2026-09-18
On-Demand Attention: Language Models Know When to Recall
针对长上下文推理中全注意力解码效率低的问题，提出按需注意力机制，选择性调用全局注意力以提升效率。
Authors: Haibo Feng, Ruiqi Liang, Hanyang Peng, Shiqi Yu | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.20734v1
Abstract
Reasoning and agentic workloads increasingly demand efficient long-context inference. Yet full-attention decoding reads the growing history at every step, regardless of its benefit to the next prediction. We show that a pretrained model's decoding states already contain information predictive of this benefit, before the global read. Building on this finding, we introduce On-Demand Attention (ODA), a local-first decoding method that uses a lightweight recall head to selectively invoke global attention as its predicted benefit changes during generation. ODA trains only the recall head, leaving pretrained weights unchanged and the complete historical KV cache available for future recall. We further implement GPU-side conditional execution in vLLM, translating reduced global reads into practical decoding speedups over full attention at long context lengths. Experiments across Qwen and Gemma models, including hybrid-attention backbones, show that selective recall recovers most of the performance lost under local attention while substantially reducing global reads. These findings support long-context inference in which pretrained models guide their own access to the information they retain.Relational Attention for Data-Efficient Language Modeling
针对数据高效语言建模，提出关系注意力机制，分离词汇与结构信息以提升模型性能。
Authors: Adrian Brasoveanu, Ece Takmaz, Jakub Dotlačil | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.20530v1
Abstract
We present Relational BabyLM, a system submission to the BabyLM 2026 challenge that combines two cognitively motivated inductive biases in a single decoder-only Transformer. Architecturally, we replace standard self-attention with a Dual Attention Transformer (DAT), which separates the routing of object-level ("sensory") lexical features from structural/relational information (Altabaa and Lafferty, 2025; Altabaa et al., 2024; Webb et al., 2024; Kerg et al., 2022; Webb et al., 2021). Relational attention (RA) disentangled from self-attention greatly increases data efficiency and out-of-training-sample generalization on purely relational tasks, but language modeling requires object-level and relational information to be integrated as well as disentangled, and RA-based LMs have remained largely unexplored. BabyLM's data-constrained training and comprehensive evaluation is an ideal testing ground for whether that data efficiency transfers. As a training intervention, we add a Next-Latent Prediction (NextLat; Teoh et al. 2026) objective that encourages hidden states to compress history incrementally into a dense belief state. Architecture is the dominant factor for structural linguistic generalization; the objective is secondary but still significant. DAT's three relational attention types (full RA vs. the simpler RCA and DisRCA variants) are largely interchangeable at 10M words; full RA pulls ahead at 100M. We also introduce a novel symbol-retrieval mechanism (RoPE-based, as opposed to learned, relative symbols) that matches learned symbol libraries while adding no parameters. On the strict (100M-word) track, our best model ranks 6th of 55 overall and 3rd of 55 on the leaderboard's NLP-task subset at the time of writing; our two strongest models outperform the GPT-2 baseline on most benchmarks, with one attaining the highest EWoK score among strict-track entries.To Copy or Not to Copy: Controlling Speculative Decoding via Intrinsic Model Signals
针对推测解码中神经草稿与上下文复制策略的权衡问题，提出通过内在模型信号控制策略选择。
Authors: Roy Eisenstadt, Ido Cohen, Edo Cohen-Karlik, Lior Wolf, Itamar Zimerman | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.20186v1
Abstract
Speculative Decoding (SD) has significantly accelerated Large Language Model (LLM) inference, yet existing approaches face a fundamental tradeoff between two drafting strategies: neural drafting and context-based copying. Neural drafts (e.g., EAGLE3) provide robust performance across diverse text settings, while copy-based methods achieve higher speedups in copy-intensive regimes by generating candidates faster and exploiting long repetition spans for near-perfect speculation. We analyze existing copy-based methods and find that they are prone to accidental repetitions where surface-level n-gram overlap does not reflect a structural intent to copy, leading to false-positive triggers that ultimately degrade throughput. We introduce SwitchSD, an adaptive framework that treats copying as a latent control signal of the LLM. By training lightweight probes on the target model's internal representations, SwitchSD identifies genuine copy-intent with high precision (AUC > 0.99). This allows the system to dynamically switch between neural drafting (e.g., EAGLE) and context-based copying. Our results across Llama and Qwen families demonstrate throughput gains of up to 15% over state-of-the-art baselines like EAGLE3, effectively turning copying from a noisy heuristic into a principled, model-aware decoding regime.Local Sparsity Enables Unsupervised LLM Safety Detection
针对LLM部署时安全检测依赖监督数据的问题，提出基于局部稀疏性的无监督异常检测方法。
Authors: Xin Chen, Gil Kur, Alexander Shevchenko, Andreas Krause | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.20129v1
Abstract
Deployment-time safety methods for large language models (LLMs) are predominantly supervised and assume access to unsafe training data. Nevertheless, new attacks and harm categories regularly arise, not captured by models trained in such a supervised fashion. An alternative approach is to view this problem through the lens of anomaly detection, namely, to rely solely on modeling safe data and flagging out-of-distribution inputs. However, LLM activations lie in a high-dimensional space, raising concerns about whether anomaly detection is statistically feasible. We show that, under the linear representation hypothesis (LRH), there may indeed be hope. In the LRH concept space, which is typically recovered via a sparse autoencoder (SAE), nearby points share a small common active support. Using this local sparsity insight, we propose a framework for locally masked SAE-based anomaly detection, supported by theoretical justifications. We validate it on various architectures and datasets, including both capability-testing datasets and safety-specific datasets. Finally, when we allow algorithms to use 1% out-of-distribution data for calibration, locally sparse methods achieve near-optimal performance, demonstrating their ability to capture meaningful safety information while using only 1-2% of SAE neurons for computation.Marginal utility, matrix factorization, and the Key-Value (KV) cache: a unified information-economic framework for sovereign geo-mining inference
针对KV缓存优化，提出统一信息经济框架，将边际效用理论应用于缓存压缩与管理。
Authors: Caroline Gans Combe | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.20068v1
Abstract
This paper builds a theoretical bridge between the economic notion of marginal utility and two machine-learning constructs, matrix factorization and the Key--Value cache of transformer language models. The singular value spectrum of a rating matrix is shown to be a diminishing marginal utility schedule for latent factors, the eigenvalue spectrum of the projected covariance operator to be the marginal utility schedule of a model's learned representation, and cache eviction and low-rank cache compression to be instances of constrained utility maximization under a memory budget. The three collapse into a single allocation rule: retain the top dimensions whose eigenvalue exceeds the shadow price of the binding constraint. The framework is applied to the automated extraction of structured information from geo-mining documents, where it motivates a multi-pass inference protocol, a layer-wise TIES model merging procedure, and a selection policy combining extraction quality, localization drift and energy, scalarized with a Conditional Value-at-Risk term on drift. Two empirical contributions are reported. An 11.2-million-parameter hierarchical classifier, trained in about five minutes on a single GPU, reaches 90.0 per cent level-1 accuracy on a held-out test set from a 973-document uranium-exploration corpus, against 92.0 per cent for a proprietary model on a fifty-document human audit of the same corpus, at a latency of 2.62 ms per card against approximately 2,000 ms for the API and at negligible cost. A diagnostic of uniform-density TIES merging exposes a reproducible degenerate mode in which the merged model returns token-identical outputs across five geographically distinct districts while declaring high confidence; re-executing the merge under layer-wise calibrated densities removes that signature on the diagnostic sample. The full-scale extraction benchmark, including LoRA fine-tuning, is reported as projected rather than measured and remains an empirical extension of this work.DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression
针对长上下文推理中KV缓存内存与带宽瓶颈，提出DeepSeek-V4.1-Flash以极限压缩KV缓存。
Authors: DeepSeek-AI, :, Anyi Xu, B. Li, Bangcai Lin, Bing Xue, BingCheng Xian, Bingzheng Xu, Bochao Wu, Bowei Zhang, Boyi Deng, C. C. Yu, Chao Jin, Chaofan Lin, Chen Dong, Chenbing Wang, Chenfan Feng, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chengyuan Zhang, Chenhao Xu, Chenqi Zhao, Chenze Shao, Chuhao Wang, Chuqi Zhang, Damai Dai, Dejian Yang, Deli Chen, Di Huang, Di Wu, Donghao Li, Erhang Li, Eric Fu, F. Zhou, Fangwei Zhou, Fangyun Lin, Fangzhou Yuan, Feiyu Xia, Fucong Dai, Guangbo Hao, Guanglin Li, Guanting Chen, Guoai Cao, Guofan Fan, Guolai Meng, Guowei Li, Haichuan Zhang, Haiyang Ma, Haiyang Shen, Han Li, Han Yu, Han Zhang, Hangyuan Deng, Hanwei Xu, Hanxiang Xu, Hanxun Zhong, Hao Guo, Hao Jiang, Hao Li, Hao Qin, Haodong Wen, Haofen Liang, Haofeng Huang, Haohua Liu, Haoling Zhang, Haoming Luo, Haoran Yang, Haotian Xu, Haotian Yuan, Haoting Huang, Haowen Luo, Haoyang Cai, Haoyu Chen, Haozhe Ji, Hengran Zhang, Hengrui Wang, Hengxu Wu, Honghui Ding, Hongxuan Tang, Huadong Wang, Huanqi Cao, Huazuo Gao, Hui Qu, Hui Zeng, J. Yang, J. H. Jin, J. H. Zhang, J. X. Zou, Jia Yu, Jiahui Zhou, Jiajun Chen, Jialiang Huang, Jialin Zhao, Jiamin Tang, Jian Zhou, Jianan Tong, Jianwen Li, Jiaqi Zhu, Jiarui Wang, Jiasheng Ye, Jiashi Li, Jiaxin Xu, Jiaying Ding, Jibai Lu, Jiewen Hu, Jin Yan, Jincheng Zhai, Jingchang Chen, Jingcheng Hu, Jingli Zhou, Jingsheng Xu, Jingting Xiang, Jingyan Yun, Jingyang Yuan, Jingyuan Cheng, Jinhua Zhu, Jinpeng Wang, Jinyi Chen, Jinyi Hu, Jiping Yu, Jueliang Guo, Junbo Pei, Junbo Sun, Junguang Jiang, Junjie Qiu, Junkang Zhou, Junqi Liu, Junren Li, Junxian Li, Junxiao Song, Junyi Guo, Kai Dong, Kaifeng Chen, Kaige Gao, Kang Guan, Kangdong Yuan, Ke Hong, Ke Xu, Kefan Zhao, Kexin Ji, Kexin Zhang, Kexing Zhou, Kuai Yu, Lan Zhang, Lean Wang, Lecong Zhang, Lei Wang, Letian Gao, Liang Zhao, Liansheng Xu, Lihua Guo, Lingxiao Luo, Lingyue Fu, Litao Deng, Litong Wang, Liyue Zhang, Longhao Chen, Lu Chen, Luotian Huang, Luyao Ma, Luyao Wang, M. S. Di, Max Mei, Menghao Ye, Miao Cui, Mingchuan Zhang, Minghua Zhang, Minghui Tang, Mingjing Zhang, Mingqi Wei, Mingshu Chen, Mingxing Liu, Mingxu Zhou, Mingyu Xu, Mingyu Yang, Mingze Wang, Muyang Chen, Ni Shentu, Ning Wang, Niufang Ning, Panpan Huang, Peixin Cong, Peiyi Wang, Peiyuan Xin, Pengfei Ren, Pengfei Yan, Pengle Zhang, Qi Kang, Qi Tang, Qiancheng Wang, Qiang Li, Qihao Zhu, Qingyang Li, Qinyu Chen, Qiushi Du, Qizhou Guo, Rongxian Xu, Rui Ding, Rui Hu, Rui Tian, Rui Yu, Ruidong Zhu, Ruifan Xu, Ruihan Yang, Ruihang Xia, Ruijie Lu, Ruilin Geng, Ruipeng Hong, Ruiqi Ge, Ruisong Zhang, Ruize Sun, Ruizhe Pan, Runji Wang, Runqian Chen, Runxin Xu, Ruohong Tian, Ruomeng Shen, Ruoyu Zhang, Ryan X., S. H. Liu, Shanghao Lu, Shangyan Zhou, Shanhuang Chen, Shaofei Cai, Shaoheng Nie, Shaoyuan Chen, Shengding Hu, Shengkai Lin, Shengwen Ran, Shengyu Liu, Shengyuan Jia, Shi Bai, Shi Feng, Shicheng Xu, Shichun Liu, Shiqiang Hu, Shirong Ma, Shiyu Wang, Shiyuan Feng, Shufan Gong, Shuhan Lin, Shuiping Yu, Shunfeng Zhou, Shuo Yang, Shuomeng Wang, Shuting Guo, Shuting Pan, Shuying Yu, Sinuo Cao, Siyi Lin, Sizhe Chen, Songyang Chen, Songyang Zhou, Tao Ni, Tao Yun, Tian Jin, Tian Pei, Tian Ye, Tianle Lin, Tianran Ji, Tianyi Cui, Tianyuan Yue, Tingting Yu, Tongrui Xiong, Wangding Zeng, Wei Liu, Wei Zhang, Weibin Xu, Weihao Zeng, Weilin Zhao, Wen Liu, Wenfeng Liang, Wenjie Pang, Wenjing Luo, Wenjing Yao, Wenjun Gao, Wenkai Shao, Wenkai Yang, Wenli Zhang, Wenlu Wang, Wenlve Huang, Wenqian Yan, Wentao Zhang, Xi Gao, Xiang He, Xiang Li, Xiangli Li, Xiangwen Wang, Xiangying Zhang, Xiankui Wei, Xiao Bi, Xiaodong Liu, Xiaohan Wang, Xiaojian Qu, Xiaokang Chen, Xiaokang Zhang, Xiaotao Nie, Xiaoyao Zou, Xiaoyuan Li, Xicheng Guo, Xieting Chu, Xin Cheng, Xin Liu, Xin Xie, Xinbo Xu, Xingchao Liu, Xingchen Liu, Xingkai Yu, Xingyou Li, Xintong Yao, Xinyang Chen, Xinyong Jiang, Xinyu Yang, Xinyu Yang, Xu Chen, Xuanyu Wang, Xubei Zhong, Xuecheng Su, Xuejie Liu, Xuheng Lin, Xujie Fan, Xuncheng Zhao, Xuwei Fu, Y. C. Yan, Y. H. Jiang, Y. T. Wu, Y. W. M., Y. Z. Wang, Yafei Gao, Yang Yang, Yang Zhang, Yanru Ma, Yanwen Huang, Yao Li, Yao Li, Yao Meng, Yao Zhao, Yaofeng Sun, Yaohui Wang, Yaoyang Ye, Yehang Yin, Yexinrui Wu, Yi Qian, Yi Tao, Yi Yu, Yichao Zhang, Yichen Jiang, Yicheng Wang, Yifan Ding, Yifan Shi, Yifeng Peng, Yifeng Zhai, Yijia Wu, Yiliang Xiong, Yilun Wang, Ying He, Ying Zhou, Yingjia Luo, Yinmin Zhong, Yiping Wang, Yisong Wang, Yixiang Zhang, Yixiao Chen, Yixuan Tan, Yixuan Wei, Yiyang Ma, Yiyao Yang, Yiyuan Liu, Yizai Cai, Yizhen Wei, Yizhi Wang, Yonglun Yang, Yongqi Zhuo, Yongqiang Guo, Yongtong Wu, Yu Wu, Yu Zhang, Yuan Bian, Yuan Cheng, Yuan Ou, Yuan Sun, Yuanfan Xu, Yuanhang Sun, Yuanhao Li, Yuchen Liu, Yuchen Yao, Yudong Han, Yuduan Wang, Yuhan Wu, Yuhao Meng, Yuheng Zou, YuKun Li, Yunchuan Wang, Yunfan Xiao, Yunfan Xiong, Yupeng Chen, Yuqian Cao, Yuqian Wang, Yuqing Chen, Yushun Zhang, Yutong Lin, Yuwei Xiao, Yuxian Gu, Yuxiang Chen, Yuxiang Huang, Yuxiang Luo, Yuxiang You, Yuxin Chen, Yuxin Xiang, Yuxuan Liu, Yuxuan Zhou, Yuyang Zhou, Yuzhe Guo, Yuzhen Huang, Yuzhuo Bai, Z. Y. Z., Zanlin Ni, Zehao Wang, Zehua Zhao, Zehui Ren, Zejun Zhao, Zhangli Sha, Zhanying Wang, Zhaochen Zhang, Zhaoshuai Du, Zhe Fu, Zhean Xu, Zhenda Xie, Zheng Liu, Zhengyan Zhang, Zhenhua Dong, Zhewen Hao, Zhibang Wang, Zhibin Gou, Zhicheng Ma, Zhihao Li, Zhihong Shao, Zhihuan Huang, Zhijie Li, Zhirui Lu, Zhixian Huang, Zhixuan Chen, Zhixuan Chen, Zhixuan Pan, Zhiyu Wu, Zhizhou Ren, Zhu He, Zhuoshu Li, Zhuping Zhang, Zian Xu, Zihao Wang, Zihui Gu, Zijia Zhu, Zili Zhang, Zilin Li, Zilong Hou, Zilong Lyu, Ziqiao Wang, Ziwei Xie, Ziya Zhang, Ziyi Gao, Zizheng Pan, Zonglin Li, Zongqing Yao, Zui Chen, Zuofan Wu, Chenchen Ling, Chengyu Hou, Chong Chen, D. Li, Di Qi, Dongjie Ji, Fang Wei, Fanyi Xia, Fei Xie, Feiyi Tan, Hailong Guo, Haiyan Zhai, Hui Zhou, Huihui Tan, Huijie Li, Jia Luo, Jia Song, Jialu Cai, Jian Liang, Jiangting Zhou, Jiaqi Gao, Jiayi Shao, Jie Chen, Jieyu Yang, Jin Chen, Jingde Zhang, Jingzi Zhou, Jinqian Wang, Jinyang Liu, JinZhao Sun, Junhua Ling, Junmin Zheng, Kaicheng Yang, Ke Xu, Le Su, Leyi Xia, Liangfeng Ding, Lin Zhuo, Linwang Ma, Linyan Zhu, Liyu Cai, Luqi Yao, M. K. Zhang, Meng Li, Miao Lin, Miaojun Wang, Min Zhang, Mingming Li, Mingming Wang, Mingze Yin, Minmin Han, Nan Cao, Ning Wang, Ningxin Ma, Panpan Wang, Peihan Lin, Peng Sun, Peng Zhang, Qian Ying, Qiang Xiang, Qiao Wang, Qingmiao Mao, Qiwei Jiang, Rongli Jin, Ruyi Chen, Sha Tao, Shangmian Sun, Shaoqing Wu, Shichao Zou, Si Lei, Tianyang Zhang, Tianyu Sun, Tingting Yin, W. L. Xiao, Wei An, Wei Li, Wei Wang, Weiwei Lin, Wenqing Hou, X. Lin, Xiangfei Meng, Xianzhu Huang, Xiao Peng, Xiaoqian Li, Xiaoting Zhang, Xiaowen Sun, Xiaoxiang Wang, Xiaoyu Ye, Xinrou Zhang, Xinyu Zhang, Xue Cao, Xueyin Chen, Yanan Zhou, Yanhong Xu, Yao Xia, Yao Xu, Yi Shao, Yihong Zhang, Yiling Ma, Ying Tang, Yining Lou, Yiru Chen, Yishi Piao, Yixuan Chen, Yong Xiong, Yuchen Xuan, Yuehan Yang, Yuer Xu, Yukun Zha, Yunxian Ma, Yuping Lin, Yuting Yan, Yutong Xie, Yuwen Sheng, Yuxuan Zhu, Zekai Zhang, Zhe Ju, Zhenzhen Lin, Zheren Gao, Zheyang Sun, Zhigang Yan, Zhongyu Wu, Zi Wang, Zihua Qu, Ziling Yan, Ziyi Wan | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.19969v1
Abstract
The widespread adoption of long-horizon agents has made model workloads increasingly input-heavy. Although prior work has substantially reduced the cost of long-context computation, prefill remains computationally expensive, and large KV caches continue to strain HBM and SSD capacity and data-transfer bandwidth. Together, these compute, storage, and bandwidth demands constitute the primary bottleneck to further lowering deployment costs. To address this challenge, we introduce DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts (MoE) model with 552B backbone parameters and support for contexts of up to one million tokens. With its Causal Encoder-Decoder (CED) architecture, the model activates 16B parameters per token during decode but only 8B parameters during prefill, substantially improving cost efficiency for agentic workloads. To push the limits of KV cache compression, DeepSeek-V4.1-Flash combines cross-layer KV cache reuse in Compressed Sparse Attention 2 (CSA2) with FP4 KV caching. These designs reduce its global KV cache footprint (always in HBM) to 890 bytes per token, roughly 1/4 of the corresponding footprint of DeepSeek-V4-Flash. Further, through a dedicated deployment optimization known as SWA Bounded Replay, DeepSeek-V4.1-Flash reduces its persistent KV cache footprint (always on SSD or in host memory) to roughly 1/8 of that of DeepSeek-V4-Flash. Despite its much smaller KV cache footprint, the model delivers substantially better performance than the baseline. In addition, we streamline the DeepSeek-V4 architecture and introduce several efficient architectural extensions. We pretrain DeepSeek-V4.1-Flash on a multimodal corpus comprising 45T tokens and conduct comprehensive post-training, yielding strong performance across diverse text-based and multimodal agentic scenarios. Model checkpoints are available at https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash.Xronos: Heterogeneity-Aware Tensor Parallelism for Collaborative LLM Fine-Tuning on Edge CPUs
针对边缘CPU设备协同微调效率低的问题，提出异构感知张量并行方法以优化性能。
Authors: Wonmi Choi, Sunjae Park, Dohyeok Kwon, Zhixiong Niu, Yeonho Yoo, Chuck Yoo, Gyeongsik Yang | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.19909v1
Abstract
Collaborative fine-tuning on edge devices adapts large language models to domain-specific data while keeping each device's data local. State-of-the-art (SOTA) collaborative fine-tuning techniques are largely designed for GPU-based edge devices and rely on pipeline parallelism (PP). However, many edge platforms, including IoT gateways, smart-home hubs, and in-vehicle computers, are primarily CPU-based. This paper reports that PP is ineffective on CPU-based edge devices because the same CPU handles both model computation and communication, which causes severe CPU contention. Our analysis shows that this leads to 5.75$\times$ higher computation stall ratios than on GPU devices on average. Tensor parallelism (TP) can alleviate this contention by separating computation and communication, but existing TP techniques assume homogeneous devices. On heterogeneous CPU edge devices, we find that this assumption causes faster workers to remain idle for up to 34% while waiting for slower devices at synchronization points. To address the limitations, we propose Xronos, a collaborative fine-tuning framework for heterogeneous CPU edge devices. Xronos uses TP as its execution backbone and combines lightweight profiling with heterogeneity-aware tensor partitioning to reduce the straggler bottleneck. Across diverse devices, models, and benchmark tasks, Xronos reduces fine-tuning time by 18% (TP) to 56% (PP) and the ratio of device idle time by $\sim$5.9$\times$ over SOTA techniques, while maintaining the accuracy.D-Quant: Driftable Entropy Coding for KV Cache Quantization
针对KV缓存量化中固定位宽限制动态范围的问题，提出可漂移熵编码以提升压缩效果。
Authors: Yi Su, Hong Liu, Guanghua Yu, Jianchen Zhu | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.19880v1
Abstract
The KV cache has become a major bottleneck in deploying LLMs, as its memory footprint grows linearly with sequence length and batch size, imposing substantial pressure on both memory capacity and bandwidth. Among various KV cache compression techniques, quantization is particularly attractive due to its effectiveness and ease of deployment. However, most existing methods rely on fixed-width quantization, where a $b$ bit representation is inherently limited to $2^b$ quantization levels. As the bit width decreases, the number of available levels shrinks exponentially, leading to severe information loss and rapid performance degradation. We further observe that fixed-width quantization fails to exploit the highly non-uniform distribution of KV cache. After rotation and normalization, KV values approximately follow a normal distribution, with most values concentrated near the center and only a small fraction appearing in the tails. Nevertheless, fixed-width coding allocates the same number of bits to frequent and rare symbols. Entropy coding naturally exploits such non-uniformity by assigning shorter codewords to frequent symbols and longer ones to rare symbols, substantially reducing the average number of bits required for representation. However, its variable-length output is not suited to highly parallel attention kernels, where efficient dequantization and computation rely on regular memory layouts and fixed-stride accesses. To bridge this gap, we propose \textbf{D-Quant}, a flexible KV cache quantization framework that introduces a \textbf{drift} mechanism to convert entropy-coded representations of each token into fixed-size bitstreams, enabling regular memory access and parallel dequantization within attention kernels.MiX: Micro-Inverted-Scaling for End-to-End Low-Bit Vision-Language Model Acceleration
针对视觉语言模型在边缘设备部署的内存带宽瓶颈，提出微反缩放量化方法以实现端到端低比特加速。
Authors: Yuan Liao, Jae-sun Seo | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.19683v1
Abstract
The deployment of Vision-Language Models (VLMs) on edge devices is severely bottlenecked by memory bandwidth, necessitating aggressive sub-8-bit quantization. Since edge accelerators are strictly constrained by area and power, they require end-to-end quantized models. However, the extreme dynamic range gap between multi-modal tokens causes standard block formats to suffer "microscaling collapse," where a single massive outlier hijacks the shared exponent, underflowing surrounding elements and destroying attention maps. To break this bottleneck, we propose Micro-Inverted-Scaling (MiX), a novel format that mathematically inverts the microscaling paradigm: rather than grouping multiple mantissas under one shared exponent, MiX groups private, per-element exponents under a single shared mantissa. To handle asymmetric VLM outlier topologies, we introduce an adaptive dual-format (MiX-MX) inference framework. By algebraically factoring out the shared MiX mantissa, this framework maps to a custom accelerator, replacing multipliers with efficient shifters. Evaluated end-to-end on multiple VLMs, our 4.5-bit MiX formulation exhibits equivalent or superior accuracy on multi-modal benchmarks compared to NVFP4. Simultaneously, the MiX accelerator delivers a 25% improvement in area efficiency over the NVFP4 baseline and a 2.3-4.5x speedup with 1.4-2.9x energy reduction across models compared to the state-of-the-art accelerator Focus, proving the inverted-scaling datapath is physically superior for efficient VLM deployment.PrefixBench-H100: Characterizing Prefix Reuse and Time-to-First-Token in H100 LLM Serving
针对LLM服务中前缀重用性能评估不足的问题，提出基准测试以表征前缀重用对首token时间的影响。
Authors: Omkar Shewale, Deepak Kumar, Divakar Kumar Yadav | Date: 2026-09-17
Link: http://arxiv.org/abs/2609.19657v1
Abstract
Repeated prompt prefixes are increasingly common in LLM serving workloads, appearing in system prompts, templated retrieval-augmented generation pipelines, agent frameworks, and multi-turn conversations. Modern inference runtimes such as vLLM and TensorRT-LLM provide mechanisms for reusing previously computed KV-cache state across requests, yet it remains unclear when prefix reuse materially improves serving performance on contemporary accelerators and when its benefits are limited by scheduling, cache granularity, concurrency, or memory pressure. This paper presents PrefixBench-H100, a reproducible benchmark and measurement framework for characterizing prefix reuse on a single NVIDIA H100. PrefixBench-H100 combines controlled synthetic traces with chat-style and retrieval-style workloads, and evaluates two widely used LLM serving runtimes under matched workload conditions. The benchmark varies shared-prefix length, suffix diversity, request arrival pattern, concurrency, output length, and cache configuration, while collecting time-to-first-token, inter-token latency, end-to-end latency, throughput, cache-hit statistics, GPU memory usage, and selected profiling traces. The goal of PrefixBench-H100 is not to introduce a new caching algorithm, but to expose the practical operating envelope of prefix reuse for H100-class LLM serving. The study identifies the regime where prefix reuse provides substantial first-token latency reductions and the regime where cache pressure erodes them, while showing that cache effectiveness itself is largely insensitive to concurrency and output length; the cross-runtime differences that remain arise above the cache, in the scheduling layer.Higher-order pruning of experts in mixture-of-experts language models
针对混合专家模型参数量大导致内存瓶颈的问题，提出高阶专家剪枝方法以协同优化专家贡献。
Authors: Alex M. Tseng, Prannay Kaul, Luca Zancato, Wei Xia, Stefano Soatto | Date: 2026-09-16
Link: http://arxiv.org/abs/2609.18916v1
Abstract
Mixture-of-Experts (MoE) language models suffer from large parameter counts, which create a significant memory bottleneck. Expert pruning is the most direct approach for reducing this parameter count, yet existing methods make pruning decisions for each expert independently, and assume experts' contributions are purely additive. In reality, expert usage in MoEs is inherently cooperative. We derive HOPE (Higher-Order Pruning of Experts), a second-order pruning objective which provably minimizes an upper bound on the error resulting from pruning. We show that REAP (a state-of-the-art first-order pruning method) is a special case of HOPE where interaction terms are ignored. Across three frontier MoE models (up to 122B parameters), two distinct calibration sets, and multiple benchmarks (including math, instruction following, coding, and an agentic suite), we demonstrate that HOPE produces better pruning decisions than existing methods, and its advantage is most pronounced at high pruning rates and on challenging agentic workloads. At 50% pruning, HOPE outperforms all baselines and achieves an average rank of 1.58 out of 5 methods (versus 2.42 for the next-best method, REAP), with gains of up to +6.1% on agentic coding. Over all conditions, HOPE again achieves the best average rank and surpasses every other method in the majority of head-to-head comparisons. By preserving cooperative expert structure that first-order methods ignore, HOPE enables aggressive compression with minimal degradation, particularly on complex tasks where diverse expert combinations are invoked over long sequences.Quantifying the Effect of HCLs on a Fixed-Microarchitecture MXFP4 Accelerator
针对硬件构建语言评估缺乏统一设计比较的问题，提出基于固定微架构加速器的量化对比方法。
Authors: Daniele Passaretti, Sajjad Tamimi, Nicola Dall'Ora | Date: 2026-09-16
Link: http://arxiv.org/abs/2609.18792v1
Abstract
Hardware Construction Languages (HCLs) aim to improve hardware design productivity while generating register-transfer-level (RTL) circuits without changing the designer's microarchitecture. However, most comparisons between HCLs are either qualitative or evaluate quality of results (QoR) across different designs, making it difficult to separate language effects from design effects. This paper compares the most widely used HCLs using the same fixed design, the OCP MXFP4 block dot product, a quantization primitive at the heart of edge Physical-AI inference, implemented as a single 12-stage, II=1 pipeline. A SystemVerilog baseline is followed by implementations in Chisel, SpinalHDL, Amaranth, Clash, Bluespec, and C++ for high-level synthesis (HLS). Every variant goes through the same flow on the same Artix-7 device set at 100 MhZ, driven by a RISC-V soft core. With the micro-architecture held constant, the comparison is clean: every variant meets timing, and the HCLs match or even undercut hand-written RTL in area. The remaining differences stem not from the algorithm but from how each back end lowers arithmetic, and from a single width choice that silently toggles DSP inference. Unlike HLS, where design decisions are limited to pragmas, the HCLs achieve comparable area and timing. Therefore, the choice comes down to ecosystem fit and interface needs rather than QoR.Beyond Truncation: Rethinking LLM Decoding as Ensemble Pruning
针对LLM解码中候选token选择忽略几何语义关系的问题，提出马氏集成解码框架以优化选择。
Authors: Dunyao Xue, Chengshuo Du, Zhengbo Wang, Wenlin Dai, Cheng Meng | Date: 2026-09-16
Link: http://arxiv.org/abs/2609.18723v1
Abstract
We introduce Mahalanobis-Ensemble Decoding (ME-Decoding), a novel Large Language Model (LLM) decoding framework that frames candidate token selection as ensemble pruning. Existing selection strategies rely predominantly on scalar probabilities, ignoring geometric semantic relationships and causing candidate redundancy. Meanwhile, current geometry-aware methods often require complex optimization or directly reweighting the original token probabilities, leading to significant computational overhead or inference instability. To address this, we formulate decoding as a subset optimization problem using a Mahalanobis distance-driven objective to enhance semantic diversity while preserving high probabilities. Specifically, we dynamically discount redundant generation paths using a token similarity matrix, constructed via an adaptive-bandwidth kernel over token embeddings. We further devise an efficient greedy selection algorithm with near-linear complexity in the candidate size under early stopping, while establishing its theoretical approximation guarantees. This renders ME-Decoding a robust, plug-and-play module with negligible inference overhead. Extensive experiments across diverse reasoning and generation tasks demonstrate that our method consistently achieves strong performance.HBFlex: A Flexible Memory System for Bridging Fine-Grained LLM States and Coarse-Grained HBF Parallel Execution
针对LLM状态与高带宽Flash并行执行不匹配的问题，提出灵活内存系统以桥接细粒度状态与粗粒度并行。
Authors: Shuzhang Zhong, Weikai Xu, Yifan Zhou, Tongbin Zhao, Tenghao Zhao, Yifei Kang, Cunyin Chang, Shu Li, Guangyu Sun, Meng Li | Date: 2026-09-16
Link: http://arxiv.org/abs/2609.18675v1