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
influential_citation_count
int64
has_code
bool
code_url
string
venue
string
quality_score
float64
e3309643bc3fcd71d8e4043679f3ad0cc54123a6bb8495797bb20bdbb6ad3099
[ "arxiv", "semantic_scholar" ]
Soft Label Pruning and Quantization for Large-Scale Dataset Distillation
Large-scale dataset distillation requires storing auxiliary soft labels that can be 30-40x larger on ImageNet-1K and 200x larger on ImageNet-21K than the condensed images, undermining the goal of dataset compression. We identify two fundamental issues necessitating such extensive labels: (1) insufficient image diversit...
[ "Xiao Lingao", "Yang He" ]
[ "cs.CV", "cs.AI", "cs.LG" ]
[ "Medicine", "Computer Science" ]
2026-04-20T00:00:00
https://arxiv.org/abs/2604.18135
https://arxiv.org/pdf/2604.18135v1
2604.18135
10.1109/TPAMI.2026.3664488
1
0
true
https://github.com/he-y/soft-label-pruning-quantization-for-dataset-distillation
IEEE Transactions on Pattern Analysis and Machine Intelligence
0.85
7b4a7be9493a7453dc6a9ceefef97e09a7b065ae832c493c599f7400da41c581
[ "arxiv", "semantic_scholar" ]
Six Llamas: Comparative Religious Ethics Through LoRA-Adapted Language Models
We present Six Llamas, a comparative study examining whether large language models fine-tuned on distinct religious corpora encode systematically different patterns of ethical reasoning. Six variants of Meta-Llama-3.1-8B are constructed: one unmodified control and five LoRA-adapted models trained exclusively on the sac...
[ "Chad Coleman", "W. Russell Neuman", "Manan Shah", "Ali Dasdan", "Matthew Crispi", "Morris Chiang", "Zack Leitman", "Mustafa Poonawala" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-04-20T00:00:00
https://arxiv.org/abs/2604.18404
https://arxiv.org/pdf/2604.18404v1
2604.18404
null
0
0
false
null
null
0.35
535f207b12e3c5db6d356d572ff15bcfdce5c0c227949ebf7a292ee113dd989e
[ "arxiv", "semantic_scholar" ]
Domain-Specialized Object Detection via Model-Level Mixtures of Experts
Mixture-of-Experts (MoE) models provide a structured approach to combining specialized neural networks and offer greater interpretability than conventional ensembles. While MoEs have been successfully applied to image classification and semantic segmentation, their use in object detection remains limited due to challen...
[ "Svetlana Pavlitska", "Malte Stüven", "Beyza Keskin", "J. Marius Zöllner" ]
[ "cs.CV", "cs.LG" ]
[ "Computer Science" ]
2026-04-20T00:00:00
https://arxiv.org/abs/2604.18256
https://arxiv.org/pdf/2604.18256v1
2604.18256
null
0
0
true
https://github.com/KASTEL-MobilityLab/mixtures-of-experts/
null
0.65
158dc6ca943fb8437ce10b495f2a7f99848c5c18a831e6c4e8485e87ed02b799
[ "arxiv", "semantic_scholar" ]
Towards Joint Quantization and Token Pruning of Vision-Language Models
Deploying Vision-Language Models (VLMs) under aggressive low-bit inference remains challenging because inference cost is dominated by the long visual-token prefix during prefill and the growing KV cache during autoregressive decoding. Token pruning and low-bit quantization are complementary for reducing these costs, ye...
[ "Xinqing Li", "Xin He", "Xindong Zhang", "Ming-Ming Cheng", "Lei Zhang", "Yun Liu" ]
[ "cs.CV" ]
[ "Computer Science" ]
2026-04-19T00:00:00
https://arxiv.org/abs/2604.17320
https://arxiv.org/pdf/2604.17320v1
2604.17320
null
0
0
false
null
null
0.35
2cc97ea717fd897bfe3ce60c9770134cdb2c18324a5a46980264d32b03298319
[ "arxiv", "semantic_scholar" ]
Predictive Multi-Tier Memory Management for KV Cache in Large-Scale GPU Inference
Key-value (KV) cache memory management is the primary bottleneck limiting throughput and cost-efficiency in large-scale GPU inference serving. Current systems suffer from three compounding inefficiencies: (1) the absence of unified KV cache sizing across all attention architectures--particularly multi-head latent atten...
[ "Sanjeev Rao Ganjihal" ]
[ "cs.AR", "cs.AI", "cs.DC", "cs.PF" ]
[ "Computer Science" ]
2026-04-19T00:00:00
https://arxiv.org/abs/2604.26968
https://arxiv.org/pdf/2604.26968v1
2604.26968
null
0
0
false
null
null
0.35
3d1f5d001658da4a52772636eb74e595d11eddff1c7b5d585dc5ca3684b2f43b
[ "arxiv", "semantic_scholar" ]
Bit-Flip Vulnerability of Shared KV-Cache Blocks in LLM Serving Systems
Rowhammer on GPU DRAM has enabled adversarial bit flips in model weights; shared KV-cache blocks in LLM serving systems present an analogous but previously unexamined target. In vLLM's Prefix Caching, these blocks exist as a single physical copy without integrity protection. Using software fault injection under ideal b...
[ "Yuji Yamamoto", "Satoshi Matsuura" ]
[ "cs.CR", "cs.AR", "cs.LG" ]
[ "Computer Science" ]
2026-04-19T00:00:00
https://arxiv.org/abs/2604.17249
https://arxiv.org/pdf/2604.17249v2
2604.17249
null
0
0
false
null
null
0.35
859db27e02a0753772942b853209312b85e46debae8cbaa6f9b1f48bb63c93e4
[ "arxiv", "semantic_scholar" ]
MoVE: Translating Laughter and Tears via Mixture of Vocalization Experts in Speech-to-Speech Translation
Recent Speech-to-Speech Translation (S2ST) systems achieve strong semantic accuracy yet consistently strip away non-verbal vocalizations (NVs), such as laughter and crying that convey pragmatic intent, which severely limits real-world utility. We address this via three contributions. First, we propose a synthesis pipel...
[ "Szu-Chi Chen", "I-Ning Tsai", "Yi-Cheng Lin", "Sung-Feng Huang", "Hung-yi Lee" ]
[ "cs.CL", "cs.AI", "cs.SD", "eess.AS" ]
[ "Computer Science", "Engineering" ]
2026-04-19T00:00:00
https://arxiv.org/abs/2604.17435
https://arxiv.org/pdf/2604.17435v1
2604.17435
null
1
0
false
null
null
0.35
a0ea47cf6f200cf6028fc80134c311b7f04bf507c4c0c0b909d205a05470b913
[ "arxiv", "semantic_scholar" ]
HieraSparse: Hierarchical Semi-Structured Sparse KV Attention
The deployment of long-context Large Language Models (LLMs) poses significant challenges due to the intense computational cost of self-attention and the substantial memory overhead of the Key-Value Cache (KV Cache). In this paper, we introduce HieraSparse, a hierarchical KV Cache compression framework with acceleration...
[ "Haoxuan Wang", "Chen Wang" ]
[ "cs.DC", "cs.AR" ]
[ "Computer Science" ]
2026-04-18T00:00:00
https://arxiv.org/abs/2604.16864
https://arxiv.org/pdf/2604.16864v1
2604.16864
null
0
0
true
https://github.com/psl-ntu/HieraSparse
null
0.65
202cadd1a6aebb28f9d7d67988c10bb22a2d8aae4d312d8bdf452787c892ebd2
[ "arxiv", "semantic_scholar" ]
Graph-Guided Adaptive Channel Elimination for KV Cache Compression
Large Language Models have revolutionized natural language processing, achieving unprecedented success across a vast range of tasks. However, their practical application in long-context scenarios is severely hampered by the formidable memory footprint of the Key-Value cache. While channel pruning has emerged as a promi...
[ "Enwei Tong", "Yao Zhu", "Yuanchao Bai", "Kai Wang", "Xianming Liu", "Xiangyang Ji" ]
[ "eess.SP" ]
[ "Engineering" ]
2026-04-18T00:00:00
https://arxiv.org/abs/2604.16983
https://arxiv.org/pdf/2604.16983v1
2604.16983
null
0
0
false
null
null
0.35
e5695158bda68c82d5fcd87f732cca08949abdfb6cbab718ec443e10592475e0
[ "arxiv", "semantic_scholar" ]
The Illusion of Equivalence: Systematic FP16 Divergence in KV-Cached Autoregressive Inference
KV caching is a ubiquitous optimization in autoregressive transformer inference, long presumed to be numerically equivalent to cache-free computation. This assumption fails under standard FP16 precision: cache-ON and cache-OFF execution paths employ different floating-point accumulation orderings which, due to FP16 non...
[ "Ranjith Chodavarapu", "Lei Xu" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-16T00:00:00
https://arxiv.org/abs/2604.15409
https://arxiv.org/pdf/2604.15409v1
2604.15409
null
0
0
false
null
null
0.3493
a77969075b58e608a7413c17a31718e8aa89e32d39277b2f865d4061d97bfa06
[ "arxiv", "semantic_scholar" ]
Dispatch-Aware Ragged Attention for Pruned Vision Transformers
Token pruning methods for Vision Transformers (ViTs) promise quadratic reductions in attention FLOPs by dropping uninformative patches. Yet standard variable-length attention APIs -- including FlashAttention-2's varlen and PyTorch's NestedTensor SDPA -- fail to translate these savings into proportional wall-clock gains...
[ "Seifeldin Abdellatif", "Ahmad Almasri" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-16T00:00:00
https://arxiv.org/abs/2604.15408
https://arxiv.org/pdf/2604.15408v2
2604.15408
null
0
0
false
null
null
0.3493
c12ba085126bb19f4fab5642a6d2bede59b56436b172294c944991f85470b666
[ "arxiv", "semantic_scholar" ]
Geometric Routing Enables Causal Expert Control in Mixture of Experts
Sparse Mixture-of-Experts (MoE) models scale parameters while fixing active computation per token, but the specialization of individual experts remains opaque. In a companion paper we showed that routing topology is quality-neutral: five structurally different configurations converge to statistically equivalent languag...
[ "Ivan Ternovtsii", "Yurii Bilak" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-04-15T00:00:00
https://arxiv.org/abs/2604.14434
https://arxiv.org/pdf/2604.14434v1
2604.14434
null
2
0
false
null
null
0.3485
188d979b73119021af4835fed38b86e7ea88d7c79ad04b41645cee6c7771f55d
[ "arxiv", "semantic_scholar" ]
ART: Attention Run-time Termination for Efficient Large Language Model Decoding
Long-context decoding in Large Language Models (LLMs) is constrained by the cost of accessing and processing the Key-Value (KV) cache. Despite the evidence that attention outputs depend jointly on keys and values, most existing KV management methods rely on key-only pruning, as incorporating values incurs prohibitive a...
[ "Chen Qiu", "Guozhong Li", "Cristian McGee", "Aritra Dutta", "Panos Kalnis" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-04-15T00:00:00
https://arxiv.org/abs/2606.00024
https://arxiv.org/pdf/2606.00024v2
2606.00024
null
0
0
false
null
null
0.3485
e8074ce7501acb1df0cdf17d52da8641dcfe2259cfef9e8edae5b0cc682aff93
[ "arxiv", "semantic_scholar" ]
KV Packet: Recomputation-Free Context-Independent KV Caching for LLMs
Large Language Models (LLMs) rely heavily on Key-Value (KV) caching to minimize inference latency. However, standard KV caches are context-dependent: reusing a cached document in a new context requires recomputing KV states to account for shifts in attention distribution. Existing solutions such as CacheBlend, EPIC, an...
[ "Chuangtao Chen", "Grace Li Zhang", "Xunzhao Yin", "Cheng Zhuo", "Bing Li", "Ulf Schlichtmann" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-14T00:00:00
https://arxiv.org/abs/2604.13226
https://arxiv.org/pdf/2604.13226v2
2604.13226
null
1
0
false
null
null
0.3478
32da91b91c77332e072107884e57c07d2e5adf9e50641274ad5aab9d3a9c68a1
[ "arxiv", "semantic_scholar" ]
RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction
Multimodal Large Language Models face severe challenges in computational efficiency and memory consumption due to the substantial expansion of the visual KV cache when processing long visual contexts. Existing KV cache compression methods typically rely on the "persistence of importance" hypothesis to prune tokens. How...
[ "Sihao Liu", "YuFan Xiong", "Zhonghua Jiang", "Zhaode Wang", "chengfei lv Shengyu Zhang" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-04-14T00:00:00
https://arxiv.org/abs/2605.04075
https://arxiv.org/pdf/2605.04075v1
2605.04075
null
0
0
false
null
null
0.3478
250d618f444b25194b1d614c4452b160d50a97b36f468ad705875b94234b2399
[ "arxiv", "semantic_scholar" ]
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning
We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemotron 3 Super is the first model in the Nemotron 3 family to 1) be pre-trained in NVFP4, 2) leverage LatentMoE, a new Mixture-of-Experts arch...
[ " NVIDIA", " :", "Aakshita Chandiramani", "Aaron Blakeman", "Abdullahi Olaoye", "Abhibha Gupta", "Abhilash Somasamudramath", "Abhinav Khattar", "Adeola Adesoba", "Adi Renduchintala", "Adil Asif", "Aditya Agrawal", "Aditya Vavre", "Ahmad Kiswani", "Aishwarya Padmakumar", "Ajay Hotchanda...
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-04-14T00:00:00
https://arxiv.org/abs/2604.12374
https://arxiv.org/pdf/2604.12374v1
2604.12374
null
9
1
true
null
null
0.6459
9ea6c3415ec79ba1f9a593eb79f8408b74ca0fac1d802159bc8ed7d248305991
[ "arxiv", "semantic_scholar" ]
Efficient Handwriting-Based Alzheimer,s Disease Diagnosis Using a Low-Rank Mixture of Experts Deep Learning Framework
Early and reliable detection of Alzheimer's disease (AD) is crucial for timely clinical intervention and improved patient management. It also supports the evaluation of emerging therapeutic strategies. In this paper, we propose a Low-Rank Mixture of Experts (LoRA-MoE) deep learning framework for Alzheimer's disease dia...
[ "Wu Wang", "Yuang Cheng", "Fouzi Harrou", "Ying Sun" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-14T00:00:00
https://arxiv.org/abs/2605.04079
https://arxiv.org/pdf/2605.04079v1
2605.04079
null
0
0
false
null
null
0.3478
20733de7296219368726d254373eb2d05e325e8c9d51c44d639a0de467111885
[ "arxiv", "semantic_scholar" ]
VFA: Relieving Vector Operations in Flash Attention with Global Maximum Pre-computation
FlashAttention-style online softmax enables exact attention computation with linear memory by streaming score tiles through on-chip memory and maintaining a running maximum and normalizer. However, as attention kernels approach peak tensor-core/cube-core throughput on modern accelerators, non-matmul components of onlin...
[ "Yupeng Sun", "Yanzhao Li", "Zhiqiang Zou", "Bai Du", "Zhiyuan Zhang", "Hui Dong", "Gaoyige Fan", "Hui Wang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-14T00:00:00
https://arxiv.org/abs/2604.12798
https://arxiv.org/pdf/2604.12798v1
2604.12798
null
0
0
false
null
null
0.3478
db441711f721eaf21d7ec517f2356278827128ba7c329def52275f52ac36afa8
[ "arxiv", "semantic_scholar" ]
Quantization Dominates Rank Reduction for KV-Cache Compression
We compare two strategies for compressing the KV cache in transformer inference: rank reduction (discard dimensions) and quantization (keep all dimensions, reduce precision). At matched storage budgets across five models (124M-14B, MHA and GQA), we find that quantization consistently outperforms rank reduction by 4-364...
[ "Samuel Salfati" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-04-13T00:00:00
https://arxiv.org/abs/2604.11501
https://arxiv.org/pdf/2604.11501v1
2604.11501
null
1
0
false
null
null
0.3471
328931df37dae3043ade81bda07eefbc99e26b02f26672b633e769dfcf46ce90
[ "arxiv", "semantic_scholar" ]
Transactional Attention: Semantic Sponsorship for KV-Cache Retention
At K=16 tokens (0.4% of a 4K context), every existing KV-cache compression method achieves 0% on credential retrieval. The failure mode is dormant tokens: credentials, API keys, and configuration values that receive near-zero attention but become essential at generation time. Because these tokens lack the statistical s...
[ "Abhinaba Basu" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2026-04-13T00:00:00
https://arxiv.org/abs/2604.11288
https://arxiv.org/pdf/2604.11288v1
2604.11288
null
0
0
false
null
null
0.3471
7f2ad71fb3e250b2d09ea81831e0360fb49920b09e9c8976bdc9feefcf20fd07
[ "arxiv", "semantic_scholar" ]
Beyond Attention Scores: SVD-Based Vision Token Pruning for Efficient Vision-Language Models
Vision-Language Models (VLMs) have revolutionized multi-modal learning by jointly processing visual and textual information. Yet, they face significant challenges due to the high computational and memory demands of processing long sequences of vision tokens. Many existing methods rely on local heuristics, such as atten...
[ "Yvon Apedo", "Martyna Poreba", "Michal Szczepanski", "Samia Bouchafa" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2026-04-13T00:00:00
https://arxiv.org/abs/2604.11530
https://arxiv.org/pdf/2604.11530v2
2604.11530
null
0
0
false
null
null
0.3471
d7f90706e9b6252fa790b9581d101e0256907f82016c713753de1448d20aac6d
[ "arxiv", "semantic_scholar" ]
IceCache: Memory-efficient KV-cache Management for Long-Sequence LLMs
Key-Value (KV) cache plays a crucial role in accelerating inference in large language models (LLMs) by storing intermediate attention states and avoiding redundant computation during autoregressive generation. However, its memory footprint scales linearly with sequence length, often leading to severe memory bottlenecks...
[ "Yuzhen Mao", "Qitong Wang", "Martin Ester", "Ke Li" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-12T00:00:00
https://arxiv.org/abs/2604.10539
https://arxiv.org/pdf/2604.10539v1
2604.10539
null
2
0
true
null
null
0.6432
9dc1d3bb0c7037cb3b58fcfd79bfa815247b2d780592aaffe16030d6cc3563b4
[ "arxiv", "semantic_scholar" ]
CodeQuant: Unified Clustering and Quantization for Enhanced Outlier Smoothing in Low-Precision Mixture-of-Experts
Outliers have emerged as a fundamental bottleneck in preserving accuracy for low-precision large models, particularly within Mixture-of-Experts (MoE) architectures that are increasingly central to large-scale language modeling. Under post-training quantization (PTQ), these outliers induce substantial quantization error...
[ "Xiangyang Yin", "Xingyu Liu", "Tianhua Xia", "Bo Bao", "Vithursan Thangarasa", "Valavan Manohararajah", "Eric Sather", "Sai Qian Zhang" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-04-12T00:00:00
https://arxiv.org/abs/2604.10496
https://arxiv.org/pdf/2604.10496v1
2604.10496
null
1
0
true
https://github.com/SAI-Lab-NYU/CodeQuant
null
0.6432
d91605ca1bb23809af027978644388951ac497adf326fc6b78b32796ce6684b0
[ "arxiv", "semantic_scholar" ]
LoopGuard: Breaking Self-Reinforcing Attention Loops via Dynamic KV Cache Intervention
Through systematic experiments on long-context generation, we observe a damaging failure mode in which decoding can collapse into persistent repetition loops. We find that this degeneration is driven by collapsed attention patterns, where a subset of heads locks onto a narrow suffix of the history, and is further stabi...
[ "Dongjie Xu", "Hao Wu", "Weijie Shi", "Yue Cui", "Yuanjun Liu", "Jiawei Li", "Haolun Ma", "An Liu", "Jia Zhu", "Jiajie Xu" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-04-11T00:00:00
https://arxiv.org/abs/2604.10044
https://arxiv.org/pdf/2604.10044v1
2604.10044
null
1
1
false
null
null
0.3456
cf5bfcfeac6d374a606385eac18eedd3c7f9aa80bb3f602ca266d9d73a9c4ae5
[ "arxiv", "semantic_scholar" ]
CodeComp: Structural KV Cache Compression for Agentic Coding
Agentic code tasks such as fault localization and patch generation require processing long codebases under tight memory constraints, where the Key-Value (KV) cache becomes the primary inference bottleneck. Existing compression methods rely exclusively on attention signals to estimate token importance, systematically di...
[ "Qiujiang Chen", "Jing Xiong", "Chenyang Zhao", "Sidi Yang", "Ngai Wong" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-04-11T00:00:00
https://arxiv.org/abs/2604.10235
https://arxiv.org/pdf/2604.10235v1
2604.10235
null
0
0
false
null
null
0.3456
7e126645626e22d06233927d125b67fcddc74d9ff5e1c5489616c789ded2c114
[ "arxiv", "semantic_scholar" ]
MoRI: Mixture of RL and IL Experts for Long-Horizon Manipulation Tasks
Reinforcement Learning (RL) and Imitation Learning (IL) are the standard frameworks for policy acquisition in manipulation. While IL offers efficient policy derivation, it suffers from compounding errors and distribution shift. Conversely, RL facilitates autonomous exploration but is frequently hindered by low sample e...
[ "Yaohang Xu", "Lianjie Ma", "Gewei Zuo", "Wentao Zhang", "Han Ding", "Lijun Zhu" ]
[ "cs.RO" ]
[ "Computer Science" ]
2026-04-11T00:00:00
https://arxiv.org/abs/2604.10165
https://arxiv.org/pdf/2604.10165v1
2604.10165
null
0
0
false
null
null
0.3456
70382e8338b39bceeb4a88e95f5dc07733d26ad2fe40e1f033e2b6919a39c82f
[ "arxiv", "semantic_scholar" ]
Sequential KV Cache Compression via Probabilistic Language Tries: Beyond the Per-Vector Shannon Limit
Recent work on KV cache quantization, culminating in TurboQuant, has approached the Shannon entropy limit for per-vector compression of transformer key-value caches. We observe that this limit applies to a strictly weaker problem than the one that actually matters: compressing the KV cache as a sequence. The tokens sto...
[ "Gregory Magarshak" ]
[ "cs.LG", "cs.AI", "cs.IT", "cs.NE" ]
[ "Computer Science", "Mathematics" ]
2026-04-10T00:00:00
https://arxiv.org/abs/2604.15356
https://arxiv.org/pdf/2604.15356v1
2604.15356
null
1
0
false
null
null
0.3449
62f0fbad79b75697b733e3d1e5e929869892cee2399efea8e9411d62e7d0b361
[ "arxiv", "semantic_scholar" ]
MP-ISMoE: Mixed-Precision Interactive Side Mixture-of-Experts for Efficient Transfer Learning
Parameter-efficient transfer learning (PETL) has emerged as a pivotal paradigm for adapting pre-trained foundation models to downstream tasks, significantly reducing trainable parameters yet suffering from substantial memory overhead caused by gradient backpropagation during fine-tuning. While memory-efficient transfer...
[ "Yutong Zhang", "Zimeng Wu", "Shangcai Liao", "Shujiang Wu", "Jiaxin Chen" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-10T00:00:00
https://arxiv.org/abs/2605.04058
https://arxiv.org/pdf/2605.04058v1
2605.04058
10.1609/aaai.v40i34.40084
2
0
false
null
AAAI Conference on Artificial Intelligence
0.542
192e1da6ca291ee2ed760d051906dc9c663fb8b4db9c7955a5c9df1fab3bbc4f
[ "arxiv", "semantic_scholar" ]
Modality-Aware Zero-Shot Pruning and Sparse Attention for Efficient Multimodal Edge Inference
Edge devices increasingly run multimodal sensing pipelines that must remain accurate despite fluctuating power budgets and unpredictable sensor dropout. Existing pruning methods fail under these conditions: they generally require fine-tuning after compression, consuming over $10\times$ the deployment energy, and they a...
[ "Yueyuan Sui", "Payal Mohapatra", "Doğaç Eldenk", "Haodong Yang", "Yiting Zhang", "Haoyan Zhang", "Qi Zhu", "Stephen Xia" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-04-10T00:00:00
https://arxiv.org/abs/2604.08971
https://arxiv.org/pdf/2604.08971v1
2604.08971
null
0
0
false
null
null
0.3449
9f3b30754b59503a5fb09b8f44edb9d4e81a1d520ab2e4cbe653017dbdd1c6f7
[ "arxiv", "semantic_scholar" ]
Plasticity-Enhanced Multi-Agent Mixture of Experts for Dynamic Objective Adaptation in UAVs-Assisted Emergency Communication Networks
Unmanned aerial vehicles serving as aerial base stations can rapidly restore connectivity after disasters, yet abrupt changes in user mobility and traffic demands shift the quality of service trade-offs and induce strong non-stationarity. Deep reinforcement learning policies suffer from plasticity loss under such shift...
[ "Wen Qiu", "Zhiqiang He", "Wei Zhao", "Hiroshi Masui" ]
[ "cs.MA", "cs.LG", "cs.NI" ]
[ "Computer Science" ]
2026-04-10T00:00:00
https://arxiv.org/abs/2604.09028
https://arxiv.org/pdf/2604.09028v1
2604.09028
10.1109/jiot.2026.3696834
0
0
false
null
IEEE Internet of Things Journal
0.542
3c29bc3970a7e4bce1ef512ff5c627ee2526ee0c79a7f5d3ccea0eca8d89907f
[ "arxiv", "semantic_scholar" ]
KV Cache Offloading for Context-Intensive Tasks
With the growing demand for long-context LLMs across a wide range of applications, the key-value (KV) cache has become a critical bottleneck for both latency and memory usage. Recently, KV-cache offloading has emerged as a promising approach to reduce memory footprint and inference latency while preserving accuracy. Pr...
[ "Andrey Bocharnikov", "Ivan Ermakov", "Denis Kuznedelev", "Vyacheslav Zhdanovskiy", "Yegor Yershov" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-04-09T00:00:00
https://arxiv.org/abs/2604.08426
https://arxiv.org/pdf/2604.08426v4
2604.08426
null
1
0
false
null
null
0.3442
219ab6374d4916c7e505d4a442b1d9194b034d9b45dbe0e21202df45f44a28bc
[ "arxiv", "semantic_scholar" ]
Seeing but Not Thinking: Routing Distraction in Multimodal Mixture-of-Experts
Multimodal Mixture-of-Experts (MoE) models have achieved remarkable performance on vision-language tasks. However, we identify a puzzling phenomenon termed Seeing but Not Thinking: models accurately perceive image content yet fail in subsequent reasoning, while correctly solving identical problems presented as pure tex...
[ "Haolei Xu", "Haiwen Hong", "Hongxing Li", "Rui Zhou", "Yang Zhang", "Longtao Huang", "Hui Xue", "Yongliang Shen", "Weiming Lu", "Yueting Zhuang" ]
[ "cs.CV", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-04-09T00:00:00
https://arxiv.org/abs/2604.08541
https://arxiv.org/pdf/2604.08541v1
2604.08541
null
2
0
false
null
null
0.3442
65aed9717acda0620036cb5f78ef6e1169172f9fdd76d78de1148e6cd5d41a8c
[ "arxiv", "semantic_scholar" ]
Can LoRA Fusion Support Cross-Domain Tasks in Cloud-Edge Collaboration?
Cloud-hosted large language models (LLMs) commonly rely on LoRA for domain adaptation, yet domain data are distributed across multiple edge devices and cannot be uploaded due to privacy constraints. This raises a fundamental question: how can knowledge from multiple private edges be integrated into a cloud LLM for cros...
[ "Yatong Wang", "Fali Wang", "Naibin Gu", "Zheng Lin", "Zhengxiao Liu", "Dingyu Yao", "Zhiwei Zhang", "Jianxin Shi", "Weiping Wang" ]
[ "cs.DC", "cs.CL" ]
[ "Computer Science" ]
2026-04-09T00:00:00
https://arxiv.org/abs/2605.23913
https://arxiv.org/pdf/2605.23913v1
2605.23913
null
0
0
false
null
null
0.3442
ccf6bb1000107964a5995f139d4f2cfcbaa98bb25485eb1072079be16f54bff3
[ "arxiv", "semantic_scholar" ]
MoBiE: Efficient Inference of Mixture of Binary Experts under Post-Training Quantization
Mixture-of-Experts (MoE) based large language models (LLMs) offer strong performance but suffer from high memory and computation costs. Weight binarization provides extreme efficiency, yet existing binary methods designed for dense LLMs struggle with MoE-specific issues, including cross-expert redundancy, task-agnostic...
[ "Zhixiong Zhao", "Zukang Xu", "Zhixuan Chen", "Dawei Yang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-08T00:00:00
https://arxiv.org/abs/2604.06798
https://arxiv.org/pdf/2604.06798v4
2604.06798
null
0
0
true
https://github.com/Kishon-zzx/MoBiE
null
0.6378
308bfc05e454a554872a2cf602f14ef95577800ee6dcaa03e87b73c2035daeee
[ "arxiv", "semantic_scholar" ]
AudioKV: KV Cache Eviction in Efficient Large Audio Language Models
Large Audio-Language Models (LALMs) have set new benchmarks in speech processing, yet their deployment is hindered by the memory footprint of the Key-Value (KV) cache during long-context inference. While general KV cache compression techniques excel in LLMs, they often fail in the audio domain by overlooking the intrin...
[ "Yuxuan Wang", "Peize He", "Xiyan Gui", "Xiaoqian Liu", "Junhao He", "Xuyang Liu", "Zichen Wen", "Xuming Hu", "Linfeng Zhang" ]
[ "cs.SD" ]
[ "Computer Science" ]
2026-04-08T00:00:00
https://arxiv.org/abs/2604.06694
https://arxiv.org/pdf/2604.06694v1
2604.06694
null
1
0
false
null
null
0.3434
6cbfcfa948788f6c102de75255ea891114756250d835c043a13f3614ea31947e
[ "arxiv", "semantic_scholar" ]
Does a Global Perspective Help Prune Sparse MoEs Elegantly?
Empirical scaling laws for language models have encouraged the development of ever-larger LLMs, despite their growing computational and memory costs. Sparse Mixture-of-Experts (MoEs) offer a promising alternative by activating only a subset of experts per forward pass, improving efficiency without sacrificing performan...
[ "Zeliang Zhang", "Nikhil Ghosh", "Jiani Liu", "Bin Yu", "Xiaodong Liu" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-04-08T00:00:00
https://arxiv.org/abs/2604.06542
https://arxiv.org/pdf/2604.06542v1
2604.06542
null
0
0
false
null
null
0.3434
17a3008c95aaa5ef05bc9849d9b88037f9e3cdc5efb3c45430990406554f06fa
[ "arxiv", "semantic_scholar" ]
Region-Graph Optimal Transport Routing for Mixture-of-Experts Whole-Slide Image Classification
Multiple Instance Learning (MIL) is the dominant framework for gigapixel whole-slide image (WSI) classification in computational pathology. However, current MIL aggregators route all instances through a shared pathway, constraining their capacity to specialise across the pathological heterogeneity inherent in each slid...
[ "Xin Tian", "Jiuliu Lu", "Ephraim Tsalik", "Bart Wanders", "Colleen Knoth", "Julian Knight" ]
[ "cs.CV", "cs.AI", "eess.IV" ]
[ "Computer Science", "Engineering" ]
2026-04-08T00:00:00
https://arxiv.org/abs/2604.07298
https://arxiv.org/pdf/2604.07298v1
2604.07298
null
0
0
false
null
null
0.3434
51229331cfd93bfc4842bf4203772d54017495689fe97703c5201ea146216d21
[ "arxiv", "semantic_scholar" ]
ForkKV: Scaling Multi-LoRA Agent Serving via Copy-on-Write Disaggregated KV Cache
The serving paradigm of large language models (LLMs) is rapidly shifting towards complex multi-agent workflows where specialized agents collaborate over massive shared contexts. While Low-Rank Adaptation (LoRA) enables the efficient co-hosting of these specialized agents on a single base model, it introduces a critical...
[ "Shao Wang", "Rui Ren", "Lin Gui" ]
[ "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2026-04-07T00:00:00
https://arxiv.org/abs/2604.06370
https://arxiv.org/pdf/2604.06370v1
2604.06370
null
0
0
false
null
null
0.3427
5a4211aedb271a766a363d381406bad463e14e3d760dc9966a47580e74f27e5e
[ "arxiv", "semantic_scholar" ]
Efficient Quantization of Mixture-of-Experts with Theoretical Generalization Guarantees
Sparse Mixture-of-Experts (MoE) allows scaling of language and vision models efficiently by activating only a small subset of experts per input. While this reduces computation, the large number of parameters still incurs substantial memory overhead during inference. Post-training quantization has been explored to addre...
[ "Mohammed Nowaz Rabbani Chowdhury", "Kaoutar El Maghraoui", "Hsinyu Tsai", "Naigang Wang", "Geoffrey W. Burr", "Liu Liu", "Meng Wang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-07T00:00:00
https://arxiv.org/abs/2604.06515
https://arxiv.org/pdf/2604.06515v1
2604.06515
null
2
0
false
null
The Fourteenth International Conference on Learning Representations, 2026
0.5385
1e395dce6b6afe0baf0fb281b998d39c30dad6a021f50e828371daf53c8d0d9d
[ "arxiv", "semantic_scholar" ]
HybridKV: Hybrid KV Cache Compression for Efficient Multimodal Large Language Model Inference
Multimodal Large Language Models (MLLMs) have advanced unified reasoning over text, images, and videos, but their inference is hindered by the rapid growth of key-value (KV) caches. Each visual input expands into thousands of tokens, causing caches to scale linearly with context length and remain resident in GPU memory...
[ "Bowen Zeng", "Feiyang Ren", "Jun Zhang", "Xiaoling Gu", "Ke Chen", "Lidan Shou", "Huan Li" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-04-07T00:00:00
https://arxiv.org/abs/2604.05887
https://arxiv.org/pdf/2604.05887v1
2604.05887
null
6
0
false
null
null
0.3427
37a5f2fa34e7cc188961d2c2c2725bf22dc9c89b1f2479af0e936e6f4fc943e8
[ "arxiv", "semantic_scholar" ]
Agentic AI-Based Joint Computing and Networking via Mixture of Experts and Large Language Models
Future sixth-generation (6G) mobile networks are envisioned to be equipped with a diverse set of powerful, yet highly specialized, optimization experts. Such a promising vision is concurrently expected to give rise to the need for scalable mechanisms that can select, combine, and orchestrate such experts based on high-...
[ "Robert-Jeron Reifert", "Alaa Alameer Ahmad", "Hayssam Dahrouj", "Aydin Sezgin" ]
[ "cs.LG", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2026-04-07T00:00:00
https://arxiv.org/abs/2605.02911
https://arxiv.org/pdf/2605.02911v1
2605.02911
null
0
0
false
null
null
0.3427
6474c051b7f535373c023060984142a39c5237dd3f132abe95ce2f8051a722f1
[ "arxiv", "semantic_scholar" ]
ALTO: Adaptive LoRA Tuning and Orchestration for Heterogeneous LoRA Training Workloads
Low-Rank Adaptation (LoRA) is now the dominant method for parameter-efficient fine-tuning of large language models, but achieving a high-quality adapter often requires systematic hyperparameter tuning because LoRA performance is highly sensitive to configuration choices. In practice, this leads to many concurrent LoRA ...
[ "Jingwei Zuo", "Xinze Feng", "Zien Liu", "Kaijian Wang", "Fanjiang Ye", "Ye Cao", "Zhuang Wang", "Yuke Wang" ]
[ "cs.LG", "cs.AI", "cs.DC" ]
[ "Computer Science" ]
2026-04-07T00:00:00
https://arxiv.org/abs/2604.05426
https://arxiv.org/pdf/2604.05426v2
2604.05426
null
2
0
false
null
null
0.3427
36fef3816df51c37bee1ad8250c6c85afc4b0d91665c2044c740b6b9140b7372
[ "arxiv", "semantic_scholar" ]
Don't Waste Bits! Adaptive KV-Cache Quantization for Lightweight On-Device LLMs
Large Language Models (LLMs) have achieved remarkable progress across reasoning, generation, and decision-making tasks, yet deploying them on mobile, embedded, and edge devices remains particularly challenging. On-device LLM inference is heavily constrained by the memory and bandwidth overhead of the key-value (KV) cac...
[ "Sayed Pedram Haeri Boroujeni", "Niloufar Mehrabi", "Patrick Woods", "Gabriel Hillesheim", "Abolfazl Razi" ]
[ "cs.CV" ]
[ "Computer Science" ]
2026-04-06T00:00:00
https://arxiv.org/abs/2604.04722
https://arxiv.org/pdf/2604.04722v1
2604.04722
null
5
0
false
null
null
0.342
6bcc5b6491f28096042705d410186d8320757c3d1bf65176e49602fc85b231d5
[ "arxiv", "semantic_scholar" ]
REAM: Merging Improves Pruning of Experts in LLMs
Mixture-of-Experts (MoE) large language models (LLMs) are among the top-performing architectures. The largest models, often with hundreds of billions of parameters, pose significant memory challenges for deployment. Traditional approaches to reduce memory requirements include weight pruning and quantization. Motivated ...
[ "Saurav Jha", "Maryam Hashemzadeh", "Ali Saheb Pasand", "Ali Parviz", "Min-Joong Lee", "Boris Knyazev" ]
[ "cs.AI", "cs.CL", "cs.LG", "cs.PF" ]
[ "Computer Science" ]
2026-04-06T00:00:00
https://arxiv.org/abs/2604.04356
https://arxiv.org/pdf/2604.04356v1
2604.04356
null
2
1
true
https://github.com/SamsungSAILMontreal/ream
null
0.6351
d9dbc10430bf6f734e04ca85badd470cab1a792eba28e69786560fb49a85cc6e
[ "arxiv", "semantic_scholar" ]
eOptShrinkQ: Near-Lossless KV Cache Compression Through Optimal Spectral Denoising and Quantization
We show that the key-value (KV) cache in transformer attention heads admits a natural decomposition into a low-rank \emph{shared context} component and a full-rank \emph{per-token} residual, well described by the spiked random matrix model. This observation leads to eOptShrinkQ, a two-stage compression pipeline: optima...
[ "Pei-Chun Su" ]
[ "cs.LG", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2026-04-06T00:00:00
https://arxiv.org/abs/2605.02905
https://arxiv.org/pdf/2605.02905v1
2605.02905
null
0
0
false
null
null
0.342
9c0aee3dfabc0aeeb633a14a1afa1e698326f29e460b541457faded15a1621b4
[ "arxiv", "semantic_scholar" ]
Comparative Characterization of KV Cache Management Strategies for LLM Inference
Efficient inference with Large Language Models (LLMs) increasingly relies on Key-Value (KV) caches to store previously computed key and value vectors at each layer. These caches are essential to minimize redundant computation during autoregressive token generation, lowering computational complexity from quadratic to li...
[ "Oteo Mamo", "Olga Kogiou", "Hyunjin Yi", "Weikuan Yu" ]
[ "cs.AR", "cs.AI" ]
[ "Computer Science" ]
2026-04-06T00:00:00
https://arxiv.org/abs/2604.05012
https://arxiv.org/pdf/2604.05012v1
2604.05012
null
0
0
false
null
null
0.342
5acb70ba0d37755d9dd17c8aa87a6c1028581d0aefa80f36972b3915d2b0696e
[ "arxiv", "semantic_scholar" ]
Rényi Attention Entropy for Patch Pruning
Transformers are strong baselines in both vision and language because self-attention captures long-range dependencies across tokens. However, the cost of self-attention grows quadratically with the number of tokens. Patch pruning mitigates this cost by estimating per-patch importance and removing redundant patches. To ...
[ "Hiroaki Aizawa", "Yuki Igaue" ]
[ "cs.CV", "cs.LG" ]
[ "Computer Science" ]
2026-04-04T00:00:00
https://arxiv.org/abs/2604.03803
https://arxiv.org/pdf/2604.03803v1
2604.03803
null
0
0
false
null
null
0.3405
093508dcf98330e7145f1b62d18510fb0be2be570700e7ca54d10912969ba7a5
[ "arxiv", "semantic_scholar" ]
Stochastic KV Routing: Enabling Adaptive Depth-Wise Cache Sharing
Serving transformer language models with high throughput requires caching Key-Values (KVs) to avoid redundant computation during autoregressive generation. The memory footprint of KV caching is significant and heavily impacts serving costs. This work proposes to lessen these memory requirements. While recent work has l...
[ "Anastasiia Filippova", "David Grangier", "Marco Cuturi", "João Monteiro" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-04-03T00:00:00
https://arxiv.org/abs/2604.22782
https://arxiv.org/pdf/2604.22782v1
2604.22782
null
0
0
false
null
null
0.3398
a1734fcfed2f160ad1549680e3c0d790c51a4eb3268a023689317cac831cdd64
[ "arxiv", "semantic_scholar" ]
QAPruner: Quantization-Aware Vision Token Pruning for Multimodal Large Language Models
Multimodal Large Language Models (MLLMs) have shown strong reasoning ability, but their high computational and memory costs hinder deployment in resource-constrained settings. While Post-Training Quantization (PTQ) and vision token pruning are standard compression techniques, they are usually treated as independent opt...
[ "Xinhao Wang", "Zhonyu Xia", "Zhiwei Lin", "Zhe Li", "Yongtao Wang" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2026-04-03T00:00:00
https://arxiv.org/abs/2604.02816
https://arxiv.org/pdf/2604.02816v1
2604.02816
null
0
0
false
null
null
0.3398
910adb7abaebfa3718d2e3b741bf4c3747dce8c460da00a598cb4e60e5f4acca
[ "arxiv", "semantic_scholar" ]
TokenDance: Scaling Multi-Agent LLM Serving via Collective KV Cache Sharing
Multi-agent LLM applications organize execution in synchronized rounds where a central scheduler gathers outputs from all agents and redistributes the combined context. This All-Gather communication pattern creates massive KV Cache redundancy, because every agent's prompt contains the same shared output blocks, yet exi...
[ "Zhuohang Bian", "Feiyang Wu", "Chengrui Zhang", "Hangcheng Dong", "Yun Liang", "Youwei Zhuo" ]
[ "cs.DC" ]
[ "Computer Science" ]
2026-04-03T00:00:00
https://arxiv.org/abs/2604.03143
https://arxiv.org/pdf/2604.03143v1
2604.03143
null
1
0
false
null
null
0.3398
a324edd365652390c578a5f03d72bfdc8719e5241e636c188f87adbd45f89fbd
[ "arxiv", "semantic_scholar" ]
Mixture-of-Experts in Remote Sensing: A Survey
Remote sensing data analysis and interpretation present unique challenges due to the diversity in sensor modalities and spatiotemporal dynamics of Earth observation data. Mixture-of-Experts (MoE) model has emerged as a powerful paradigm that addresses these challenges by dynamically routing inputs to specialized expert...
[ "Yongchuan Cui", "Peng Liu", "Lajiao Chen" ]
[ "cs.CV" ]
[ "Computer Science" ]
2026-04-03T00:00:00
https://arxiv.org/abs/2604.03342
https://arxiv.org/pdf/2604.03342v1
2604.03342
null
0
0
false
null
https://www.icck.org/article/abs/jgrs.2025.140654
0.534
a62dab8eb8e64f4c5e996644409c8b60bd38b04c9876d49cf6b00644f12629f7
[ "arxiv", "semantic_scholar" ]
The Expert Strikes Back: Interpreting Mixture-of-Experts Language Models at Expert Level
Mixture-of-Experts (MoE) architectures have become the dominant choice for scaling Large Language Models (LLMs), activating only a subset of parameters per token. While MoE architectures are primarily adopted for computational efficiency, it remains an open question whether their sparsity makes them inherently easier t...
[ "Jeremy Herbst", "Stefan Wermter", "Jae Hee Lee" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2026-04-02T00:00:00
https://arxiv.org/abs/2604.02178
https://arxiv.org/pdf/2604.02178v2
2604.02178
null
5
0
true
https://github.com/jerryy33/MoE_analysis
null
0.6297
f60562b1504d3ed5cb95bc02897ba005ed8fdaf8f74f86f08730c4fcd0d70658
[ "arxiv", "semantic_scholar" ]
Attention to Mamba: A Recipe for Cross-Architecture Distillation
State Space Models (SSMs) such as Mamba have become a popular alternative to Transformer models, due to their reduced memory consumption and higher throughput at generation compared to their Attention-based counterparts. On the other hand, the community has built up a considerable body of knowledge on how to train Tran...
[ "Abhinav Moudgil", "Ningyuan Huang", "Eeshan Gunesh Dhekane", "Pau Rodríguez", "Luca Zappella", "Federico Danieli" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2026-04-01T00:00:00
https://arxiv.org/abs/2604.14191
https://arxiv.org/pdf/2604.14191v1
2604.14191
null
1
0
false
null
null
0.3383
1d2e9986934d368ad386fc2e878f262e2c5616acbfca731dceb5fe1e507ad859
[ "arxiv", "semantic_scholar" ]
Routing-Free Mixture-of-Experts
Standard Mixture-of-Experts (MoE) models rely on centralized routing mechanisms that introduce rigid inductive biases. We propose Routing-Free MoE which eliminates any hard-coded centralized designs including external routers, Softmax, Top-K and load balancing, instead encapsulating all activation functionalities withi...
[ "Yilun Liu", "Jinru Han", "Sikuan Yan", "Volker Tresp", "Yunpu Ma" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-04-01T00:00:00
https://arxiv.org/abs/2604.00801
https://arxiv.org/pdf/2604.00801v1
2604.00801
null
0
0
true
https://github.com/liuyilun2000/RoutingFreeMoE/tree/release
null
0.6283
011fd8755dbdc3b571d133cc955add35d6a97cd5aa0e16945a88b68857cd44de
[ "arxiv", "semantic_scholar" ]
Cost-Penalized Fitness in FMA-Orchestrated Mixture of Experts: Experimental Evidence for Molecular Memory in Domain Adaptation
We present experimental results from seven controlled runs of nanoFMT, a Free-Market Algorithm (FMA) orchestrated transformer with dynamic Mixture-of-Experts (MoE) management. The experiments address a fundamental question for advanced LLM development: how should an MoE system manage its expert pool when operating at f...
[ "Martin Jaraiz" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-04-01T00:00:00
https://arxiv.org/abs/2604.00812
https://arxiv.org/pdf/2604.00812v1
2604.00812
null
0
0
false
null
null
0.3383
b8aecc7782f427271f02090f4830d1029c9afffe3a1be42efda589167dbdcb3f
[ "arxiv", "semantic_scholar" ]
IWP: Token Pruning as Implicit Weight Pruning in Large Vision Language Models
Large Vision Language Models show impressive performance across image and video understanding tasks, yet their computational cost grows rapidly with the number of visual tokens. Existing token pruning methods mitigate this issue through empirical approaches while overlooking the internal mechanism of attention. In this...
[ "Dong-Jae Lee", "Sunghyun Baek", "Junmo Kim" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2026-04-01T00:00:00
https://arxiv.org/abs/2604.00757
https://arxiv.org/pdf/2604.00757v1
2604.00757
null
0
0
false
null
null
0.3383
4feedb03b6b9207675d85a30b385886576a915f156dce9f8375467c65f26112c
[ "arxiv", "semantic_scholar" ]
Quantization with Unified Adaptive Distillation to enable multi-LoRA based one-for-all Generative Vision Models on edge
Generative Artificial Intelligence (GenAI) features such as image editing, object removal, and prompt-guided image transformation are increasingly integrated into mobile applications. However, deploying Large Vision Models (LVMs) for such tasks on resource-constrained devices remains challenging due to their high memor...
[ "Sowmya Vajrala", "Aakash Parmar", "Prasanna R", "Sravanth Kodavanti", "Manjunath Arveti", "Srinivas Soumitri Miriyala", "Ashok Senapati" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2026-03-31T00:00:00
https://arxiv.org/abs/2603.29535
https://arxiv.org/pdf/2603.29535v1
2603.29535
null
0
0
false
null
null
0.3376
fd27f171f34c8f163c221f6302243f1560363adf5287ed45cd2482786ee9cd70
[ "arxiv", "semantic_scholar" ]
On the Role of Encoder Depth: Pruning Whisper and LoRA Fine-Tuning in SLAM-ASR
Automatic speech recognition (ASR) has advanced rapidly in recent years, driven by large-scale pretrained models and end-to-end architectures such as SLAM-ASR. A key component of SLAM-ASR systems is the Whisper speech encoder, which provides robust acoustic representations. While model pruning has been explored for the...
[ "Ganesh Pavan Kartikeya Bharadwaj Kolluri", "Michael Kampouridis", "Ravi Shekhar" ]
[ "cs.CL", "cs.SD" ]
[ "Computer Science" ]
2026-03-30T00:00:00
https://arxiv.org/abs/2603.27981
https://arxiv.org/pdf/2603.27981v1
2603.27981
null
0
0
false
null
null
0.3369
c75bedd21fe657046b7aacafac1a6f00ab2e41bcedcf9662dce2ef498807745d
[ "arxiv", "semantic_scholar" ]
IsoQuant: Hardware-Aligned SO(4) Isoclinic Rotations for LLM KV Cache Compression
Orthogonal feature decorrelation is effective for low-bit online vector quantization, but dense random orthogonal transforms incur prohibitive $O(d^2)$ storage and compute. RotorQuant reduces this cost with blockwise $3$D Clifford rotors, yet the resulting $3$D partition is poorly aligned with modern hardware and offer...
[ "Zhongping Ji" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2026-03-30T00:00:00
https://arxiv.org/abs/2603.28430
https://arxiv.org/pdf/2603.28430v1
2603.28430
null
2
1
false
null
null
0.3369
6b37f8878e1b8c13df5c169f008fea4531d46e0cd1f1d61cd090a68aa9da925a
[ "arxiv", "semantic_scholar" ]
Low-Latency Edge LLM Handover via Joint KV Cache Transfer and Token Prefill
Edge deployment of large language models (LLMs) can reduce latency for interactive services, but mobility introduces service interruptions when an user equipment (UE) hands over between base stations (BSs). To promptly resume decoding, the target-side edge server must recover the UE context state, which can be provisio...
[ "Seunghun Lee", "Jihong Park", "Ce Zheng", "Hyuncheol Park" ]
[ "eess.SP" ]
[ "Engineering" ]
2026-03-30T00:00:00
https://arxiv.org/abs/2603.28018
https://arxiv.org/pdf/2603.28018v1
2603.28018
null
1
1
false
null
null
0.3369
45f5bec4189d537fc0ceb1a84d25d17b1df005f2eab557b46525ebb8b5d668fd
[ "arxiv", "semantic_scholar" ]
KV Cache Quantization for Self-Forcing Video Generation: A 33-Method Empirical Study
Self-forcing video generation extends a short-horizon video model to longer rollouts by repeatedly feeding generated content back in as context. This scaling path immediately exposes a systems bottleneck: the key-value (KV) cache grows with rollout length, so longer videos require not only better generation quality but...
[ "Suraj Ranganath", "Vaishak Menon", "Anish Patnaik" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-29T00:00:00
https://arxiv.org/abs/2603.27469
https://arxiv.org/pdf/2603.27469v1
2603.27469
null
2
0
true
https://github.com/suraj-ranganath/kv-quant-longhorizon/
null
0.6243
de79a316f833cba93384f952cde4c399b561cc29f2f6bd0d7a84a425e4583bae
[ "arxiv", "semantic_scholar" ]
KVSculpt: KV Cache Compression as Distillation
KV cache compression is critical for efficient long-context LLM inference. Approaches that reduce the per-pair footprint -- quantization and low-rank decomposition -- are orthogonal to those that reduce the sequence length of the cache. Along the sequence-length dimension, existing methods range from pure eviction -- s...
[ "Bo Jiang", "Sian Jin" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2026-03-29T00:00:00
https://arxiv.org/abs/2603.27819
https://arxiv.org/pdf/2603.27819v1
2603.27819
null
0
0
false
null
null
0.3361
195271ef729656e923cf5f02544ee6086b42e078b17382ad0228b602e4cbbd9e
[ "arxiv", "semantic_scholar" ]
TurboAngle: Near-Lossless KV Cache Compression via Uniform Angle Quantization
We compress KV cache entries by quantizing angles in the Fast Walsh-Hadamard domain, where a random diagonal rotation makes consecutive element pairs approximately uniformly distributed on the unit circle. We extend this angular quantizer with per-layer early-boost, which independently configures K and V codebook sizes...
[ "Dipkumar Patel" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-29T00:00:00
https://arxiv.org/abs/2603.27467
https://arxiv.org/pdf/2603.27467v1
2603.27467
null
2
0
false
null
null
0.3361
dddaee0a13abd30e37ea3ffd2f1facd8d17bdcddb97e29a535dbb97fe6cf0e83
[ "arxiv", "semantic_scholar" ]
ScoutAttention: Efficient KV Cache Offloading via Layer-Ahead CPU Pre-computation for LLM Inference
Large language models encounter critical GPU memory capacity constraints during long-context inference, where KV cache memory consumption severely limits decode batch sizes. While existing research has explored offloading KV cache to DRAM, these approaches either demand frequent GPU-CPU data transfers or impose extensi...
[ "Qiuyang Zhang", "Kai Zhou", "Ding Tang", "Kai Lu", "Cheng Li", "Zhenyu Yang", "Peng Xu", "Jiguang Wan" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-03-28T00:00:00
https://arxiv.org/abs/2603.27138
https://arxiv.org/pdf/2603.27138v1
2603.27138
null
0
0
false
null
null
0.3354
29ac0b5282c6eef232598d0df8e417496472c94112baad74f59aa6a81b4325ab
[ "arxiv", "semantic_scholar" ]
TurboESM: Ultra-Efficient 3-Bit KV Cache Quantization for Protein Language Models with Orthogonal Rotation and QJL Correction
The rapid scaling of Protein Language Models (PLMs) has unlocked unprecedented accuracy in protein structure prediction and design, but the quadratic memory growth of the Key-Value (KV) cache during inference remains a prohibitive barrier for single-GPU deployment and high-throughput generation. While 8-bit quantizatio...
[ "Yue Hu", "Junqing Wang", "Yingchao Liu" ]
[ "q-bio.QM" ]
[ "Biology" ]
2026-03-27T00:00:00
https://arxiv.org/abs/2603.26110
https://arxiv.org/pdf/2603.26110v1
2603.26110
null
0
0
false
null
null
0.3347
afed2caba7ddda81f5c0da9f292ebf40a68ef5edd83468504f10576a804e3c1a
[ "arxiv", "semantic_scholar" ]
TTKV: Temporal-Tiered KV Cache for Long-Context LLM Inference
Key-value (KV) caching is critical for efficient inference in large language models (LLMs), yet its memory footprint scales linearly with context length, resulting in a severe scalability bottleneck. Existing approaches largely treat KV states as equally important across time, implicitly assuming uniform precision and ...
[ "Gradwell Dzikanyanga", "Weihao Yang", "Hao Huang", "Donglei Wu", "Shihao Wang", "Wen Xia", "Sanjeeb K C" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2026-03-27T00:00:00
https://arxiv.org/abs/2604.19769
https://arxiv.org/pdf/2604.19769v1
2604.19769
null
0
0
false
null
null
0.3347
6c4934016b13a0b356a18c4473ac9eecda572f44dc87c41d9301485ed767a327
[ "arxiv", "semantic_scholar" ]
Bayesian estimation of optical constants using mixtures of Gaussian process experts
We propose modeling absorption spectrum measurements as mixtures of Gaussian process experts. This enables us to construct a flexible statistical model for interpolating and extrapolating measurements, facilitating statistical integration of Kramers-Kronig relations to estimate the whole complex refractive index. Addit...
[ "Teemu Härkönen", "Hui Chen", "Erik Vartiainen" ]
[ "stat.AP", "physics.data-an" ]
[ "Mathematics", "Physics" ]
2026-03-27T00:00:00
https://arxiv.org/abs/2603.26334
https://arxiv.org/pdf/2603.26334v1
2603.26334
null
0
0
false
null
null
0.3347
f3b5b82a8737ec2dc0b25dce7e024e60ff54d8cf0fdaf797a73a193e81137698
[ "arxiv", "semantic_scholar" ]
How Pruning Reshapes Features: Sparse Autoencoder Analysis of Weight-Pruned Language Models
Weight pruning is a standard technique for compressing large language models, yet its effect on learned internal representations remains poorly understood. We present the first systematic study of how unstructured pruning reshapes the feature geometry of language models, using Sparse Autoencoders (SAEs) as interpretabi...
[ "Hector Borobia", "Elies Seguí-Mas", "Guillermina Tormo-Carbó" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-26T00:00:00
https://arxiv.org/abs/2603.25325
https://arxiv.org/pdf/2603.25325v1
2603.25325
null
2
1
true
https://github.com/hborobia/sae-pruning-paper
null
0.6202
bcf6a3e028a27d0e6c9e7c0b3c711b4019234fcd93e71ef99136128515d40236
[ "arxiv", "semantic_scholar" ]
MoE-Sieve: Routing-Guided LoRA for Efficient MoE Fine-Tuning
Standard LoRA fine-tuning of Mixture-of-Experts (MoE) models applies adapters to every expert, yet our profiling shows that per-layer expert routing is highly skewed: a small subset of experts handles most tokens in each layer, while many others are rarely activated ("cold"). We propose MoE-Sieve, a simple routing-guid...
[ "Andrea Manzoni" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2026-03-25T00:00:00
https://arxiv.org/abs/2603.24044
https://arxiv.org/pdf/2603.24044v1
2603.24044
null
0
0
false
null
null
0.3332
62df5e004553baa4d283ee0b70c1b4eeeec422d0d171e20a5ce568623093473b
[ "arxiv", "semantic_scholar" ]
EchoKV: Efficient KV Cache Compression via Similarity-Based Reconstruction
The increasing memory demand of the Key-Value (KV) cache poses a significant bottleneck for Large Language Models (LLMs) in long-context applications. Existing low-rank KV compression methods reduce this footprint by modifying model projections, limiting the flexibility to switch back to standard full-cache inference w...
[ "Shiyu Ji", "Yixuan Wang", "Yijun Liu", "Qingfu Zhu", "Wanxiang Che" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-03-24T00:00:00
https://arxiv.org/abs/2603.22910
https://arxiv.org/pdf/2603.22910v2
2603.22910
null
0
0
false
null
null
0.3325
dec77340d7e040a40a873747bdb5bd3cb23b234de68ffb37624da3b3ac4fa688
[ "arxiv", "semantic_scholar" ]
PCR: A Prefetch-Enhanced Cache Reuse System for Low-Latency RAG Serving
Retrieval-Augmented Generation (RAG) systems enhance the performance of large language models (LLMs) by incorporating supplementary retrieved documents, enabling more accurate and context-aware responses. However, integrating these external documents often results in very long input sequences, which significantly incre...
[ "Wenfeng Wang", "Xiaofeng Hou", "Peng Tang", "Hengyi Zhou", "Jing Wang", "Xinkai Wang", "Chao Li", "Minyi Guo" ]
[ "cs.DC" ]
[ "Computer Science" ]
2026-03-24T00:00:00
https://arxiv.org/abs/2603.23049
https://arxiv.org/pdf/2603.23049v1
2603.23049
null
1
0
false
null
null
0.3325
f25c2bf34020c76bf7e0dd7b28de3e3c3d51026e3ce76f6505ed71994b7adea1
[ "arxiv", "semantic_scholar" ]
Knowledge Packs: Zero-Token Knowledge Delivery via KV Cache Injection
RAG wastes tokens. We propose Knowledge Packs: pre-computed KV caches that deliver the same knowledge at zero token cost. For causal transformers, the KV cache from a forward pass on text F is identical to what a joint pass on F+q would produce - this follows directly from the causal mask. The equivalence is exact but ...
[ "Andrey Pustovit" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-03-22T00:00:00
https://arxiv.org/abs/2604.03270
https://arxiv.org/pdf/2604.03270v1
2604.03270
null
0
0
true
https://github.com/cnails/kv-knowledge-packs
null
0.6148
912b19da08c9ca08ebdcf35648f0ee66f59e9de9f271f3d5d81510bada193aef
[ "arxiv", "semantic_scholar" ]
HELIX: Scaling Raw Audio Understanding with Hybrid Mamba-Attention Beyond the Quadratic Limit
Audio representation learning typically evaluates design choices such as input frontend, sequence backbone, and sequence length in isolation. We show that these axes are coupled, and conclusions from one setting often do not transfer to others. We introduce HELIX, a controlled framework comparing pure Mamba, pure atten...
[ " Khushiyant", "Param Thakkar" ]
[ "cs.SD", "cs.LG", "eess.AS" ]
[ "Computer Science", "Engineering" ]
2026-03-22T00:00:00
https://arxiv.org/abs/2603.21316
https://arxiv.org/pdf/2603.21316v1
2603.21316
null
0
0
false
null
null
0.331
a0e433d3b6b7886b962a68ddb00bfe5a79a718a5e9893fab7604f24f34e805d9
[ "arxiv", "semantic_scholar" ]
Beyond Token Eviction: Mixed-Dimension Budget Allocation for Efficient KV Cache Compression
Key-value (KV) caching is widely used to accelerate transformer inference, but its memory cost grows linearly with input length, limiting long-context deployment. Existing token eviction methods reduce memory by discarding less important tokens, which can be viewed as a coarse form of dimensionality reduction that assi...
[ "Ruijie Miao", "Zhiming Wang", "Wang Li", "Shiwei Wu", "Shufan Liu", "Yanbing Jiang", "Tong Yang" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-03-21T00:00:00
https://arxiv.org/abs/2603.20616
https://arxiv.org/pdf/2603.20616v1
2603.20616
null
0
0
false
null
null
0.3303
59ca1cb0036a67455fcc20e04efbce80ffa00238eb739f7304e3bf32f38e62f6
[ "arxiv", "semantic_scholar" ]
KV Cache Optimization Strategies for Scalable and Efficient LLM Inference
The key-value (KV) cache is a foundational optimization in Transformer-based large language models (LLMs), eliminating redundant recomputation of past token representations during autoregressive generation. However, its memory footprint scales linearly with context length, imposing critical bottlenecks on GPU memory ca...
[ "Yichun Xu", "Navjot K. Khaira", "Tejinder Singh" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-20T00:00:00
https://arxiv.org/abs/2603.20397
https://arxiv.org/pdf/2603.20397v1
2603.20397
null
1
0
false
null
null
0.3296
29d2bec40fe3070d962991082a5d3e20871d93a9cc36aac85d2a82a41a4adc8d
[ "arxiv", "semantic_scholar" ]
The Residual Stream Is All You Need: On the Redundancy of the KV Cache in Transformer Inference
The key-value (KV) cache is widely treated as essential state in transformer inference, and a large body of work engineers policies to compress, evict, or approximate its entries. We prove that this state is entirely redundant: keys and values at every layer are deterministic projections of the residual stream, and rec...
[ "Kaleem Ullah Qasim", "Jiashu Zhang", "Muhammad Kafeel Shaheen", "Razan Alharith", "Heying Zhang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-20T00:00:00
https://arxiv.org/abs/2603.19664
https://arxiv.org/pdf/2603.19664v1
2603.19664
null
1
0
true
https://github.com/Kaleemullahqasim/KV-Direct
null
0.6121
184cb8a2cfcc3629fd3cdbf524eea69b8b4bb7fad1d4eff7095a5288de5fa696
[ "arxiv", "semantic_scholar" ]
Mixture of Experts with Soft Nearest Neighbor Loss: Resolving Expert Collapse via Representation Disentanglement
The Mixture-of-Experts (MoE) model uses a set of expert networks that specialize on subsets of a dataset under the supervision of a gating network. A common issue in MoE architectures is ``expert collapse'' where overlapping class boundaries in the raw input feature space cause multiple experts to learn redundant repre...
[ "Abien Fred Agarap", "Arnulfo P. Azcarraga" ]
[ "cs.NE", "cs.LG" ]
[ "Computer Science" ]
2026-03-20T00:00:00
https://arxiv.org/abs/2603.26734
https://arxiv.org/pdf/2603.26734v1
2603.26734
null
0
0
false
null
null
0.3296
505e1a8e965845c6e4a5ceb2d5777a1e8a7d9508dda4192e4697f2fdd5caadba
[ "arxiv", "semantic_scholar" ]
EntropyCache: Decoded Token Entropy Guided KV Caching for Diffusion Language Models
Diffusion-based large language models (dLLMs) rely on bidirectional attention, which prevents lossless KV caching and requires a full forward pass at every denoising step. Existing approximate KV caching methods reduce this cost by selectively updating cached states, but their decision overhead scales with context leng...
[ "Minsoo Cheong", "Donghyun Son", "Woosang Lim", "Sungjoo Yoo" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-03-19T00:00:00
https://arxiv.org/abs/2603.18489
https://arxiv.org/pdf/2603.18489v1
2603.18489
null
0
0
true
https://github.com/mscheong01/EntropyCache
null
0.6107
c0380f0d11c31a45214ba7131ec6637859c727b80e9233598d9d881bfd15d4ec
[ "arxiv", "semantic_scholar" ]
Prune-then-Quantize or Quantize-then-Prune? Understanding the Impact of Compression Order in Joint Model Compression
What happens when multiple compression methods are combined-does the order in which they are applied matter? Joint model compression has emerged as a powerful strategy to achieve higher efficiency by combining multiple methods such as pruning and quantization. A central but underexplored factor in joint model compressi...
[ "Minjun Kim", "Jaehyeon Choi", "Hyunwoo Yang", "Jongjin Kim", "Jinho Song", "U Kang" ]
[ "cs.AI" ]
[ "Computer Science" ]
2026-03-19T00:00:00
https://arxiv.org/abs/2603.18426
https://arxiv.org/pdf/2603.18426v1
2603.18426
null
1
0
false
null
null
0.3289
2bafc89e70d327ebae6bd46cce96bc73e1cc5b86ca09c6bfd3703d252c892d1e
[ "arxiv", "semantic_scholar" ]
AIMER: Calibration-Free Task-Agnostic MoE Pruning
Mixture-of-Experts (MoE) language models increase parameter capacity without proportional per-token compute, but the deployment still requires storing all experts, making expert pruning important for reducing memory and serving overhead. Existing task-agnostic expert pruning methods are typically calibration-dependent:...
[ "Zongfang Liu", "Shengkun Tang", "Yifan Shen", "Huan Wang", "Xin Yuan" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-03-19T00:00:00
https://arxiv.org/abs/2603.18492
https://arxiv.org/pdf/2603.18492v2
2603.18492
null
1
0
false
null
null
0.3289
d5826824fb9625f03c2bc48b0891d0a16ff27cae15ceb86b1b9e3a7e0dc3eb44
[ "arxiv", "semantic_scholar" ]
DyMoE: Dynamic Expert Orchestration with Mixed-Precision Quantization for Efficient MoE Inference on Edge
Despite the computational efficiency of MoE models, the excessive memory footprint and I/O overhead inherent in multi-expert architectures pose formidable challenges for real-time inference on resource-constrained edge platforms. While existing static methods struggle with a rigid latency-accuracy trade-off, we observe...
[ "Yuegui Huang", "Zhiyuan Fang", "Weiqi Luo", "Ruoyu Wu", "Wuhui Chen", "Zibin Zheng" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-03-19T00:00:00
https://arxiv.org/abs/2603.19172
https://arxiv.org/pdf/2603.19172v1
2603.19172
null
0
0
false
null
null
0.3289
edecef9494ffc7ff8a6f1b3e9f2a2e42b0cbd6fdf33f919d51a66512ad4b170e
[ "arxiv", "semantic_scholar" ]
Path-Constrained Mixture-of-Experts
Sparse Mixture-of-Experts (MoE) architectures route each token through a subset of experts at each layer independently. We propose viewing MoE computation through the lens of \emph{expert paths} -- the sequence of expert selections a token makes across all layers. This perspective reveals that, despite $N^L$ possible p...
[ "Zijin Gu", "Tatiana Likhomanenko", "Vimal Thilak", "Jason Ramapuram", "Navdeep Jaitly" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-03-18T00:00:00
https://arxiv.org/abs/2603.18297
https://arxiv.org/pdf/2603.18297v2
2603.18297
null
0
0
false
null
null
0.3281
b7ba6ad5fa85713fd7297c51c54422d952690b12f87badb8ec451851174c9f93
[ "arxiv", "semantic_scholar" ]
FEMBA on the Edge: Physiologically-Aware Pre-Training, Quantization, and Deployment of a Bidirectional Mamba EEG Foundation Model on an Ultra-low Power Microcontroller
Objective: To enable continuous, long-term neuro-monitoring on wearable devices by overcoming the computational bottlenecks of Transformer-based Electroencephalography (EEG) foundation models and the quantization challenges inherent to State-Space Models (SSMs). Methods: We present FEMBA, a bidirectional Mamba architec...
[ "Anna Tegon", "Nicholas Lehmann", "Yawei Li", "Andrea Cossettini", "Luca Benini", "Thorir Mar Ingolfsson" ]
[ "eess.SP", "cs.LG" ]
[ "Medicine", "Engineering", "Computer Science" ]
2026-03-18T00:00:00
https://arxiv.org/abs/2603.26716
https://arxiv.org/pdf/2603.26716v1
2603.26716
10.1109/tbme.2026.3683482
0
0
false
null
null
0.3281
7ecf1ba867b0007e1516b3029303e6402d54604bbb40f1d38c557828acdcbc6e
[ "arxiv", "semantic_scholar" ]
VQKV: High-Fidelity and High-Ratio Cache Compression via Vector-Quantization
The growing context length of Large Language Models (LLMs) enlarges the Key-Value (KV) cache, limiting deployment in resource-limited environments. Prior training-free approaches for KV cache compression typically rely on low-rank approximation or scalar quantization, which fail to simultaneously achieve high compressi...
[ "Yixuan Wang", "Qingyu Shi", "Jiayu Zhou", "Dianbo Liu", "Ziwei He", "Zhouhan Lin" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-03-17T00:00:00
https://arxiv.org/abs/2603.16435
https://arxiv.org/pdf/2603.16435v1
2603.16435
null
1
0
false
null
null
0.3274
87ad767d85ab2cba95f89ccd152b39026819399144b726e289237fb7f09697ef
[ "arxiv", "semantic_scholar" ]
Mixture of Style Experts for Diverse Image Stylization
Diffusion-based stylization has advanced significantly, yet existing methods are limited to color-driven transformations, neglecting complex semantics and material details. We introduce StyleExpert, a semantic-aware framework based on the Mixture of Experts (MoE). Our framework employs a unified style encoder, trained ...
[ "Shihao Zhu", "Ziheng Ouyang", "Yijia Kang", "Qilong Wang", "Mi Zhou", "Bo Li", "Ming-Ming Cheng", "Qibin Hou" ]
[ "cs.CV" ]
[ "Computer Science" ]
2026-03-17T00:00:00
https://arxiv.org/abs/2603.16649
https://arxiv.org/pdf/2603.16649v3
2603.16649
null
0
0
false
null
null
0.3274
bbc2f2a580c4694e729e3902134e7945e870c799bbea61f7b4184b4627cd2368
[ "arxiv", "semantic_scholar" ]
Frequency Matters: Fast Model-Agnostic Data Curation for Pruning and Quantization
Post-training model compression is essential for enhancing the portability of Large Language Models (LLMs) while preserving their performance. While several compression approaches have been proposed, less emphasis has been placed on selecting the most suitable set of data (the so-called \emph{calibration data}) for fin...
[ "Francesco Pio Monaco", "Elia Cunegatti", "Flavio Vella", "Giovanni Iacca" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2026-03-17T00:00:00
https://arxiv.org/abs/2603.16105
https://arxiv.org/pdf/2603.16105v3
2603.16105
null
1
0
true
https://github.com/FrancescoMonaco/ZipCal.}
null
0.608
73a6c57ab61a3d6802ae33462d4ba4d3c57ae63152da5fcbb770245e3be57e25
[ "arxiv", "semantic_scholar" ]
Mostly Text, Smart Visuals: Asymmetric Text-Visual Pruning for Large Vision-Language Models
Network pruning is an effective technique for enabling lightweight Large Vision-Language Models (LVLMs), which primarily incorporates both weights and activations into the importance metric. However, existing efforts typically process calibration data from different modalities in a unified manner, overlooking modality-...
[ "Sijie Li", "Biao Qian", "Jungong Han" ]
[ "cs.CV", "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2026-03-16T00:00:00
https://arxiv.org/abs/2603.16001
https://arxiv.org/pdf/2603.16001v1
2603.16001
null
0
0
true
https://github.com/LezJ/ATV-Pruning
null
0.6067
943a64d8001b65702238b54306a6c7b8e336656930e75411ac05ae1dc913342c
[ "arxiv", "semantic_scholar" ]
Bridging Local and Global Knowledge: Cascaded Mixture-of-Experts Learning for Near-Shortest Path Routing
While deep learning models that leverage local features have demonstrated significant potential for near-optimal routing in dense Euclidean graphs, they struggle to generalize well in sparse networks where topological irregularities require broader structural awareness. To address this limitation, we train a Cascaded M...
[ "Yung-Fu Chen", "Anish Arora" ]
[ "cs.LG", "cs.NI" ]
[ "Computer Science" ]
2026-03-16T00:00:00
https://arxiv.org/abs/2603.15541
https://arxiv.org/pdf/2603.15541v1
2603.15541
null
0
0
false
null
null
0.3267
4fc59cfa68ba8cce5c5e3f44cdf82e25a4ae633fd0a96c9f95dd611ef236dfba
[ "arxiv", "semantic_scholar" ]
ASAP: Attention-Shift-Aware Pruning for Efficient LVLM Inference
While Large Vision-Language Models (LVLMs) demonstrate exceptional multi-modal capabilities, the quadratic computational cost of processing high-resolution visual tokens remains a critical bottleneck. Though recent token reduction strategies attempt to accelerate inference, such methods inadequately exploit attention v...
[ "Surendra Pathak", "Bo Han" ]
[ "cs.CV", "cs.LG" ]
[ "Computer Science" ]
2026-03-15T00:00:00
https://arxiv.org/abs/2603.14549
https://arxiv.org/pdf/2603.14549v2
2603.14549
null
1
0
false
null
null
0.3259
0a71a7ce56e0e696fe4c5519909e4f79843a225ea39dac30b0e23d3314d40ab1
[ "arxiv", "semantic_scholar" ]
OxyGen: Unified KV Cache Management for VLA Inference under Multi-Task Parallelism
Embodied AI agents increasingly require parallel execution of multiple tasks, such as manipulation, conversation, and memory construction, from shared observations under distinct time constraints. Recent Mixture-of-Transformers (MoT) Vision-Language-Action Models (VLAs) architecturally support such heterogeneous output...
[ "Xiangyu Li", "Huaizhi Tang", "Xin Ding", "Weijun Wang", "Ting Cao", "Yunxin Liu" ]
[ "cs.RO", "cs.AI" ]
[ "Computer Science" ]
2026-03-15T00:00:00
https://arxiv.org/abs/2603.14371
https://arxiv.org/pdf/2603.14371v2
2603.14371
null
1
0
false
null
null
0.3259
c1302e1af1fe9142aad81a012d3adea61c5891458cc967ac2f3d8f1ae9876018
[ "arxiv", "semantic_scholar" ]
Flood Risk Follows Valleys, Not Grids: Graph Neural Networks for Flash Flood Susceptibility Mapping in Himachal Pradesh with Conformal Uncertainty Quantification
Flash floods are the most destructive natural hazard in Himachal Pradesh (HP), India, causing over 400 fatalities and $1.2 billion in losses in the 2023 monsoon season alone. Existing risk maps treat every pixel independently, ignoring the basic fact that flooding upstream raises risk downstream. We address this with a...
[ "Paras Sharma", "Swastika Sharma" ]
[ "cs.LG" ]
[ "Computer Science" ]
2026-03-15T00:00:00
https://arxiv.org/abs/2603.15681
https://arxiv.org/pdf/2603.15681v1
2603.15681
null
0
0
true
https://github.com/Parassharmaa/flash-flood-zones-hp
null
0.6053
c75f50d846f9cb1c4b64eab5fa7875ca11c39204effdbbf0d0f2325a0f6ae2cb
[ "arxiv", "semantic_scholar" ]
SemantiCache: Efficient KV Cache Compression via Semantic Chunking and Clustered Merging
Existing KV cache compression methods generally operate on discrete tokens or non-semantic chunks. However, such approaches often lead to semantic fragmentation, where linguistically coherent units are disrupted, causing irreversible information loss and degradation in model performance. To address this, we introduce S...
[ "Shunlong Wu", "Hai Lin", "Shaoshen Chen", "Tingwei Lu", "Yongqin Zeng", "Shaoxiong Zhan", "Hai-Tao Zheng", "Hong-Gee Kim" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-03-15T00:00:00
https://arxiv.org/abs/2603.14303
https://arxiv.org/pdf/2603.14303v1
2603.14303
10.1109/icassp55912.2026.11464823
0
0
false
null
IEEE International Conference on Acoustics, Speech, and Signal Processing
0.5122
8a35ce815134b0eb7c9379640f1b13c463bdea33e44353a7b9292700326225fd
[ "arxiv", "semantic_scholar" ]
LightMoE: Reducing Mixture-of-Experts Redundancy through Expert Replacing
Mixture-of-Experts (MoE) based Large Language Models (LLMs) have demonstrated impressive performance and computational efficiency. However, their deployment is often constrained by substantial memory demands, primarily due to the need to load numerous expert modules. While existing expert compression techniques like pr...
[ "Jiawei Hao", "Zhiwei Hao", "Jianyuan Guo", "Li Shen", "Yong Luo", "Han Hu", "Dan Zeng" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-13T00:00:00
https://arxiv.org/abs/2603.12645
https://arxiv.org/pdf/2603.12645v1
2603.12645
null
1
0
false
null
null
0.3245
c5681253aeca3c3526727669d7ccd97b59709a5a19dde3ca709154d43b176ad1
[ "arxiv", "semantic_scholar" ]
MoEKD: Mixture-of-Experts Knowledge Distillation for Robust and High-Performing Compressed Code Models
Large language models for code have achieved strong performance across diverse software analytics tasks, yet their real-world adoption remains limited by high computational demands, slow inference speeds, significant energy consumption, and environmental impact. Knowledge distillation (KD) offers a practical solution b...
[ "Md. Abdul Awal", "Mrigank Rochan", "Chanchal K. Roy" ]
[ "cs.SE" ]
[ "Computer Science" ]
2026-03-13T00:00:00
https://arxiv.org/abs/2603.13213
https://arxiv.org/pdf/2603.13213v1
2603.13213
null
1
0
false
null
null
0.3245
62afd323071f0d51198d4e800fa9cd40bb245d855841b80f2031e4e146287b86
[ "arxiv", "semantic_scholar" ]
LongFlow: Efficient KV Cache Compression for Reasoning Models
Recent reasoning models such as OpenAI-o1 and DeepSeek-R1 have shown strong performance on complex tasks including mathematical reasoning and code generation. However, this performance gain comes with substantially longer output sequences, leading to significantly increased deployment costs. In particular, long outputs...
[ "Yi Su", "Zhenxu Tian", "Dan Qiao", "Yuechi Zhou", "Juntao Li", "Min Zhang" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2026-03-12T00:00:00
https://arxiv.org/abs/2603.11504
https://arxiv.org/pdf/2603.11504v2
2603.11504
null
1
0
false
null
null
0.3237
6aff84c78c181251f6eeeb77ad0b6b4ece78cbc5be7da07d040fa99d080eedad
[ "arxiv", "semantic_scholar" ]
Accelerating Suffix Jailbreak attacks with Prefix-Shared KV-cache
Suffix jailbreak attacks serve as a systematic method for red-teaming Large Language Models (LLMs) but suffer from prohibitive computational costs, as a large number of candidate suffixes need to be evaluated before identifying a jailbreak suffix. This paper presents Prefix-Shared KV Cache (PSKV), a plug-and-play infer...
[ "Xinhai Wang", "Shaopeng Fu", "Shu Yang", "Liangyu Wang", "Tianhang Zheng", "Di Wang" ]
[ "cs.CR", "cs.AI" ]
[ "Computer Science" ]
2026-03-12T00:00:00
https://arxiv.org/abs/2603.13420
https://arxiv.org/pdf/2603.13420v2
2603.13420
null
0
0
false
null
null
0.3237
5408c2ea0d17de6c9257d3dbc5f23288f22735208c6ff399b70106d19995cfff
[ "arxiv", "semantic_scholar" ]
Where Matters More Than What: Decoding-aligned KV Cache Compression via Position-aware Pseudo Queries
The Key-Value (KV) cache is crucial for efficient Large Language Models (LLMs) inference, but excessively long contexts drastically increase KV cache memory footprint. Existing KV cache compression methods typically rely on input-side attention patterns within a prompt observation window to estimate token importance du...
[ "Zhenxu Tian", "Yi Su", "Juntao Li", "Min Zhang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2026-03-12T00:00:00
https://arxiv.org/abs/2603.11564
https://arxiv.org/pdf/2603.11564v1
2603.11564
null
1
0
false
null
null
0.3237
bf0f16d8e660f4559def9b285d5c616764e7222fa47ee96e3ccdf52716efda04
[ "arxiv", "semantic_scholar" ]
Task-Conditioned Routing Signatures in Sparse Mixture-of-Experts Transformers
Sparse Mixture-of-Experts (MoE) architectures enable efficient scaling of large language models through conditional computation, yet the routing mechanisms responsible for expert selection remain poorly understood. In this work, we introduce routing signatures, a vector representation summarizing expert activation patt...
[ "Mynampati Sri Ranganadha Avinash" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-11T00:00:00
https://arxiv.org/abs/2603.11114
https://arxiv.org/pdf/2603.11114v1
2603.11114
null
0
0
false
null
null
0.323
c9b6dafdbad61df80fa18fcf60b0597295f43516d54b8601ce8d673842787915
[ "arxiv", "semantic_scholar" ]
Optimal Expert-Attention Allocation in Mixture-of-Experts: A Scalable Law for Dynamic Model Design
This paper presents a novel extension of neural scaling laws to Mixture-of-Experts (MoE) models, focusing on the optimal allocation of compute between expert and attention sub-layers. As MoE architectures have emerged as an efficient method for scaling model capacity without proportionally increasing computation, deter...
[ "Junzhuo Li", "Peijie Jiang", "Changxin Tian", "Jia Liu", "Zhiqiang Zhang", "Xuming Hu" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-11T00:00:00
https://arxiv.org/abs/2603.10379
https://arxiv.org/pdf/2603.10379v1
2603.10379
null
0
0
false
null
null
0.323
004c0e3045d469626b727ab0354521a734e9eb15fba6a20724f141dc0a0976b0
[ "arxiv", "semantic_scholar" ]
LookaheadKV: Fast and Accurate KV Cache Eviction by Glimpsing into the Future without Generation
Transformer-based large language models (LLMs) rely on key-value (KV) caching to avoid redundant computation during autoregressive inference. While this mechanism greatly improves efficiency, the cache size grows linearly with the input sequence length, quickly becoming a bottleneck for long-context tasks. Existing sol...
[ "Jinwoo Ahn", "Ingyu Seong", "Akhil Kedia", "Junhan Kim", "Hyemi Jang", "Kangwook Lee", "Yongkweon Jeon" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2026-03-11T00:00:00
https://arxiv.org/abs/2603.10899
https://arxiv.org/pdf/2603.10899v1
2603.10899
null
2
1
true
https://github.com/SamsungLabs/LookaheadKV
null
0.5999