language Models (MoE-LLMs) achieve strong performance but incur substantial memory overhead due to massive expert parameters. Mixed-precision quantization mitigates this cost by allocating expert-wise bit-widths based on their importance, approaching the accuracy-memory Pareto frontier and enab...	[ "Jianing Deng", "Song Wang", "Dongwei Wang", "Zijie Liu", "Tianlong Chen", "Huanrui Yang", "Jingtong Hu" ]	[ "cs.LG", "cs.CL" ]	[ "Computer Science" ]	2026-05-21T00:00:00	https://arxiv.org/abs/2605.23078	https://arxiv.org/pdf/2605.23078v1	2605.23078	null	0	true	https://github.com/jndeng/GEMQ	null	0.65
bc1ba87a17c6a785190da579061e491810076b1cc820e3cc275cf94bd16ab83c	[ "arxiv", "semantic_scholar" ]	Meta-Soft: Leveraging Composable Meta-Tokens for Context-Preserving KV Cache Compression	The KV cache used in large language models has linearly growing time complexity, so LLMs face memory blow-up and reduced decoding efficiency when they process long contexts. Current KV Cache eviction has become an important research direction; however, existing methods based on fixed Soft Tokens (e.g., Judge Q) rely on...	[ "Wei Luo", "Yi Huang", "Songchen Ma", "Huanyu Qu", "Jiang Cai", "Mingkun Xu" ]	[ "cs.AI" ]	[ "Computer Science" ]	2026-05-21T00:00:00	https://arxiv.org/abs/2605.22337	https://arxiv.org/pdf/2605.22337v2	2605.22337	null	0	false	null	null	0.35
993ff84b0b0ed3e9729055f674de7e3202b268c2c43d9d393d87e1172a4986b3	[ "arxiv", "semantic_scholar" ]	ASAP: Attention Sink Anchored Pruning	Vision Transformers (ViTs) face severe computational bottlenecks due to the quadratic complexity of self-attention at high resolutions. Existing token reduction methods rely on local metrics - such as single-layer attention scores - that are inherently vulnerable to the attention sink phenomenon, where uninformative to...	[ "Jaehyuk Lee", "Hanyoung Kim", "Yanggee Kim", "Donghun Lee" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-05-21T00:00:00	https://arxiv.org/abs/2605.22372	https://arxiv.org/pdf/2605.22372v1	2605.22372	null	0	false	null	null	0.35
f9867d713eb57acc60f2d7ae6bba814dff6763a2228dc5e0c950b93f2a0a4a06	[ "arxiv", "semantic_scholar" ]	ArborKV: Structure-Aware KV Cache Management for Scaling Tree-based LLM Reasoning	Recent progress in LLM reasoning has increasingly shifted from single-pass generation to explicit search over intermediate reasoning states. Tree-of-Thoughts (ToT) organizes inference to tree-structured search with branching and backtracking, but it substantially amplifies the Key--Value (KV) cache: retaining KV states...	[ "Yeqiu Chen", "Ziyan Liu", "Zhenxin Huang", "Runquan Gui", "Hong Wang", "Lei Liu" ]	[ "cs.AI" ]	[ "Computer Science" ]	2026-05-21T00:00:00	https://arxiv.org/abs/2605.22106	https://arxiv.org/pdf/2605.22106v1	2605.22106	null	1	false	null	null	0.35
13455e8850b9cbb8bd37f0a3470cc974f9c6e5653e514ab11a01a3d68dc50612	[ "arxiv", "semantic_scholar" ]	OCTOPUS: Optimized KV Cache for Transformers via Octahedral Parametrization Under optimal Squared error quantization	The key-value (KV) cache dominates memory bandwidth and footprint in long-context autoregressive inference. Recent rotation-preconditioned codecs (TurboQuant, PolarQuant) show that a structured random rotation followed by a per-coordinate scalar quantizer matched to an analytically tractable marginal is a near-optimal ...	[ "Mark Boss", "Vikram Voleti", "Simon Donné", "Shimon Vainer" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-20T00:00:00	https://arxiv.org/abs/2605.21226	https://arxiv.org/pdf/2605.21226v1	2605.21226	null	0	false	null	null	0.35
3a26771612b52e609a6d4d4a7d8d2c6aa59d7afeb2a2e71d55f7966fcfcf8dbe	[ "arxiv", "semantic_scholar" ]	Adaptive KV Cache Reuse for Fast Long-Context LLM Serving	In long-context Large Language Model (LLM) inference, the Time-To-First-Token (TTFT) latency incurred by the prefill stage has become the foremost bottleneck limiting interactive performance and deployment cost. KV Cache reuse offers a direct path to reduce redundant prefill, yet traditional prefix caching applies only...	[ "Fei li", "Song Liu", "Yan Liu", "Jinhua Cui", "Shiqiang Nie", "Jinyu Wang", "Weiguo Wu" ]	[ "cs.AR", "cs.DC" ]	[ "Computer Science" ]	2026-05-20T00:00:00	https://arxiv.org/abs/2605.24022	https://arxiv.org/pdf/2605.24022v1	2605.24022	null	0	false	null	null	0.35
7d9ba1716a760bed2899e2bf180fc5feedc8040c5da3f0965cbefb44a88cd3ef	[ "arxiv", "semantic_scholar" ]	Runtime-Certified Bounded-Error Quantized Attention	KV cache quantization reduces the memory cost of long-context LLM inference, but introduces approximation error that is typically validated only empirically. Existing systems rely on average-case robustness, with no mechanism to detect or recover from failures at runtime. We present a tiered KV cache architecture that ...	[ "Dean Calver" ]	[ "cs.LG", "cs.AI", "eess.SY" ]	[ "Computer Science", "Engineering" ]	2026-05-20T00:00:00	https://arxiv.org/abs/2605.20868	https://arxiv.org/pdf/2605.20868v1	2605.20868	null	0	false	null	null	0.35
5097a4413db04ebf41c66e7a11c7b2c24f0cdb425f5bd8a0a45d1f335c872aa8	[ "arxiv", "semantic_scholar" ]	NanoCP: Request-Level Dynamic Context Parallelism for Data-Expert Parallel Decoding	Modern serving systems for Mixture-of-Experts (MoE) models adopt hybrid data-expert parallelism: expert parallelism (EP) shards experts across GPUs to scale capacity, while data parallelism (DP) replicates attention layers across instances to process independent requests. Existing systems bind each request's attention,...	[ "Jiefei Chen", "Binbin Lin", "Jinming Ma", "Jiangfei Duan", "Haojie Duanmu", "Hao Liu", "Qinxiu Cheng", "Xiuhong Li", "Zhilin Pei", "Hui Wang", "Xingcheng Zhang", "Dahua Lin" ]	[ "cs.DC" ]	[ "Computer Science" ]	2026-05-20T00:00:00	https://arxiv.org/abs/2605.21100	https://arxiv.org/pdf/2605.21100v1	2605.21100	null	0	false	null	null	0.35
45851014eee01e7aaf26c7a5b5b3da4a2c89178f3f9461434be4251569b4ae22	[ "arxiv", "semantic_scholar" ]	Dynamic TMoE: A Drift-Aware Dynamic Mixture of Experts Framework for Non-Stationary Time Series Forecasting	Non-stationary time series forecasting is challenged by evolving distribution shifts that static models struggle to capture. While Mixture-of-Experts (MoE) architectures offer a promising paradigm for decoupling complex drift patterns, existing approaches are limited by fixed expert pools and memoryless routing, hamper...	[ "Jiawen Zhu", "Shuhan Liu", "Di Weng", "Yingcai Wu" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-20T00:00:00	https://arxiv.org/abs/2605.20678	https://arxiv.org/pdf/2605.20678v1	2605.20678	null	0	true	https://github.com/andone-07/Dynamic-TMoE	null	0.65
4e5862621c61c300f9ab95b4f8fa94ef1ce14225651c8649ef5568fb11cbfc53	[ "arxiv", "semantic_scholar" ]	Beyond Routing: Characterising Expert Tuning and Representation in Vision Mixture-of-Experts	Mixture-of-Experts (MoE) models are often interpreted by analysing which categories are routed to which experts. However, routing alone does not reveal what each expert actually encodes. We train sparsely-gated convolutional MoE models with a contrastive objective on natural images and characterise expert specialisatio...	[ "Gene Tangtartharakul", "Katherine R. Storrs" ]	[ "cs.CV", "cs.AI" ]	[ "Computer Science" ]	2026-05-20T00:00:00	https://arxiv.org/abs/2605.20610	https://arxiv.org/pdf/2605.20610v1	2605.20610	null	0	false	null	null	0.35
c08824c4c5c184721917a400dc5a9eb4e992dd4df7274f6748aad169dd56983a	[ "arxiv", "semantic_scholar" ]	Task-Routed Mixture-of-Experts with Cognitive Appraisal for Implicit Sentiment Analysis	Implicit sentiment analysis is challenging because sentiment toward an aspect is often inferred from events rather than expressed through explicit opinion words. Existing models typically learn from the final polarity label, which provides limited guidance for reasoning about sentiment from the context. Motivated by co...	[ "Yaping Chai", "Haoran Xie", "Joe S. Qin" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-05-20T00:00:00	https://arxiv.org/abs/2605.20916	https://arxiv.org/pdf/2605.20916v1	2605.20916	null	0	true	https://github.com/yaping166/TRMoE-ISA	null	0.65
0fab166949be451c217f839504866a72e3538be8fca90d34a2b6ad7907d92cc1	[ "arxiv", "semantic_scholar" ]	OScaR: The Occam's Razor for Extreme KV Cache Quantization in LLMs and Beyond	The rapid advancement toward long-context reasoning and multi-modal intelligence has made the memory footprint of the Key-Value (KV) cache a dominant memory bottleneck for efficient deployment. While the established per-channel quantization effectively accommodates intrinsic channel-wise outliers in Key tensors, its ef...	[ "Zunhai Su", "Rui Yang", "Chao Zhang", "Yaxiu Liu", "Yifan Zhang", "Wei Wu", "Jing Xiong", "Dayou Du", "Xialie Zhuang", "Yulei Qian", "Yuchen Xie", "Yik-Chung Wu", "Hongxia Yang", "Ngai Wong" ]	[ "cs.LG", "cs.CL" ]	[ "Computer Science" ]	2026-05-19T00:00:00	https://arxiv.org/abs/2605.19660	https://arxiv.org/pdf/2605.19660v1	2605.19660	null	0	true	https://github.com/ZunhaiSu/OScaR-KV-Quant	null	0.65

End of preview. Expand in Data Studio

Efficient LLM Papers — FineSet

A research-paper dataset on Efficient LLM Papers, assembled, deduplicated, and quality-scored by FineSet from arXiv and Semantic Scholar.

📸 This is a dated snapshot — generated 2026-06-12. It is not auto-updated. Research on Efficient LLM Papers moves fast — new papers land on arXiv every week. Want this same dataset refreshed daily, on a topic you choose? See the bottom. ↓

Why this dataset

Quality-scored: quality_score float (0–1), blends citations with recency + code/venue signals — filter out the noise
Papers with code: 435 flagged via has_code — find reproducible work fast
Deduplicated: arXiv + Semantic Scholar cross-referenced, duplicate records merged
Clean JSONL: 1734 records, one per line, normalized fields — no encoding garbage

Dataset details

Records: 1734
Date range: 2022–2026
Snapshot date: 2026-06-12 (frozen — see note above)
Sources: arXiv, Semantic Scholar (cross-referenced, duplicates merged)
arXiv categories: cs.LG
Quality scoring: citations + recency + code/venue blend, 0–1 (p50=0.34, p90=0.55)
Format: JSONL, one record per line

Fields

Field	Type	Description
id	string	Deterministic SHA256 record id
sources	list	Which sources contributed (`arxiv`, `semantic_scholar`)
title	string	Paper title
abstract	string	Full abstract
authors	list	Author names
categories	list	arXiv category codes
fields_of_study	list	Semantic Scholar field tags
published_date	string	ISO 8601 date
url	string	arXiv abstract URL
pdf_url	string\|null	Open-access PDF if available
arxiv_id	string\|null	arXiv identifier
doi	string\|null	DOI if available
citation_count	int	Citation count (Semantic Scholar)
influential_citation_count	int	Influential citations (Semantic Scholar)
has_code	bool	Code repo detected in the arXiv comment
code_url	string\|null	GitHub URL if detected
venue	string\|null	Publication venue
quality_score	float	0–1, blended (citations + recency + code/venue)

Quality score methodology

quality_score = max(impact, freshness), clamped to [0, 1], where:

impact = max( log10(citations+1)/4 , log10(influential_citations+1)/2 ) — realized impact (0.5 at 100 citations, ~0.75 at 1,000, 1.0 at 10,000+).
freshness = recency × (0.35 + 0.30·has_code + 0.20·has_venue) — a baseline for recent papers (so a strong paper published this week isn't scored 0 just for lacking citations), where recency is 1.0 for papers ≤60 days old and decays linearly to 0 by ~18 months.

Old highly-cited papers score on impact; brand-new papers score on freshness; old uncited papers score ~0. Useful for filtering training data by quality, not just age.

👉 Want this on YOUR topic, updated daily?

This snapshot is frozen at 2026-06-12. The live FineSet pipeline keeps a dataset like this refreshed every day on whatever topic you describe — new papers in, dedup and quality scoring automatic, export as JSONL/Parquet or push straight to the Hub.

Tell me the topic you'd want and I'll run the pipeline on it — open a discussion on this dataset, it's free and it's how I decide what to build next.

→ fineset.io — describe what you want to train on, get a dataset. Early-access waitlist open (referral skip available).

Downloads last month: 13