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
07ba491326287450a2b5b2a096cc34c9cd80454793b44913b10dd0fc4e6c8b9b
[ "arxiv", "semantic_scholar" ]
TokenTiming: A Dynamic Alignment Method for Universal Speculative Decoding Model Pairs
Accelerating the inference of large language models (LLMs) has been a critical challenge in generative AI. Speculative decoding (SD) substantially improves LLM inference efficiency. However, its utility is limited by a fundamental constraint: the draft and target models must share the same vocabulary, thus limiting the...
[ "Sibo Xiao", "Jinyuan Fu", "Zhongle Xie", "Lidan Shou" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-10-17T00:00:00
https://arxiv.org/abs/2510.15545
https://arxiv.org/pdf/2510.15545v4
2510.15545
10.48550/arXiv.2510.15545
0
0
false
null
arXiv.org
0.338
298e73b7622084d7adc69fef8929deb05ecbcba4f87219495486c72d0b961898
[ "arxiv", "semantic_scholar" ]
Accelerating Mobile Language Model via Speculative Decoding and NPU-Coordinated Execution
Performing Retrieval-Augmented Generation (RAG) directly on mobile devices is promising for data privacy and responsiveness but is hindered by the architectural constraints of mobile NPUs. Specifically, current hardware struggles with the variable workloads intrinsic to RAG: the transition between processing extensive ...
[ "Zhiyang Chen", "Daliang Xu", "Haiyang Shen", "Chiheng Lou", "Mengwei Xu", "Shangguang Wang", "Xin Jin", "Yun Ma" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-10-17T00:00:00
https://arxiv.org/abs/2510.15312
https://arxiv.org/pdf/2510.15312v4
2510.15312
10.48550/arXiv.2510.15312
1
0
false
null
arXiv.org
0.338
3eb2c2605a527ff6d7d0ed87ce7cb34d50ab97004508c8102709f00873bf9cbb
[ "arxiv", "semantic_scholar" ]
Mirror Speculative Decoding: Breaking the Serial Barrier in LLM Inference
Speculative decoding accelerates LLM inference by using a draft model to look ahead, but gains are capped by the cost of autoregressive draft generation: increasing draft size elevates acceptance rates but introduces additional latency overhead exacerbating the speed-accuracy tradeoff. Prior methods (Medusa, Hydra, EAG...
[ "Nikhil Bhendawade", "Kumari Nishu", "Arnav Kundu", "Chris Bartels", "Minsik Cho", "Irina Belousova" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-10-15T00:00:00
https://arxiv.org/abs/2510.13161
https://arxiv.org/pdf/2510.13161v2
2510.13161
10.48550/arXiv.2510.13161
1
0
false
null
arXiv.org
0.3357
3612acd84e7b6b937ac87cfba9ca43fd383814c99bbb4b680cc37da8a185387d
[ "arxiv", "semantic_scholar" ]
STAR: Decode-Phase Rescheduling for LLM Inference
Large Language Model (LLM) inference has emerged as a fundamental paradigm, however, variations in output length cause severe workload imbalance in the decode phase, particularly for long-output reasoning tasks. Existing systems, such as PD disaggregation architectures, rely on static prefill-to-decode scheduling, whic...
[ "Zhibin Wang", "Zetao Hong", "Xue Li", "Zibo Wang", "Shipeng Li", "Qingkai Meng", "Qing Wang", "Chengying Huan", "Rong Gu", "Sheng Zhong", "Chen Tian" ]
[ "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2025-10-15T00:00:00
https://arxiv.org/abs/2510.13668
https://arxiv.org/pdf/2510.13668v2
2510.13668
10.1145/3806645.3807813
1
0
false
null
null
0.2136
677bbdf95edb9fb3fc816aad5a69d531d06ba2c3e79d82b452f45b4115b93fb9
[ "arxiv", "semantic_scholar" ]
3-Model Speculative Decoding
Speculative Decoding (SD) accelerates inference in large language models by using a smaller draft model to propose tokens, which are then verified by a larger target model. However, the throughput gains of SD are fundamentally limited by a trade-off between draft model size and token acceptance: smaller draft models ge...
[ "Sanghyun Byun", "Mohanad Odema", "Jung Ick Guack", "Baisub Lee", "Jacob Song", "Woo Seong Chung" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-10-14T00:00:00
https://arxiv.org/abs/2510.12966
https://arxiv.org/pdf/2510.12966v1
2510.12966
10.48550/arXiv.2510.12966
0
0
false
null
arXiv.org
0.3346
03d69efd712c80783e10e157380ccc82ce2725ec8d2c05a5391551d536a31933
[ "arxiv", "semantic_scholar" ]
Efficient LLM Inference over Heterogeneous Edge Networks with Speculative Decoding
Large language model (LLM) inference at the network edge is a promising serving paradigm that leverages distributed edge resources to run inference near users and enhance privacy. Existing edge-based LLM inference systems typically adopt autoregressive decoding (AD), which only generates one token per forward pass. Thi...
[ "Bingjie Zhu", "Zhixiong Chen", "Liqiang Zhao", "Hyundong Shin", "Arumugam Nallanathan" ]
[ "eess.SY" ]
[ "Computer Science", "Engineering" ]
2025-10-13T00:00:00
https://arxiv.org/abs/2510.11331
https://arxiv.org/pdf/2510.11331v1
2510.11331
10.48550/arXiv.2510.11331
6
0
false
null
arXiv.org
0.3334
a0952e40b1e21fa0f8ac6fbd30cf6e1b0f777b2b584f590ee412447baffdf733
[ "arxiv", "semantic_scholar" ]
Conformal Sparsification for Bandwidth-Efficient Edge-Cloud Speculative Decoding
Edge-cloud speculative decoding (SD) accelerates inference by having a cloud-based large language model (LLM) that verifies draft tokens generated by a resource-constrained small language model (SLM) at the edge. A central bottleneck is the limited bandwidth of the edge-cloud link, which necessitates efficient compress...
[ "Payel Bhattacharjee", "Fengwei Tian", "Meiyu Zhong", "Guangyi Zhang", "Osvaldo Simeone", "Ravi Tandon" ]
[ "cs.LG", "cs.AI", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2025-10-11T00:00:00
https://arxiv.org/abs/2510.09942
https://arxiv.org/pdf/2510.09942v1
2510.09942
10.48550/arXiv.2510.09942
2
0
false
null
arXiv.org
0.3311
2f9527dcc8cb49e9c70e043c19e102b010890bb1d59e93d8122eaa4d8c8eca76
[ "arxiv", "semantic_scholar" ]
SP-MoE: Speculative Decoding and Prefetching for Accelerating MoE-based Model Inference
The Mixture-of-Experts (MoE) architecture has been widely adopted in large language models (LLMs) to reduce computation cost through model sparsity. Employing speculative decoding (SD) can further accelerate MoE inference by drafting multiple tokens per step and verifying them in parallel. However, combining MoE with S...
[ "Liangkun Chen", "Zijian Wen", "Tian Wu", "Xiaoxi Zhang", "Chuan Wu" ]
[ "cs.DC" ]
[ "Computer Science" ]
2025-10-11T00:00:00
https://arxiv.org/abs/2510.10302
https://arxiv.org/pdf/2510.10302v2
2510.10302
10.48550/arXiv.2510.10302
3
0
false
null
arXiv.org
0.3311
cce4ea6eecbe8e5b871c8896f310d805fba3867618365d3436a8e3f8d0df76ca
[ "arxiv", "semantic_scholar" ]
Speculative Jacobi-Denoising Decoding for Accelerating Autoregressive Text-to-image Generation
As a new paradigm of visual content generation, autoregressive text-to-image models suffer from slow inference due to their sequential token-by-token decoding process, often requiring thousands of model forward passes to generate a single image. To address this inefficiency, we propose Speculative Jacobi-Denoising Deco...
[ "Yao Teng", "Fuyun Wang", "Xian Liu", "Zhekai Chen", "Han Shi", "Yu Wang", "Zhenguo Li", "Weiyang Liu", "Difan Zou", "Xihui Liu" ]
[ "cs.CV" ]
[ "Computer Science" ]
2025-10-10T00:00:00
https://arxiv.org/abs/2510.08994
https://arxiv.org/pdf/2510.08994v1
2510.08994
10.48550/arXiv.2510.08994
4
1
false
null
arXiv.org
0.33
8d87ee13b01f1e8580143616019a08c09c2bfc308ef3bce5f74003b398aa8186
[ "arxiv", "semantic_scholar" ]
SPAD: Specialized Prefill and Decode Hardware for Disaggregated LLM Inference
Large Language Models (LLMs) have gained popularity in recent years, driving up the demand for inference. LLM inference is composed of two phases with distinct characteristics: a compute-bound prefill phase followed by a memory-bound decode phase. To efficiently serve LLMs, prior work proposes prefill-decode disaggrega...
[ "Hengrui Zhang", "Pratyush Patel", "August Ning", "David Wentzlaff" ]
[ "cs.AR", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2025-10-09T00:00:00
https://arxiv.org/abs/2510.08544
https://arxiv.org/pdf/2510.08544v1
2510.08544
10.48550/arXiv.2510.08544
5
0
false
null
arXiv.org
0.3289
7b5859c00f7474158aece9cbeaa91dd3e811b1ed862e9f5029de4480cd32954e
[ "arxiv", "semantic_scholar" ]
OWL: Overcoming Window Length-Dependence in Speculative Decoding for Long-Context Inputs
Speculative decoding promises faster inference for large language models (LLMs), yet existing methods fail to generalize to real-world settings. Benchmarks typically assume short contexts (e.g., 2K tokens), whereas practical workloads involve long contexts. We find current approaches degrade severely with long contexts...
[ "Jaeseong Lee", "seung-won hwang", "Aurick Qiao", "Gabriele Oliaro", "Ye Wang", "Samyam Rajbhandari" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-10-08T00:00:00
https://arxiv.org/abs/2510.07535
https://arxiv.org/pdf/2510.07535v1
2510.07535
10.48550/arXiv.2510.07535
0
0
false
null
arXiv.org
0.3277
346cd617ccf424b28e1e4eab0cb897f62ecd7b2ffc7deaa9c316d6e58cd43eab
[ "arxiv", "semantic_scholar" ]
Draft, Verify, and Improve: Toward Training-Aware Speculative Decoding
Autoregressive (AR) decoding is a major latency bottleneck for large language models. Speculative decoding (SD) accelerates AR by letting a drafter propose multi-token blocks that a verifier accepts or rejects. However, many SD systems require heavy offline training or extra components. These choices raise data/compute...
[ "Shrenik Bhansali", "Larry Heck" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-10-06T00:00:00
https://arxiv.org/abs/2510.05421
https://arxiv.org/pdf/2510.05421v1
2510.05421
10.48550/arXiv.2510.05421
1
0
false
null
arXiv.org
0.3254
c984a01a98ad2856f3e307a10fb15d4261bcc017d099d8052df0df7a8fbecb20
[ "arxiv", "semantic_scholar" ]
Self Speculative Decoding for Diffusion Large Language Models
Diffusion-based Large Language Models (dLLMs) have emerged as a competitive alternative to autoregressive models, offering unique advantages through bidirectional attention and parallel generation paradigms. However, the generation results of current parallel decoding methods deviate from stepwise decoding, introducing...
[ "Yifeng Gao", "Ziang Ji", "Yuxuan Wang", "Biqing Qi", "Hanlin Xu", "Linfeng Zhang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-10-05T00:00:00
https://arxiv.org/abs/2510.04147
https://arxiv.org/pdf/2510.04147v1
2510.04147
10.48550/arXiv.2510.04147
25
5
true
null
arXiv.org
0.5011
de9adfe6916a2b78d1d444f1e3a0bf5f0401e3ea6ebe9b23dbfb7ff81dd9e499
[ "arxiv", "semantic_scholar" ]
The Disparate Impacts of Speculative Decoding
The practice of speculative decoding, whereby inference is probabilistically supported by a smaller, cheaper, ``drafter'' model, has become a standard technique for systematically reducing the decoding time of large language models. This paper conducts an analysis of speculative decoding through the lens of its potenti...
[ "Jameson Sandler", "Ahmet Üstün", "Marco Romanelli", "Sara Hooker", "Ferdinando Fioretto" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-10-02T00:00:00
https://arxiv.org/abs/2510.02128
https://arxiv.org/pdf/2510.02128v1
2510.02128
10.48550/arXiv.2510.02128
2
0
false
null
arXiv.org
0.3208
6efead990ebdce075638f64911d7b96e7a2d66789cb4e07586b47875b8310f56
[ "arxiv", "semantic_scholar" ]
HiSpec: Hierarchical Speculative Decoding for LLMs
Speculative decoding accelerates LLM inference by using a smaller draft model to speculate tokens that a larger target model verifies. Verification is often the bottleneck (e.g. verification is $4\times$ slower than token generation when a 3B model speculates for a 70B target model), but most prior works focus only on ...
[ "Avinash Kumar", "Sujay Sanghavi", "Poulami Das" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-10-01T00:00:00
https://arxiv.org/abs/2510.01336
https://arxiv.org/pdf/2510.01336v2
2510.01336
10.48550/arXiv.2510.01336
1
0
false
null
arXiv.org
0.3197
6324420d09a0a05ef0eb7a32dc54aa143c05a99733a2f0a8fa9211292c0ab04e
[ "arxiv", "semantic_scholar" ]
Speculative Verification: Exploiting Information Gain to Refine Speculative Decoding
LLMs have low GPU efficiency and high latency due to autoregressive decoding. Speculative decoding (SD) mitigates this using a small draft model to speculatively generate multiple tokens, which are then verified in parallel by a target model. However, when speculation accuracy is low, the overhead from rejected tokens ...
[ "Sungkyun Kim", "Jaemin Kim", "Dogyung Yoon", "Jiho Shin", "Junyeol Lee", "Jiwon Seo" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-09-29T00:00:00
https://arxiv.org/abs/2509.24328
https://arxiv.org/pdf/2509.24328v2
2509.24328
10.48550/arXiv.2509.24328
1
0
false
null
arXiv.org
0.3174
88eebe9d27ee6615bf039d2474fe8fe49463ebac3a4fc2cc3374f8217b3a18f4
[ "arxiv", "semantic_scholar" ]
HiViS: Hiding Visual Tokens from the Drafter for Speculative Decoding in Vision-Language Models
Speculative decoding has proven effective for accelerating inference in Large Language Models (LLMs), yet its extension to Vision-Language Models (VLMs) remains limited by the computational burden and semantic inconsistency introduced by visual tokens. Recent studies reveal that visual tokens in large VLMs are highly r...
[ "Zhinan Xie", "Peisong Wang", "Shuang Qiu", "Jian Cheng" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-09-28T00:00:00
https://arxiv.org/abs/2509.23928
https://arxiv.org/pdf/2509.23928v2
2509.23928
10.48550/arXiv.2509.23928
1
0
false
null
arXiv.org
0.3162
bb334f7fd49d3b2f29971cd30b20b09e1cf88cc54f08bdf3490153c93a60e261
[ "arxiv", "semantic_scholar" ]
DiffuSpec: Unlocking Diffusion Language Models for Speculative Decoding
As large language models (LLMs) scale up, accuracy improves, but the autoregressive (AR) nature of decoding increases latency since each token requires a serial forward pass. Speculative decoding addresses this by employing a fast drafter to propose multi-token drafts, which are then verified in parallel by the target ...
[ "Guanghao Li", "Zhihui Fu", "Min Fang", "Qibin Zhao", "Ming Tang", "Chun Yuan", "Jun Wang" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-09-28T00:00:00
https://arxiv.org/abs/2510.02358
https://arxiv.org/pdf/2510.02358v1
2510.02358
10.48550/arXiv.2510.02358
21
0
false
null
arXiv.org
0.3356
3f8daa231589327dd1ca1e40074b63c58eb37c1d9102560caa85c1a6b4310e1c
[ "arxiv", "semantic_scholar" ]
SelfJudge: Faster Speculative Decoding via Self-Supervised Judge Verification
Speculative decoding accelerates LLM inference by verifying candidate tokens from a draft model against a larger target model. Recent judge decoding boosts this process by relaxing verification criteria by accepting draft tokens that may exhibit minor discrepancies from target model output, but existing methods are res...
[ "Kanghoon Yoon", "Minsub Kim", "Sungjae Lee", "Joonhyung Lee", "Sunghyeon Woo", "Yeonjun In", "Se Jung Kwon", "Chanyoung Park", "Dongsoo Lee" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-09-26T00:00:00
https://arxiv.org/abs/2510.02329
https://arxiv.org/pdf/2510.02329v2
2510.02329
10.48550/arXiv.2510.02329
4
0
false
null
arXiv.org
0.314
179a0e42cc9b324bef67d7efee9db40d2a6f7f241d7d97a3730063a1142cb1de
[ "arxiv", "semantic_scholar" ]
Self-Speculative Biased Decoding for Faster Re-Translation
Large language models achieve strong machine translation quality but incur high inference cost and latency, posing challenges for simultaneous translation. Re-translation provides a practical solution for off-the-shelf LLMs by repeatedly regenerating the target output as the source input grows, but it suffers from subs...
[ "Linxiao Zeng", "Haoyun Deng", "Kangyuan Shu", "Shizhen Wang" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-09-26T00:00:00
https://arxiv.org/abs/2509.21740
https://arxiv.org/pdf/2509.21740v2
2509.21740
null
0
0
false
null
null
0.1998
9afae801d6aac95feff6025a3697734664b926c472fb09096026d6a08ddc1eeb
[ "arxiv", "semantic_scholar" ]
Bridging Draft Policy Misalignment: Group Tree Optimization for Speculative Decoding
Speculative decoding accelerates large language model (LLM) inference by letting a lightweight draft model propose multiple tokens that the target model verifies in parallel. Yet existing training objectives optimize only a single greedy draft path, while decoding follows a tree policy that re-ranks and verifies multip...
[ "Shijing Hu", "Jingyang Li", "Zhihui Lu", "Pan Zhou" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-09-26T00:00:00
https://arxiv.org/abs/2509.22134
https://arxiv.org/pdf/2509.22134v2
2509.22134
10.48550/arXiv.2509.22134
5
1
true
https://github.com/hsj576/GTO
arXiv.org
0.4852
67ab3e7150b2cc3e9d9a8baddf77e7a1fcb7305099a5866674099a9a01d94896
[ "arxiv", "semantic_scholar" ]
FastGRPO: Accelerating Policy Optimization via Concurrency-aware Speculative Decoding and Online Draft Learning
Group relative policy optimization (GRPO) has demonstrated significant potential in improving the reasoning capabilities of large language models (LLMs) via reinforcement learning. However, its practical deployment is impeded by an excessively slow training process, primarily attributed to the computationally intensive...
[ "Yizhou Zhang", "Ning Lv", "Teng Wang", "Jisheng Dang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-09-26T00:00:00
https://arxiv.org/abs/2509.21792
https://arxiv.org/pdf/2509.21792v1
2509.21792
10.48550/arXiv.2509.21792
4
0
true
https://github.com/yedaotian9/GRPO_speculative
arXiv.org
0.4852
16a83b33cd2e7a6fb6a32499669a7656b00c00d685df9c4410e31d3b14ae4f32
[ "arxiv", "semantic_scholar" ]
SpecMER: Fast Protein Generation with K-mer Guided Speculative Decoding
Autoregressive models have transformed protein engineering by enabling the generation of novel protein sequences beyond those found in nature. However, their sequential inference introduces significant latency, limiting their utility in high-throughput protein screening. Speculative decoding accelerates generation by e...
[ "Thomas Walton", "Darin Tsui", "Aryan Musharaf", "Amirali Aghazadeh" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-09-25T00:00:00
https://arxiv.org/abs/2509.21689
https://arxiv.org/pdf/2509.21689v1
2509.21689
10.48550/arXiv.2509.21689
2
0
false
null
arXiv.org
0.3128
215ab202dca4ce7f7f2df32aebc83b69958af1530c743a54610fd951c29ae036
[ "arxiv", "semantic_scholar" ]
FastEagle: Cascaded Drafting for Accelerating Speculative Decoding
Speculative decoding accelerates generation by drafting candidates and verifying them in parallel, yet state-of-the-art drafters (e.g., EAGLE) still require N sequential passes to propose N tokens. We present FastEagle, a non-autoregressive cascaded drafter that emits an entire draft in a single forward pass. FastEagle...
[ "Haiduo Huang", "Jiangcheng Song", "Wenzhe Zhao", "Pengju Ren" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-09-24T00:00:00
https://arxiv.org/abs/2509.20416
https://arxiv.org/pdf/2509.20416v1
2509.20416
10.48550/arXiv.2509.20416
1
0
false
null
IEEE International Conference on Acoustics, Speech, and Signal Processing
0.3117
ad9c0d380cd671f34f669c66ce4e8491b51ea016c658e69440c0729be00810de
[ "arxiv", "semantic_scholar" ]
SpecMamba: Accelerating Mamba Inference on FPGA with Speculative Decoding
The growing demand for efficient long-sequence modeling on edge devices has propelled widespread adoption of State Space Models (SSMs) like Mamba, due to their superior computational efficiency and scalability. As its autoregressive generation process remains memory-bound, speculative decoding has been proposed that in...
[ "Linfeng Zhong", "Songqiang Xu", "Huifeng Wen", "Tong Xie", "Qingyu Guo", "Yuan Wang", "Meng Li" ]
[ "cs.AR" ]
[ "Computer Science" ]
2025-09-24T00:00:00
https://arxiv.org/abs/2509.19873
https://arxiv.org/pdf/2509.19873v1
2509.19873
10.1109/ICCAD66269.2025.11240945
2
0
false
null
null
0.1983
793293c4df76a6ccaa97b98fe6b9e18f2de67267134acfe7dccfe7d73c7bada5
[ "arxiv", "semantic_scholar" ]
Steering Multimodal Large Language Models Decoding for Context-Aware Safety
Multimodal Large Language Models (MLLMs) are increasingly deployed in real-world applications, yet their ability to make context-aware safety decisions remains limited. Existing methods often fail to balance oversensitivity (unjustified refusals of benign queries) and undersensitivity (missed detection of visually grou...
[ "Zheyuan Liu", "Zhangchen Xu", "Guangyao Dou", "Xiangchi Yuan", "Zhaoxuan Tan", "Radha Poovendran", "Meng Jiang" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-09-23T00:00:00
https://arxiv.org/abs/2509.19212
https://arxiv.org/pdf/2509.19212v1
2509.19212
10.48550/arXiv.2509.19212
4
0
false
null
arXiv.org
0.3105
15fe8618c8fc53ca1fe892f6ff252d7ec1b9775da3493ce59408dc27b77ab19b
[ "arxiv", "semantic_scholar" ]
Structuring The Future: Diffusion LLM Speculative Decoding via Calibrated Draft Graphs
Diffusion LLMs (dLLMs) have recently emerged as a powerful alternative to autoregressive LLMs (AR-LLMs) with the potential to operate at significantly higher token-generation rates. To unlock this potential, we present Spiffy, a speculative decoding algorithm to accelerate dLLM inference while provably preserving the m...
[ "Sudhanshu Agrawal", "Risheek Garrepalli", "Raghavv Goel", "Christopher Lott", "Fatih Porikli", "Mingu Lee" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2025-09-22T00:00:00
https://arxiv.org/abs/2509.18085
https://arxiv.org/pdf/2509.18085v4
2509.18085
null
17
3
false
null
null
0.3138
c2fba4d457deaf2bfba4a6d9493c8f177835aa264aa1d0ab71932ca7ab5ea63f
[ "arxiv", "semantic_scholar" ]
Speculate Deep and Accurate: Lossless and Training-Free Acceleration for Offloaded LLMs via Substitute Speculative Decoding
The immense model sizes of large language models (LLMs) challenge deployment on memory-limited consumer GPUs. Although model compression and parameter offloading are common strategies to address memory limitations, compression can degrade quality, and offloading maintains quality but suffers from slow inference. Specul...
[ "Pei-Shuo Wang", "Jian-Jia Chen", "Chun-Che Yang", "Chi-Chih Chang", "Ning-Chi Huang", "Mohamed S. Abdelfattah", "Kai-Chiang Wu" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-09-22T00:00:00
https://arxiv.org/abs/2509.18344
https://arxiv.org/pdf/2509.18344v2
2509.18344
10.48550/arXiv.2509.18344
1
0
false
null
arXiv.org
0.3094
365425d46ade5ed71301cc1ba85a2c3fee56538c7158dbb811c009a6308f407c
[ "arxiv", "semantic_scholar" ]
Pipeline Parallelism is All You Need for Optimized Early-Exit Based Self-Speculative Decoding
Large language models (LLMs) deliver impressive generation quality, but incur very high inference cost because each output token is generated auto-regressively through all model layers. Early-exit based self-speculative decoding (EESD) has emerged to mitigate this cost. However, in practice, many approaches struggle to...
[ "Ruanjun Li", "Ziheng Liu", "Yuanming Shi", "Jiawei Shao", "Chi Zhang", "Xuelong Li" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-09-19T00:00:00
https://arxiv.org/abs/2509.19368
https://arxiv.org/pdf/2509.19368v1
2509.19368
10.48550/arXiv.2509.19368
1
1
false
null
arXiv.org
0.3059
2f5c5a74ee7ee207824de4bd86f71705e6030b2a36960756cb9f72dfe50d7115
[ "arxiv", "semantic_scholar" ]
ViSpec: Accelerating Vision-Language Models with Vision-Aware Speculative Decoding
Speculative decoding is a widely adopted technique for accelerating inference in large language models (LLMs), yet its application to vision-language models (VLMs) remains underexplored, with existing methods achieving only modest speedups (<1.5x). This gap is increasingly significant as multimodal capabilities become ...
[ "Jialiang Kang", "Han Shu", "Wenshuo Li", "Yingjie Zhai", "Xinghao Chen" ]
[ "cs.CV", "cs.CL" ]
[ "Computer Science" ]
2025-09-17T00:00:00
https://arxiv.org/abs/2509.15235
https://arxiv.org/pdf/2509.15235v5
2509.15235
10.48550/arXiv.2509.15235
15
4
true
https://github.com/KangJialiang/ViSpec
arXiv.org
0.4693
861ef963718f14257628c3a92f01282981a171aae5b147d92221f8ef465dde3f
[ "arxiv", "semantic_scholar" ]
FastMTP: Accelerating LLM Inference with Enhanced Multi-Token Prediction
As large language models (LLMs) become increasingly powerful, the sequential nature of autoregressive generation creates a fundamental throughput bottleneck that limits the practical deployment. While Multi-Token Prediction (MTP) has demonstrated remarkable benefits for model training efficiency and performance, its in...
[ "Yuxuan Cai", "Xiaozhuan Liang", "Xinghua Wang", "Jin Ma", "Haijin Liang", "Jinwen Luo", "Xinyu Zuo", "Lisheng Duan", "Yuyang Yin", "Xi Chen" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-09-16T00:00:00
https://arxiv.org/abs/2509.18362
https://arxiv.org/pdf/2509.18362v1
2509.18362
10.48550/arXiv.2509.18362
9
0
false
null
arXiv.org
0.3025
5637bbb97ebe7f7da09af4390909e01a3877beed82e4a714cbf0fe37066ddc7c
[ "arxiv", "semantic_scholar" ]
Spec-LLaVA: Accelerating Vision-Language Models with Dynamic Tree-Based Speculative Decoding
Vision-Language Models (VLMs) enable powerful multimodal reasoning but suffer from slow autoregressive inference, limiting their deployment in real-time applications. We introduce Spec-LLaVA, a system that applies speculative decoding to accelerate VLMs without sacrificing output quality. Spec-LLaVA pairs a lightweight...
[ "Mingxiao Huo", "Jiayi Zhang", "Hewei Wang", "Jinfeng Xu", "Zheyu Chen", "Huilin Tai", "Yijun Chen" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-09-15T00:00:00
https://arxiv.org/abs/2509.11961
https://arxiv.org/pdf/2509.11961v1
2509.11961
10.48550/arXiv.2509.11961
5
1
false
null
arXiv.org
0.3014
894f7a3f734e1b05aa2fc523968b6e32e79bd6352c39ee554082de048d7992aa
[ "arxiv", "semantic_scholar" ]
SpecVLM: Fast Speculative Decoding in Vision-Language Models
Speculative decoding is a powerful way to accelerate autoregressive large language models (LLMs), but directly porting it to vision-language models (VLMs) faces unique systems constraints: the prefill stage is dominated by visual tokens whose count scales with image resolution and video length, inflating both compute a...
[ "Haiduo Huang", "Fuwei Yang", "Zhenhua Liu", "Xuanwu Yin", "Dong Li", "Pengju Ren", "Emad Barsoum" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2025-09-15T00:00:00
https://arxiv.org/abs/2509.11815
https://arxiv.org/pdf/2509.11815v2
2509.11815
10.48550/arXiv.2509.11815
4
2
true
https://github.com/haiduo/SpecVLM
arXiv.org
0.4657
260f3fc8beeb9cd64567ffe55ff4df7e5c33e1573922c3a460a9e3dc0b8fdf18
[ "arxiv", "semantic_scholar" ]
Communication-Efficient Collaborative LLM Inference via Distributed Speculative Decoding
Speculative decoding is an emerging technique that accelerates large language model (LLM) inference by allowing a smaller draft model to predict multiple tokens in advance, which are then verified or corrected by a larger target model. In AI-native radio access networks (AI-RAN), this paradigm is well-suited for collab...
[ "Ce Zheng", "Tingting Yang" ]
[ "eess.SP" ]
[ "Engineering" ]
2025-09-04T00:00:00
https://arxiv.org/abs/2509.04576
https://arxiv.org/pdf/2509.04576v2
2509.04576
10.1109/WCSP68525.2025.1010651
6
0
false
null
International Conference on Wireless Communications and Signal Processing
0.2888
887a9689b0b953df920246b4e40c7c8527c9c53738df7c46de2c0135b29dca2b
[ "arxiv", "semantic_scholar" ]
Can LLMs Lie? Investigation beyond Hallucination
Large language models (LLMs) have demonstrated impressive capabilities across a variety of tasks, but their increasing autonomy in real-world applications raises concerns about their trustworthiness. While hallucinations-unintentional falsehoods-have been widely studied, the phenomenon of lying, where an LLM knowingly ...
[ "Haoran Huan", "Mihir Prabhudesai", "Mengning Wu", "Shantanu Jaiswal", "Deepak Pathak" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-09-03T00:00:00
https://arxiv.org/abs/2509.03518
https://arxiv.org/pdf/2509.03518v1
2509.03518
10.48550/arXiv.2509.03518
6
0
false
null
arXiv.org
0.2876
bb7f67abf4e114b9bdedb911990b4d6d39f22f8a60ac6048b1d98f7f167376df
[ "arxiv", "semantic_scholar" ]
DSDE: Dynamic Speculative Decoding with KLD Stability for Real-World Serving
Speculative decoding accelerates large language model inference, but its reliance on a fixed speculation length is suboptimal in large-batch serving environments with diverse requests. This paper explores a new direction for dynamic adaptation by investigating a novel class of post-hoc, diagnostic signals. We propose D...
[ "Mingyu Yang", "Jae-Young Choi", "Kihyo Moon", "Minsung Jang", "Eunjoo Jeon" ]
[ "cs.DC", "cs.AI", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2025-09-01T00:00:00
https://arxiv.org/abs/2509.01083
https://arxiv.org/pdf/2509.01083v3
2509.01083
10.1109/BigData66926.2025.11400767
0
0
false
null
BigData Congress [Services Society]
0.2853
0c7bd14df643412aeb3523fe539bdd2c708cd6a197acad42833a7fe4134a4c34
[ "arxiv", "semantic_scholar" ]
Scaling Up, Speeding Up: A Benchmark of Speculative Decoding for Efficient LLM Test-Time Scaling
Test-time scaling has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models (LLMs) by allocating additional computational resources during inference. However, this paradigm is inherently inefficient due to the generation of redundant and repetitive reasoning traces, leading to...
[ "Shengyin Sun", "Yiming Li", "Xing Li", "Yingzhao Lian", "Weizhe Lin", "Hui-Ling Zhen", "Zhiyuan Yang", "Chen Chen", "Xianzhi Yu", "Mingxuan Yuan", "Chen Ma" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-08-30T00:00:00
https://arxiv.org/abs/2509.04474
https://arxiv.org/pdf/2509.04474v1
2509.04474
10.48550/arXiv.2509.04474
5
1
false
null
arXiv.org
0.283
15d6a2f927d4c2ebd145bd467769306cfdc6d0ccd385a856fcf02f73aae45288
[ "arxiv", "semantic_scholar" ]
Accelerating Mixture-of-Experts Inference by Hiding Offloading Latency with Speculative Decoding
Recent advancements in Mixture of Experts (MoE) models have significantly increased their parameter scale as well as model performance. Extensive offloading techniques have been proposed to address the GPU memory limitations of MoE inference. However, due to the I/O bottleneck and sparse computation of MoE models, exis...
[ "Zhibin Wang", "Zhonghui Zhang", "Yuhang Zhou", "Zibo Wang", "Mo Zhou", "Peng Jiang", "Weilin Cai", "Chengying Huan", "Rong Gu", "Sheng Zhong", "Chen Tian" ]
[ "cs.DC" ]
[ "Computer Science" ]
2025-08-29T00:00:00
https://arxiv.org/abs/2508.21706
https://arxiv.org/pdf/2508.21706v2
2508.21706
10.48550/arXiv.2508.21706
2
0
false
null
arXiv.org
0.2819
f54025c03f3782e8ebc58ad19d73578c6a0901e475080f92d8f495e8e518ca06
[ "arxiv", "semantic_scholar" ]
Speculative Safety-Aware Decoding
Despite extensive efforts to align Large Language Models (LLMs) with human values and safety rules, jailbreak attacks that exploit certain vulnerabilities continuously emerge, highlighting the need to strengthen existing LLMs with additional safety properties to defend against these attacks. However, tuning large model...
[ "Xuekang Wang", "Shengyu Zhu", "Xueqi Cheng" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-08-25T00:00:00
https://arxiv.org/abs/2508.17739
https://arxiv.org/pdf/2508.17739v2
2508.17739
10.48550/arXiv.2508.17739
2
1
true
https://github.com/k-k1w-w1x-x/Speculative-Safety-Aware-Decoding
Conference on Empirical Methods in Natural Language Processing
0.4285
582310ee96001a9bd00e7ea681e1420d7ad51ece472ffc4e9d87a8f3056da3ee
[ "arxiv", "semantic_scholar" ]
SpecVLM: Enhancing Speculative Decoding of Video LLMs via Verifier-Guided Token Pruning
Video large language models (Vid-LLMs) have shown strong capabilities in understanding video content. However, their reliance on dense video token representations introduces substantial memory and computational overhead in both prefilling and decoding. To mitigate the information loss of recent video token reduction me...
[ "Yicheng Ji", "Jun Zhang", "Heming Xia", "Jinpeng Chen", "Lidan Shou", "Gang Chen", "Huan Li" ]
[ "cs.CV", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2025-08-22T00:00:00
https://arxiv.org/abs/2508.16201
https://arxiv.org/pdf/2508.16201v2
2508.16201
10.48550/arXiv.2508.16201
18
3
true
https://github.com/zju-jiyicheng/SpecVLM
Conference on Empirical Methods in Natural Language Processing
0.4232
df8a049eebf8bd12ee5724231258017456ae08970dde5a3d8e428415223a47b5
[ "arxiv", "semantic_scholar" ]
Confidence-Modulated Speculative Decoding for Large Language Models
Speculative decoding has emerged as an effective approach for accelerating autoregressive inference by parallelizing token generation through a draft-then-verify paradigm. However, existing methods rely on static drafting lengths and rigid verification criteria, limiting their adaptability across varying model uncertai...
[ "Jaydip Sen", "Subhasis Dasgupta", "Hetvi Waghela" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-08-21T00:00:00
https://arxiv.org/abs/2508.15371
https://arxiv.org/pdf/2508.15371v1
2508.15371
10.1109/INDISCON66021.2025.11254640
1
0
false
null
null
0.1735
8c765d98efbd5162ad8be764e6c51a4f0c79dc8246c332914d98bb9bdc8fcef7
[ "arxiv", "semantic_scholar" ]
READER: Retrieval-Assisted Drafter for Efficient LLM Inference
Autoregressive Language Models instantiate a factorized likelihood over token sequences, yet their strictly sequential decoding process imposes an intrinsic lower bound on inference latency. This bottleneck has emerged as a central obstacle to the scalable deployment of large-scale generative models. Existing accelerat...
[ "Maxim Divilkovskiy", "Vitaly Malygin", "Sergey Zlobin", "Stanislav Ilyushin", "Sultan Isali", "Vasily Kalugin", "Nuriza Aitassova", "Fei Yi", "Weidi Zeng" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-08-12T00:00:00
https://arxiv.org/abs/2508.09072
https://arxiv.org/pdf/2508.09072v2
2508.09072
10.48550/arXiv.2508.09072
0
0
false
null
arXiv.org
0.2624
f8812d9aa15cb01e05acf34eacaec0b9426cbdb76b6db7cd19cc7507d7a0cf6d
[ "arxiv", "semantic_scholar" ]
ASPD: Unlocking Adaptive Serial-Parallel Decoding by Exploring Intrinsic Parallelism in LLMs
The increasing scale and complexity of large language models (LLMs) pose significant inference latency challenges, primarily due to their autoregressive decoding paradigm characterized by the sequential nature of next-token prediction. By re-examining the outputs of autoregressive models, we observed that some segments...
[ "Keyu Chen", "Zhifeng Shen", "Daohai Yu", "Haoqian Wu", "Wei Wen", "Jianfeng He", "Ruizhi Qiao", "Xing Sun" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-08-12T00:00:00
https://arxiv.org/abs/2508.08895
https://arxiv.org/pdf/2508.08895v2
2508.08895
10.48550/arXiv.2508.08895
6
2
false
null
arXiv.org
0.2624
3e9fa77cdc3447792bc8baf19bbffe001e1efaf16379fe10d5dc1544e350cc6b
[ "arxiv", "semantic_scholar" ]
Efficient Speculative Decoding for Llama at Scale: Challenges and Solutions
Speculative decoding is a standard method for accelerating the inference speed of large language models. However, scaling it for production environments poses several engineering challenges, including efficiently implementing different operations (e.g., tree attention and multi-round speculative decoding) on GPU. In th...
[ "Bangsheng Tang", "Carl Chengyan Fu", "Fei Kou", "Grigory Sizov", "Haoci Zhang", "Jason Park", "Jiawen Liu", "Jie You", "Qirui Yang", "Sachin Mehta", "Shengyong Cai", "Xiaodong Wang", "Xingyu Liu", "Yunlu Li", "Yanjun Zhou", "Wei Wei", "Zhiwei Zhao", "Zixi Qi", "Adolfo Victoria",...
[ "cs.CL" ]
[ "Computer Science" ]
2025-08-11T00:00:00
https://arxiv.org/abs/2508.08192
https://arxiv.org/pdf/2508.08192v1
2508.08192
10.48550/arXiv.2508.08192
8
2
false
null
arXiv.org
0.2612
d56d5385f04fa3dd872f88fe9cdf2f251ffdcbf64156bff3310095f5975bc234
[ "arxiv", "semantic_scholar" ]
Grouped Speculative Decoding for Autoregressive Image Generation
Recently, autoregressive (AR) image models have demonstrated remarkable generative capabilities, positioning themselves as a compelling alternative to diffusion models. However, their sequential nature leads to long inference times, limiting their practical scalability. In this work, we introduce Grouped Speculative De...
[ "Junhyuk So", "Juncheol Shin", "Hyunho Kook", "Eunhyeok Park" ]
[ "cs.CV" ]
[ "Computer Science" ]
2025-08-11T00:00:00
https://arxiv.org/abs/2508.07747
https://arxiv.org/pdf/2508.07747v1
2508.07747
10.1109/ICCV51701.2025.01426
13
6
true
https://github.com/junhyukso/GSD
IEEE International Conference on Computer Vision
0.4225
43d2589c16df6f1d0cdafaf9faa07e9245abc1198310a3e99ea6eae9fb251354
[ "arxiv", "semantic_scholar" ]
LP-Spec: Leveraging LPDDR PIM for Efficient LLM Mobile Speculative Inference with Architecture-Dataflow Co-Optimization
LLM inference on mobile devices faces extraneous challenges due to limited memory bandwidth and computational resources. To address these issues, speculative inference and processing-in-memory (PIM) techniques have been explored at the algorithmic and hardware levels. However, speculative inference results in more comp...
[ "Siyuan He", "Zhantong Zhu", "Yandong He", "Tianyu Jia" ]
[ "cs.AR" ]
[ "Computer Science" ]
2025-08-10T00:00:00
https://arxiv.org/abs/2508.07227
https://arxiv.org/pdf/2508.07227v3
2508.07227
10.1109/ICCAD66269.2025.11240889
6
0
false
null
null
0.2113
041b82597f1ba8ca7a864e962a2dad0f6bcebaad68080ae301c91dc6af7b3b7f
[ "arxiv", "semantic_scholar" ]
CARD: A Cache-Assisted Parallel Speculative Decoding Framework via Query-and-Correct Paradigm for Accelerating LLM Inference
Speculative decoding (SD), where a draft model provides multiple candidate tokens for the target model to verify in parallel, has demonstrated significant potential for accelerating LLM inference. Yet, existing SD approaches adhere to a strict draft-then-verify paradigm, enforcing a sequential process that hampers perf...
[ "Enyu Zhou", "Kai Sheng", "Hao Chen", "Xin He" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-08-06T00:00:00
https://arxiv.org/abs/2508.04462
https://arxiv.org/pdf/2508.04462v2
2508.04462
null
0
0
false
null
null
0.1626
c6c07b080e6989ff636f08bcf3b5c24573c2cd66272f03e7f1a517e50ae82fb3
[ "arxiv", "semantic_scholar" ]
XSpecMesh: Quality-Preserving Auto-Regressive Mesh Generation Acceleration via Multi-Head Speculative Decoding
Current auto-regressive models can generate high-quality, topologically precise meshes; however, they necessitate thousands-or even tens of thousands-of next-token predictions during inference, resulting in substantial latency. We introduce XSpecMesh, a quality-preserving acceleration method for auto-regressive mesh ge...
[ "Dian Chen", "Yansong Qu", "Xinyang Li", "Ming Li", "Shengchuan Zhang" ]
[ "cs.GR", "cs.CV", "cs.LG" ]
[ "Computer Science" ]
2025-07-31T00:00:00
https://arxiv.org/abs/2507.23777
https://arxiv.org/pdf/2507.23777v2
2507.23777
10.48550/arXiv.2507.23777
4
0
false
null
arXiv.org
0.2486
3a2d2d259747232165d5fd7ac6603592bd111242e4cadacb951d2f7a58c63a1a
[ "arxiv", "semantic_scholar" ]
Hierarchical Verification of Speculative Beams for Accelerating LLM Inference
Large language models (LLMs) have achieved remarkable success across diverse natural language processing tasks but face persistent challenges in inference efficiency due to their autoregressive nature. While speculative decoding and beam sampling offer notable improvements, traditional methods verify draft sequences se...
[ "Jaydip Sen", "Harshitha Puvvala", "Subhasis Dasgupta" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-07-30T00:00:00
https://arxiv.org/abs/2508.03726
https://arxiv.org/pdf/2508.03726v1
2508.03726
10.1007/978-3-032-07735-6_19
2
0
false
null
arXiv.org
0.2475
85a18671c07272727909d6c3889128af62a1391015f10e7c52d404e7ac5c7d23
[ "arxiv", "semantic_scholar" ]
Spec-VLA: Speculative Decoding for Vision-Language-Action Models with Relaxed Acceptance
Vision-Language-Action (VLA) models have made substantial progress by leveraging the robust capabilities of Visual Language Models (VLMs). However, VLMs' significant parameter size and autoregressive (AR) decoding nature impose considerable computational demands on VLA models. While Speculative Decoding (SD) has shown ...
[ "Songsheng Wang", "Rucheng Yu", "Zhihang Yuan", "Chao Yu", "Feng Gao", "Yu Wang", "Derek F. Wong" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-07-30T00:00:00
https://arxiv.org/abs/2507.22424
https://arxiv.org/pdf/2507.22424v2
2507.22424
10.48550/arXiv.2507.22424
25
2
false
null
Conference on Empirical Methods in Natural Language Processing
0.3537
28a30d0efea8f062a95654d87654af295152898cc6351bc8d3810006e5bc765f
[ "arxiv", "semantic_scholar" ]
Model-free Speculative Decoding for Transformer-based ASR with Token Map Drafting
End-to-end automatic speech recognition (ASR) systems based on transformer architectures, such as Whisper, offer high transcription accuracy and robustness. However, their autoregressive decoding is computationally expensive, hence limiting deployment on CPU-based and resource-constrained devices. Speculative decoding ...
[ "Tuan Vu Ho", "Hiroaki Kokubo", "Masaaki Yamamoto", "Yohei Kawaguchi" ]
[ "cs.CL", "cs.SD", "eess.AS" ]
[ "Computer Science", "Engineering" ]
2025-07-29T00:00:00
https://arxiv.org/abs/2507.21522
https://arxiv.org/pdf/2507.21522v1
2507.21522
10.48550/arXiv.2507.21522
0
0
false
null
European Signal Processing Conference
0.2464
0b3c5255fbfb5f2b3587ffe13e9e62fd1ed173b4ca757e9de08b52c5cbf100d7
[ "arxiv", "semantic_scholar" ]
Enhancing Jailbreak Attacks on LLMs via Persona Prompts
Jailbreak attacks aim to exploit large language models (LLMs) by inducing them to generate harmful content, thereby revealing their vulnerabilities. Understanding and addressing these attacks is crucial for advancing the field of LLM safety. Previous jailbreak approaches have mainly focused on direct manipulations of h...
[ "Zheng Zhang", "Peilin Zhao", "Deheng Ye", "Hao Wang" ]
[ "cs.CR", "cs.AI" ]
[ "Computer Science" ]
2025-07-28T00:00:00
https://arxiv.org/abs/2507.22171
https://arxiv.org/pdf/2507.22171v3
2507.22171
10.48550/arXiv.2507.22171
8
1
true
https://github.com/CjangCjengh/Generic_Persona
arXiv.org
0.379
d422d6c1cc9967ba5dc818d0636667a38fa381d0d7f215ee46de2d7a29bacec8
[ "arxiv", "semantic_scholar" ]
SpecASR: Accelerating LLM-based Automatic Speech Recognition via Speculative Decoding
Large language model (LLM)-based automatic speech recognition (ASR) has recently attracted a lot of attention due to its high recognition accuracy and enhanced multi-dialect support. However, the high decoding latency of LLMs challenges the real-time ASR requirements. Although speculative decoding has been explored for...
[ "Linye Wei", "Shuzhang Zhong", "Songqiang Xu", "Runsheng Wang", "Ru Huang", "Meng Li" ]
[ "eess.AS", "cs.SD" ]
[ "Engineering", "Computer Science" ]
2025-07-24T00:00:00
https://arxiv.org/abs/2507.18181
https://arxiv.org/pdf/2507.18181v2
2507.18181
10.1109/DAC63849.2025.11132579
3
0
false
null
Design Automation Conference
0.2406
f26ab641d82fc6750345b54d32f8c8e45e634b5d5fe1c950b291d0bdade436bc
[ "arxiv", "semantic_scholar" ]
DSSD: Efficient Edge-Device LLM Deployment and Collaborative Inference via Distributed Split Speculative Decoding
Large language models (LLMs) have transformed natural language processing but face critical deployment challenges in device-edge systems due to resource limitations and communication overhead. To address these issues, collaborative frameworks have emerged that combine small language models (SLMs) on devices with LLMs a...
[ "Jiahong Ning", "Ce Zheng", "Tingting Yang" ]
[ "eess.SP" ]
[ "Engineering" ]
2025-07-16T00:00:00
https://arxiv.org/abs/2507.12000
https://arxiv.org/pdf/2507.12000v2
2507.12000
null
15
1
true
https://github.com/JasonNing96/DSSD-Efficient-Edge-Computing
null
0.301
f9e6dd0ec9d17211d214885a175a019a55f6ebec66d0e2c187983e23c9a2083d
[ "arxiv", "semantic_scholar" ]
TPP-SD: Accelerating Transformer Point Process Sampling with Speculative Decoding
We propose TPP-SD, a novel approach that accelerates Transformer temporal point process (TPP) sampling by adapting speculative decoding (SD) techniques from language models. By identifying the structural similarities between thinning algorithms for TPPs and speculative decoding for language models, we develop an effici...
[ "Shukai Gong", "Yiyang Fu", "Fengyuan Ran", "Quyu Kong", "Feng Zhou" ]
[ "cs.LG", "stat.ML" ]
[ "Computer Science", "Mathematics" ]
2025-07-12T00:00:00
https://arxiv.org/abs/2507.09252
https://arxiv.org/pdf/2507.09252v3
2507.09252
10.48550/arXiv.2507.09252
2
0
false
null
arXiv.org
0.2269
4d126d963477bc25ccf29dd451b567121cff15502148c062dcb405cdbcb2c70b
[ "arxiv", "semantic_scholar" ]
On Evaluating Performance of LLM Inference Serving Systems
The rapid evolution of Large Language Model (LLM) inference systems has yielded significant efficiency improvements. However, our systematic analysis reveals that current evaluation methodologies frequently exhibit fundamental flaws, often manifesting as common evaluation anti-patterns that obscure true performance cha...
[ "Amey Agrawal", "Nitin Kedia", "Anmol Agarwal", "Jayashree Mohan", "Nipun Kwatra", "Souvik Kundu", "Ramachandran Ramjee", "Alexey Tumanov" ]
[ "cs.LG", "cs.AI", "cs.DC" ]
[ "Computer Science" ]
2025-07-11T00:00:00
https://arxiv.org/abs/2507.09019
https://arxiv.org/pdf/2507.09019v1
2507.09019
10.48550/arXiv.2507.09019
12
0
false
null
arXiv.org
0.2785
d98830d27b072526e589de5b88b174d3dea6db03e181cb63aecb1157d517a1b0
[ "arxiv", "semantic_scholar" ]
On the Convergence Speed of Spatially Coupled LDPC Ensembles Under Window Decoding
It is known that windowed decoding (WD) can effectively balance the performance and complexity of spatially coupled low-density parity-check (LDPC) codes. In this study, we show that information can propagate in a wave-like manner at a constant speed under WD. Additionally, we provide an upper bound for the information...
[ "Qingqing Peng", "Dongxu Chang", "Guanghui Wang", "Guiying Yan" ]
[ "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2025-07-09T00:00:00
https://arxiv.org/abs/2507.06635
https://arxiv.org/pdf/2507.06635v1
2507.06635
10.1109/ITW62417.2025.11240495
0
0
false
null
Information Theory Workshop
0.2234
fd59f8b8e506cc9333dfce85a86e5d3b2b6848c31954a0c821c378a20756df50
[ "arxiv", "semantic_scholar" ]
Cascade: Token-Sharded Private LLM Inference
As LLMs continue to increase in parameter size, the computational resources required to run them are available to fewer parties. Therefore, third-party inference services -- where LLMs are hosted by third parties with significant computational resources -- are becoming increasingly popular. However, third party inferen...
[ "Rahul Thomas", "Louai Zahran", "Erica Choi", "Akilesh Potti", "Micah Goldblum", "Arka Pal" ]
[ "cs.LG", "cs.CR" ]
[ "Computer Science" ]
2025-07-07T00:00:00
https://arxiv.org/abs/2507.05228
https://arxiv.org/pdf/2507.05228v1
2507.05228
10.48550/arXiv.2507.05228
0
0
false
null
arXiv.org
0.2211
acc11b4c8d216e52fc403a24e82d5b0663ad59410c0c58a7b715eb71805f270a
[ "arxiv", "semantic_scholar" ]
FlowSpec: Continuous Pipelined Speculative Decoding for Efficient Distributed LLM Inference
Distributed inference serves as a promising approach to enabling the inference of large language models (LLMs) at the network edge. It distributes the inference process to multiple devices to ensure that the LLMs can fit into the device memory. Recent pipeline-based approaches have the potential to parallelize communic...
[ "Xing Liu", "Lizhuo Luo", "Ming Tang", "Chao Huang", "Xu Chen" ]
[ "cs.DC", "cs.AI" ]
[ "Computer Science" ]
2025-07-03T00:00:00
https://arxiv.org/abs/2507.02620
https://arxiv.org/pdf/2507.02620v3
2507.02620
10.48550/arXiv.2507.02620
3
0
true
https://github.com/Leosang-lx/FlowSpec#}{https://github.com/Leosang-lx/FlowSpec\#}
arXiv.org
0.3347
03c79b7304a4a68c2e95603bf14f6f657d890ae10004856802a230949e8e5ce8
[ "arxiv", "semantic_scholar" ]
OmniDraft: A Cross-vocabulary, Online Adaptive Drafter for On-device Speculative Decoding
Speculative decoding generally dictates having a small, efficient draft model that is either pretrained or distilled offline to a particular target model series, for instance, Llama or Qwen models. However, within online deployment settings, there are two major challenges: 1) usage of a target model that is incompatibl...
[ "Ramchalam Kinattinkara Ramakrishnan", "Zhaocong Yuan", "Shaojie Zhuo", "Chen Feng", "Yicheng Lin", "Chenzheng Su", "Xiaopeng Zhang" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2025-07-03T00:00:00
https://arxiv.org/abs/2507.02659
https://arxiv.org/pdf/2507.02659v3
2507.02659
10.48550/arXiv.2507.02659
2
1
false
null
arXiv.org
0.2166
2709f59c07981dedde83e331f291c26d02eab73f5d5018db8144488ffb26dbcd
[ "arxiv", "semantic_scholar" ]
LogitSpec: Accelerating Retrieval-based Speculative Decoding via Next Next Token Speculation
Speculative decoding (SD), where a small draft model is employed to propose draft tokens in advance and then the target model validates them in parallel, has emerged as a promising technique for LLM inference acceleration. Many endeavors to improve SD are to eliminate the need for a draft model and generate draft token...
[ "Tianyu Liu", "Qitan Lv", "Hao Li", "Xing Gao", "Xiao Sun", "Xiaoyan Sun" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-07-02T00:00:00
https://arxiv.org/abs/2507.01449
https://arxiv.org/pdf/2507.01449v3
2507.01449
10.48550/arXiv.2507.01449
9
0
true
https://github.com/smart-lty/LogitSpec
arXiv.org
0.3329
10de40d14a89417d36b9abe9e64486d9c28a6ddb30f0fd56675e1db16a1a3cd2
[ "arxiv", "semantic_scholar" ]
Quantize-Sample-and-Verify: LLM Acceleration via Adaptive Edge-Cloud Speculative Decoding
In edge-cloud speculative decoding (SD), edge devices equipped with small language models (SLMs) generate draft tokens that are verified by large language models (LLMs) in the cloud. A key bottleneck in such systems is the limited communication bandwidth between edge and cloud, which necessitates quantization of the in...
[ "Guangyi Zhang", "Yunlong Cai", "Guanding Yu", "Petar Popovski", "Osvaldo Simeone" ]
[ "eess.SP" ]
[ "Engineering", "Computer Science" ]
2025-07-01T00:00:00
https://arxiv.org/abs/2507.00605
https://arxiv.org/pdf/2507.00605v3
2507.00605
10.1109/LCOMM.2026.3651580
3
0
false
null
IEEE Communications Letters
0.2143
2c284cbcbe0ae757f1f882c4c5f4f452c14971af867075f9d3c6e700c46a2b72
[ "arxiv", "semantic_scholar" ]
VOCABTRIM: Vocabulary Pruning for Efficient Speculative Decoding in LLMs
In this paper, we introduce a simple training-free technique to improve the performance of drafter-based speculative decoding (SpD) methods that incorporates language modeling head (LM head) during drafting process. A drafter-based speculative decoding leverages one or more smaller language models, a.k.a. drafters or d...
[ "Raghavv Goel", "Sudhanshu Agrawal", "Mukul Gagrani", "Junyoung Park", "Yifan Zao", "He Zhang", "Tian Liu", "Yiping Yang", "Xin Yuan", "Jiuyan Lu", "Chris Lott", "Mingu Lee" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2025-06-28T00:00:00
https://arxiv.org/abs/2506.22694
https://arxiv.org/pdf/2506.22694v2
2506.22694
10.48550/arXiv.2506.22694
9
4
false
null
arXiv.org
0.3495
e6883442625a1a239ccbfe2556c5fe1042d67d1625a9172f85b01bf7d1ccc512
[ "arxiv", "semantic_scholar" ]
Scaling Speculative Decoding with Lookahead Reasoning
Reasoning models excel by generating long chain-of-thoughts, but decoding the resulting thousands of tokens is slow. Token-level speculative decoding (SD) helps, but its benefit is capped, because the chance that an entire $γ$-token guess is correct falls exponentially as $γ$ grows. This means allocating more compute f...
[ "Yichao Fu", "Rui Ge", "Zelei Shao", "Zhijie Deng", "Hao Zhang" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2025-06-24T00:00:00
https://arxiv.org/abs/2506.19830
https://arxiv.org/pdf/2506.19830v1
2506.19830
10.48550/arXiv.2506.19830
10
2
true
https://github.com/hao-ai-lab/LookaheadReasoning
arXiv.org
0.3187
59b9ab8a4339927323ccfcfe0e162a7eb92a2d95747ee79e4d9fca1dcd8d0800
[ "arxiv", "semantic_scholar" ]
Utility-Driven Speculative Decoding for Mixture-of-Experts
GPU memory bandwidth is the main bottleneck for low-latency Large Language Model (LLM) inference. Speculative decoding leverages idle GPU compute by using a lightweight drafter to propose K tokens, which the LLM verifies in parallel, boosting token throughput. In conventional dense LLMs, all model weights are fetched e...
[ "Anish Saxena", "Po-An Tsai", "Hritvik Taneja", "Aamer Jaleel", "Moinuddin Qureshi" ]
[ "cs.DC", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-06-17T00:00:00
https://arxiv.org/abs/2506.20675
https://arxiv.org/pdf/2506.20675v1
2506.20675
10.48550/arXiv.2506.20675
5
0
false
null
arXiv.org
0.1982
4cee255d37b52f67629ac5e9ba942bc4a74f9e1768cf2f12d21483cfcd9e03f9
[ "arxiv", "semantic_scholar" ]
SwiftSpec: Ultra-Low Latency LLM Decoding by Scaling Asynchronous Speculative Decoding
Low-latency decoding for large language models (LLMs) is crucial for applications like chatbots and code assistants, yet generating long outputs remains slow in single-query settings. Prior work on speculative decoding (which combines a small draft model with a larger target model) and tensor parallelism has each accel...
[ "Ziyi Zhang", "Ziheng Jiang", "Chengquan Jiang", "Menghan Yu", "Size Zheng", "Haibin Lin", "Henry Hoffmann", "Xin Liu" ]
[ "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2025-06-12T00:00:00
https://arxiv.org/abs/2506.11309
https://arxiv.org/pdf/2506.11309v1
2506.11309
10.48550/arXiv.2506.11309
9
0
false
null
arXiv.org
0.25
62644411d7e6043fc8613a92a5d0dd6817192671126ef3a4bcc2ab1e4fed6d8d
[ "arxiv", "semantic_scholar" ]
SLED: A Speculative LLM Decoding Framework for Efficient Edge Serving
The growing gap between the increasing complexity of large language models (LLMs) and the limited computational budgets of edge devices poses a key challenge for efficient on-device inference, despite gradual improvements in hardware capabilities. Existing strategies, such as aggressive quantization, pruning, or remote...
[ "Xiangchen Li", "Dimitrios Spatharakis", "Saeid Ghafouri", "Jiakun Fan", "Hans Vandierendonck", "Deepu John", "Bo Ji", "Dimitrios Nikolopoulos" ]
[ "cs.DC", "cs.AI", "cs.LG", "cs.NI" ]
[ "Computer Science" ]
2025-06-11T00:00:00
https://arxiv.org/abs/2506.09397
https://arxiv.org/pdf/2506.09397v5
2506.09397
10.1145/3769102.3770608
14
0
false
null
IFIP International Information Security Conference
0.294
2fca7375b69324286b366eeac3a3ac3a13d246e78ff08a497ff95adf7c868c56
[ "arxiv", "semantic_scholar" ]
Tokenized Bandit for LLM Decoding and Alignment
We introduce the tokenized linear bandit (TLB) and multi-armed bandit (TMAB), variants of linear and stochastic multi-armed bandit problems inspired by LLM decoding and alignment. In these problems, at each round $t \in [T]$, a user submits a query (context), and the decision maker (DM) sequentially selects a token irr...
[ "Suho Shin", "Chenghao Yang", "Haifeng Xu", "Mohammad T. Hajiaghayi" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-06-08T00:00:00
https://arxiv.org/abs/2506.07276
https://arxiv.org/pdf/2506.07276v1
2506.07276
10.48550/arXiv.2506.07276
4
1
false
null
International Conference on Machine Learning
0.1879
d8ef199644ac8edfa91b31da024db82fdecc3a2bbf7d67ad738ac90a15c4b566
[ "arxiv", "semantic_scholar" ]
Saffron-1: Safety Inference Scaling
Existing safety assurance research has primarily focused on training-phase alignment to instill safe behaviors into LLMs. However, recent studies have exposed these methods' susceptibility to diverse jailbreak attacks. Concurrently, inference scaling has significantly advanced LLM reasoning capabilities but remains une...
[ "Ruizhong Qiu", "Gaotang Li", "Tianxin Wei", "Jingrui He", "Hanghang Tong" ]
[ "cs.LG", "cs.AI", "cs.CR" ]
[ "Computer Science" ]
2025-06-06T00:00:00
https://arxiv.org/abs/2506.06444
https://arxiv.org/pdf/2506.06444v2
2506.06444
null
0
0
true
https://github.com/q-rz/saffron
null
0.2194
4698d1054b05f5c534029fd6af09c688963224b517d759940013f6d1a5511b9e
[ "arxiv", "semantic_scholar" ]
Guided Speculative Inference for Efficient Test-Time Alignment of LLMs
We propose Guided Speculative Inference (GSI), a novel algorithm for efficient reward-guided decoding in large language models. GSI combines soft best-of-$n$ test-time scaling with a reward model $r(x,y)$ and speculative samples from a small auxiliary model $π_S(y\mid x)$. We provably approximate both the optimal tilte...
[ "Jonathan Geuter", "Youssef Mroueh", "David Alvarez-Melis" ]
[ "cs.LG", "stat.ML" ]
[ "Computer Science", "Mathematics" ]
2025-06-04T00:00:00
https://arxiv.org/abs/2506.04118
https://arxiv.org/pdf/2506.04118v3
2506.04118
10.48550/arXiv.2506.04118
10
0
true
https://github.com/j-geuter/GSI
arXiv.org
0.2833
fe08fc4092c711017b17a89fb27ecc129373485c19cee6465d437b3a07667550
[ "arxiv", "semantic_scholar" ]
AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism
Large language models (LLMs) are increasingly used for long-content generation (e.g., long Chain-of-Thought reasoning) where decoding efficiency becomes a critical bottleneck: Autoregressive decoding is inherently limited by its sequential token generation process, where each token must be generated before the next can...
[ "Zhepei Wei", "Wei-Lin Chen", "Xinyu Zhu", "Yu Meng" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-06-04T00:00:00
https://arxiv.org/abs/2506.03700
https://arxiv.org/pdf/2506.03700v1
2506.03700
10.48550/arXiv.2506.03700
4
1
true
https://github.com/weizhepei/AdaDecode
International Conference on Machine Learning
0.2833
5cdbafd441b0b061c80b1cb46dd0e6430083407ed0a465b789cb1acc7f40b542
[ "arxiv", "semantic_scholar" ]
POSS: Position Specialist Generates Better Draft for Speculative Decoding
Speculative decoding accelerates Large Language Model (LLM) inference by using a small draft model to predict multiple tokens, and a large target model to verify these tokens in parallel. Recent studies leverage the hidden state of the target model to enhance draft model prediction accuracy. However, existing methods s...
[ "Langlin Huang", "Chengsong Huang", "Jixuan Leng", "Di Huang", "Jiaxin Huang" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-06-04T00:00:00
https://arxiv.org/abs/2506.03566
https://arxiv.org/pdf/2506.03566v1
2506.03566
10.48550/arXiv.2506.03566
2
0
true
https://github.com/shrango/PosS
arXiv.org
0.2833
4d5c930042067467b583f610a26655d7dadaf0962c6ce491f61a85f0fad218d5
[ "arxiv", "semantic_scholar" ]
EfficientEdit: Accelerating Code Editing via Edit-Oriented Speculative Decoding
Large Language Models (LLMs) have demonstrated remarkable capabilities in code editing, substantially enhancing software development productivity. However, the inherent complexity of code editing tasks forces existing approaches to rely on LLMs' autoregressive end-to-end generation, where decoding speed plays a critica...
[ "Peiding Wang", "Li Zhang", "Fang Liu", "Yinghao Zhu", "Wang Xu", "Lin Shi", "Xiaoli Lian", "Minxiao Li", "Bo Shen", "An Fu" ]
[ "cs.SE" ]
[ "Computer Science" ]
2025-06-03T00:00:00
https://arxiv.org/abs/2506.02780
https://arxiv.org/pdf/2506.02780v2
2506.02780
10.1109/ASE63991.2025.00215
3
0
true
https://github.com/zhu-zhu-ding/EfficientEdit
International Conference on Automated Software Engineering
0.2816
da97645368cb2c9e6d05c6ecbaa7af3bd7dd6250ec14fa59144fb9fb14f6fba3
[ "arxiv", "semantic_scholar" ]
Out-of-Vocabulary Sampling Boosts Speculative Decoding
Speculative decoding relies on fast and accurate drafters. Recent state-of-the-art language models employ larger and larger vocabularies, which significantly slows down drafters. One promising approach to boost the efficiency of speculative decoding is to use drafters with smaller vocabularies. However, existing sampli...
[ "Nadav Timor", "Jonathan Mamou", "Oren Pereg", "Hongyang Zhang", "David Harel" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2025-06-02T00:00:00
https://arxiv.org/abs/2506.03206
https://arxiv.org/pdf/2506.03206v1
2506.03206
10.48550/arXiv.2506.03206
2
0
false
null
arXiv.org
0.181
874e52828838f468316cc8e891a356c6678045d5a9b238734eee6674c6a960db
[ "arxiv", "semantic_scholar" ]
Mamba Drafters for Speculative Decoding
Speculative decoding has emerged as a promising approach to accelerating large language model (LLM) generation using a fast drafter while maintaining alignment with the target model's distribution. However, existing approaches face a trade-off: external drafters offer flexibility but can suffer from slower drafting, wh...
[ "Daewon Choi", "Seunghyuk Oh", "Saket Dingliwal", "Jihoon Tack", "Kyuyoung Kim", "Woomin Song", "Seojin Kim", "Insu Han", "Jinwoo Shin", "Aram Galstyan", "Shubham Katiyar", "Sravan Babu Bodapati" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-06-01T00:00:00
https://arxiv.org/abs/2506.01206
https://arxiv.org/pdf/2506.01206v1
2506.01206
10.48550/arXiv.2506.01206
3
0
false
null
Conference on Empirical Methods in Natural Language Processing
0.1799
71084d20e540158f60ef7789548939cd9ad6b697d7471784241443e8f6482eb7
[ "arxiv", "semantic_scholar" ]
Speculative Reward Model Boosts Decision Making Ability of LLMs Cost-Effectively
Effective decision-making in Large Language Models (LLMs) is essential for handling intricate tasks. However, existing approaches prioritize performance but often overlook the balance between effectiveness and computational cost. To address this, we first introduce the 3E Criteria to systematically assess the cost-effe...
[ "Jiawei Gu", "Shangsong Liang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-31T00:00:00
https://arxiv.org/abs/2506.00396
https://arxiv.org/pdf/2506.00396v1
2506.00396
10.48550/arXiv.2506.00396
1
0
false
null
Annual Meeting of the Association for Computational Linguistics
0.1788
74bcc2ccd109bbc84516624ed182d6ba9ba5baeb7bca4ec8a5a9cb01e88bb445
[ "arxiv", "semantic_scholar" ]
Cross-Attention Speculative Decoding
Speculative decoding (SD) is a widely adopted approach for accelerating inference in large language models (LLMs), particularly when the draft and target models are well aligned. However, state-of-the-art SD methods typically rely on tightly coupled, self-attention-based Transformer decoders, often augmented with auxil...
[ "Wei Zhong", "Manasa Bharadwaj", "Yixiao Wang", "Yipeng Ji", "Chul Lee" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-05-30T00:00:00
https://arxiv.org/abs/2505.24544
https://arxiv.org/pdf/2505.24544v4
2505.24544
10.48550/arXiv.2505.24544
1
0
false
null
arXiv.org
0.1776
2ff89627942428ce4fc33b1f6eb6811df97ed2ea05c84693bb23d5a8f387c873
[ "arxiv", "semantic_scholar" ]
CLaSp: In-Context Layer Skip for Self-Speculative Decoding
Speculative decoding (SD) is a promising method for accelerating the decoding process of Large Language Models (LLMs). The efficiency of SD primarily hinges on the consistency between the draft model and the verify model. However, existing drafting approaches typically require additional modules to be trained, which ca...
[ "Longze Chen", "Renke Shan", "Huiming Wang", "Lu Wang", "Ziqiang Liu", "Run Luo", "Jiawei Wang", "Hamid Alinejad-Rokny", "Min Yang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-30T00:00:00
https://arxiv.org/abs/2505.24196
https://arxiv.org/pdf/2505.24196v1
2505.24196
10.48550/arXiv.2505.24196
7
2
false
null
Annual Meeting of the Association for Computational Linguistics
0.2386
9f29098256d9ca7169695ee9ceaaa861a3f9309f7b35d57d5c3c25e8d30d8cd3
[ "arxiv", "semantic_scholar" ]
Ghidorah: Fast LLM Inference on Edge with Speculative Decoding and Hetero-Core Parallelism
In-situ LLM inference on end-user devices has gained significant interest due to its privacy benefits and reduced dependency on external infrastructure. However, as the decoding process is memory-bandwidth-bound, the diverse processing units in modern end-user devices cannot be fully exploited, resulting in slow LLM in...
[ "Jinhui Wei", "Ye Huang", "Yuhui Zhou", "Jiazhi Jiang", "Jiangsu Du", "Yutong Lu" ]
[ "cs.DC" ]
[ "Computer Science" ]
2025-05-29T00:00:00
https://arxiv.org/abs/2505.23219
https://arxiv.org/pdf/2505.23219v2
2505.23219
10.1109/ICCD65941.2025.00043
1
0
false
null
ICCD
0.1765
1196216932e701c81a5f2fd544c7a856a00633534f51c6722f91decec81b03a5
[ "arxiv", "semantic_scholar" ]
DINGO: Constrained Inference for Diffusion LLMs
Diffusion LLMs have emerged as a promising alternative to conventional autoregressive LLMs, offering significant potential for improved runtime efficiency. However, existing diffusion models lack the ability to provably enforce user-specified formal constraints, such as regular expressions, which makes them unreliable ...
[ "Tarun Suresh", "Debangshu Banerjee", "Shubham Ugare", "Sasa Misailovic", "Gagandeep Singh" ]
[ "cs.LG", "cs.PL", "cs.SE" ]
[ "Computer Science" ]
2025-05-29T00:00:00
https://arxiv.org/abs/2505.23061
https://arxiv.org/pdf/2505.23061v1
2505.23061
10.48550/arXiv.2505.23061
8
1
false
null
arXiv.org
0.2386
8da9a8cdfb15682fbc485755ae1b5428f504050c6ced422164b531f5663a7fad
[ "arxiv", "semantic_scholar" ]
Speculative Decoding Meets Quantization: Compatibility Evaluation and Hierarchical Framework Design
Speculative decoding and quantization effectively accelerate memory-bound inference of large language models. Speculative decoding mitigates the memory bandwidth bottleneck by verifying multiple tokens within a single forward pass, which increases computational effort. Quantization achieves this optimization by compres...
[ "Yudi Zhang", "Weilin Zhao", "Xu Han", "Tiejun Zhao", "Wang Xu", "Hailong Cao", "Conghui Zhu" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-05-28T00:00:00
https://arxiv.org/abs/2505.22179
https://arxiv.org/pdf/2505.22179v2
2505.22179
10.48550/arXiv.2505.22179
2
0
true
https://github.com/AI9Stars/SpecMQuant
arXiv.org
0.2709
d0c45ed4e43af45f82f41a0619860bc541db51b4fc50c61709407786c315229b
[ "arxiv", "semantic_scholar" ]
Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding
Diffusion-based large language models (Diffusion LLMs) have shown promise for non-autoregressive text generation with parallel decoding capabilities. However, the practical inference speed of open-sourced Diffusion LLMs often lags behind autoregressive models due to the lack of Key-Value (KV) Cache and quality degradat...
[ "Chengyue Wu", "Hao Zhang", "Shuchen Xue", "Zhijian Liu", "Shizhe Diao", "Ligeng Zhu", "Ping Luo", "Song Han", "Enze Xie" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-28T00:00:00
https://arxiv.org/abs/2505.22618
https://arxiv.org/pdf/2505.22618v3
2505.22618
10.48550/arXiv.2505.22618
311
82
true
null
arXiv.org
0.9595
d8697f3996050709469d1ec0dd35c9c83d5e75ef77b2a0603c2c606bf8b3dbb9
[ "arxiv", "semantic_scholar" ]
SpecExtend: A Drop-in Enhancement for Speculative Decoding of Long Sequences
Speculative decoding is a widely used technique for accelerating inference in large language models (LLMs), but its performance degrades as input length grows, with significant drops even at moderate lengths. Yet, this early degradation has remained largely underexplored. We introduce SpecExtend, a drop-in enhancement ...
[ "Jungyoub Cha", "Hyunjong Kim", "Sungzoon Cho" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-05-27T00:00:00
https://arxiv.org/abs/2505.20776
https://arxiv.org/pdf/2505.20776v4
2505.20776
10.48550/arXiv.2505.20776
2
0
true
https://github.com/jycha98/SpecExtend
arXiv.org
0.2692
625fbff4f44ed7c5296ae2f6873b21382ecac6a5adadedcdf50da5e70e39a239
[ "arxiv", "semantic_scholar" ]
Fast and Cost-effective Speculative Edge-Cloud Decoding with Early Exits
Large Language Models (LLMs) enable various applications on edge devices such as smartphones, wearables, and embodied robots. However, their deployment often depends on expensive cloud-based APIs, creating high operational costs, which limit access for smaller organizations and raise sustainability concerns. Certain LL...
[ "Yeshwanth Venkatesha", "Souvik Kundu", "Priyadarshini Panda" ]
[ "cs.RO", "cs.AI", "cs.DC" ]
[ "Computer Science" ]
2025-05-27T00:00:00
https://arxiv.org/abs/2505.21594
https://arxiv.org/pdf/2505.21594v1
2505.21594
10.48550/arXiv.2505.21594
11
1
false
null
null
0.2698
3a0a51d7fa8eb3b28c631ee8cd0731056da881c7a554f0f9ac49196d788974e7
[ "arxiv", "semantic_scholar" ]
HAMburger: Accelerating LLM Inference via Token Smashing
The growing demand for efficient Large Language Model (LLM) inference requires a holistic optimization on algorithms, systems, and hardware. However, very few works have fundamentally changed the generation pattern: each token needs one forward pass and one KV cache. This can be sub-optimal because we found that LLMs a...
[ "Jingyu Liu", "Ce Zhang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-26T00:00:00
https://arxiv.org/abs/2505.20438
https://arxiv.org/pdf/2505.20438v1
2505.20438
10.48550/arXiv.2505.20438
6
1
false
null
arXiv.org
0.2113
6afdd01eb8661345d865259684968da6504b0bb13a0e7c61598822dad82a788e
[ "arxiv", "semantic_scholar" ]
MoESD: Unveil Speculative Decoding's Potential for Accelerating Sparse MoE
Large Language Models (LLMs) have achieved remarkable success across many applications, with Mixture of Experts (MoE) models demonstrating great potential. Compared to traditional dense models, MoEs achieve better performance with less computation. Speculative decoding (SD) is a widely used technique to accelerate LLM ...
[ "Zongle Huang", "Lei Zhu", "Zongyuan Zhan", "Ting Hu", "Weikai Mao", "Xianzhi Yu", "Yongpan Liu", "Tianyu Zhang" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-05-26T00:00:00
https://arxiv.org/abs/2505.19645
https://arxiv.org/pdf/2505.19645v4
2505.19645
10.48550/arXiv.2505.19645
9
0
false
null
arXiv.org
0.25
5b215da05d0a4327df82fb95463fbd3b4684e88b69d41f90d9a66334247cd2ae
[ "arxiv", "semantic_scholar" ]
DREAM: Drafting with Refined Target Features and Entropy-Adaptive Cross-Attention Fusion for Multimodal Speculative Decoding
Speculative decoding (SD) has emerged as a powerful method for accelerating autoregressive generation in large language models (LLMs), yet its integration into vision-language models (VLMs) remains underexplored. We introduce DREAM, a novel speculative decoding framework tailored for VLMs that combines three key innova...
[ "Yunhai Hu", "Tianhua Xia", "Zining Liu", "Rahul Raman", "Xingyu Liu", "Bo Bao", "Eric Sather", "Vithursan Thangarasa", "Sai Qian Zhang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-25T00:00:00
https://arxiv.org/abs/2505.19201
https://arxiv.org/pdf/2505.19201v3
2505.19201
10.48550/arXiv.2505.19201
8
1
true
https://github.com/SAI-Lab-NYU/DREAM.git
arXiv.org
0.2656
febcbb40753457f6a03531857e015cf0b2c86b1cc4081e40c58036c049225ef4
[ "arxiv", "semantic_scholar" ]
Think Before You Accept: Semantic Reflective Verification for Faster Speculative Decoding
Large language models (LLMs) suffer from high inference latency due to the auto-regressive decoding process. Speculative decoding accelerates inference by generating multiple draft tokens using a lightweight model and verifying them in parallel. However, existing verification methods rely heavily on distributional cons...
[ "Yixuan Wang", "Yijun Liu", "Shiyu ji", "Yuzhuang Xu", "Yang Xu", "Qingfu Zhu", "Wanxiang Che" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-05-24T00:00:00
https://arxiv.org/abs/2505.18629
https://arxiv.org/pdf/2505.18629v1
2505.18629
10.48550/arXiv.2505.18629
5
1
false
null
arXiv.org
0.1945
56268a8b2bb2defcd8a6f5a472dc49a1b25baf01282358ad82f808106bcf39fa
[ "arxiv", "semantic_scholar" ]
A Survey of LLM $\times$ DATA
The integration of large language model (LLM) and data management (DATA) is rapidly redefining both domains. In this survey, we comprehensively review the bidirectional relationships. On the one hand, DATA4LLM, spanning large-scale data processing, storage, and serving, feeds LLMs with high quality, diversity, and time...
[ "Xuanhe Zhou", "Junxuan He", "Wei Zhou", "Haodong Chen", "Zirui Tang", "Haoyu Zhao", "Xin Tong", "Guoliang Li", "Youmin Chen", "Jun Zhou", "Zhaojun Sun", "Binyuan Hui", "Shuo Wang", "Conghui He", "Zhiyuan Liu", "Jingren Zhou", "Fan Wu" ]
[ "cs.DB", "cs.AI", "cs.CL", "cs.IR", "cs.LG" ]
[ "Computer Science" ]
2025-05-24T00:00:00
https://arxiv.org/abs/2505.18458
https://arxiv.org/pdf/2505.18458v3
2505.18458
null
1
0
true
https://github.com/weAIDB/awesome-data-llm
null
0.2018
6b408b0f7041c30a9c4b35fed8034860dede9999c4702ca30d3c6a7036adee59
[ "arxiv", "semantic_scholar" ]
RoleRAG: Enhancing LLM Role-Playing via Graph Guided Retrieval
Large Language Models (LLMs) have shown promise in character imitation, enabling immersive and engaging conversations. However, they often generate content that is irrelevant or inconsistent with a character's background. We attribute these failures to: (1) the inability to accurately recall character-specific knowledg...
[ "Yongjie Wang", "Jonathan Leung", "Zhiqi Shen" ]
[ "cs.AI" ]
[ "Computer Science" ]
2025-05-24T00:00:00
https://arxiv.org/abs/2505.18541
https://arxiv.org/pdf/2505.18541v1
2505.18541
10.48550/arXiv.2505.18541
8
0
false
null
arXiv.org
0.2386
9cc1c5c242da58e18e22c66688df5aa9cca6e8699efecd57d0411b8bf8b8f012
[ "arxiv", "semantic_scholar" ]
An Attack to Break Permutation-Based Private Third-Party Inference Schemes for LLMs
Recent advances in Large Language Models (LLMs) have led to the widespread adoption of third-party inference services, raising critical privacy concerns. Existing methods of performing private third-party inference, such as Secure Multiparty Computation (SMPC), often rely on cryptographic methods. However, these method...
[ "Rahul Thomas", "Louai Zahran", "Erica Choi", "Akilesh Potti", "Micah Goldblum", "Arka Pal" ]
[ "cs.CR", "cs.LG" ]
[ "Computer Science" ]
2025-05-23T00:00:00
https://arxiv.org/abs/2505.18332
https://arxiv.org/pdf/2505.18332v1
2505.18332
10.48550/arXiv.2505.18332
2
0
false
null
arXiv.org
0.1696
dba740c4fbc963adc8ce65e393c7a45636fa2f85b5eb4c59b17d93c9b454c287
[ "arxiv", "semantic_scholar" ]
Reinforcement Speculative Decoding for Fast Ranking
Large Language Models (LLMs) have been widely adopted in ranking systems such as information retrieval (IR) systems and recommender systems (RSs). To alleviate the latency of auto-regressive decoding, some studies explore the single (first) token decoding for ranking approximation, but they suffer from severe degradati...
[ "Yingpeng Du", "Tianjun Wei", "Zhu Sun", "Jie Zhang" ]
[ "cs.AI" ]
[ "Computer Science" ]
2025-05-23T00:00:00
https://arxiv.org/abs/2505.20316
https://arxiv.org/pdf/2505.20316v1
2505.20316
10.1145/3770854.3780197
4
0
false
null
Knowledge Discovery and Data Mining
0.1747
80997e8b8e921f7183e83e59b57f243090f1f2bf1358c7c2ce515dad577ee0bc
[ "arxiv", "semantic_scholar" ]
KNN-SSD: Enabling Dynamic Self-Speculative Decoding via Nearest Neighbor Layer Set Optimization
Speculative Decoding (SD) has emerged as a widely used paradigm to accelerate the inference of large language models (LLMs) without compromising generation quality. It works by efficiently drafting multiple tokens using a compact model and then verifying them in parallel using the target LLM. Notably, Self-Speculative ...
[ "Mingbo Song", "Heming Xia", "Jun Zhang", "Chak Tou Leong", "Qiancheng Xu", "Wenjie Li", "Sujian Li" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-22T00:00:00
https://arxiv.org/abs/2505.16162
https://arxiv.org/pdf/2505.16162v2
2505.16162
10.48550/arXiv.2505.16162
7
1
false
null
Conference of the European Chapter of the Association for Computational Linguistics
0.2258
3d88c8e0624bfcf7a85816dbfc7cb8fffbae4a820c8f0cc0ef6dfbe71c1db87f
[ "arxiv", "semantic_scholar" ]
BanditSpec: Adaptive Speculative Decoding via Bandit Algorithms
Speculative decoding has emerged as a popular method to accelerate the inference of Large Language Models (LLMs) while retaining their superior text generation performance. Previous methods either adopt a fixed speculative decoding configuration regardless of the prefix tokens, or train draft models in an offline or on...
[ "Yunlong Hou", "Fengzhuo Zhang", "Cunxiao Du", "Xuan Zhang", "Jiachun Pan", "Tianyu Pang", "Chao Du", "Vincent Y. F. Tan", "Zhuoran Yang" ]
[ "cs.LG", "cs.AI", "stat.ML" ]
[ "Computer Science", "Mathematics" ]
2025-05-21T00:00:00
https://arxiv.org/abs/2505.15141
https://arxiv.org/pdf/2505.15141v2
2505.15141
10.48550/arXiv.2505.15141
12
2
false
null
International Conference on Machine Learning
0.2785
41fbeadd3e9a1a262543b4bc6ee522f452d040120be0817b0f6e732c7cfae48c
[ "arxiv", "semantic_scholar" ]
Accelerating Autoregressive Speech Synthesis Inference With Speech Speculative Decoding
Modern autoregressive speech synthesis models leveraging language models have demonstrated remarkable performance. However, the sequential nature of next token prediction in these models leads to significant latency, hindering their deployment in scenarios where inference speed is critical. In this work, we propose Spe...
[ "Zijian Lin", "Yang Zhang", "Yougen Yuan", "Yuming Yan", "Jinjiang Liu", "Zhiyong Wu", "Pengfei Hu", "Qun Yu" ]
[ "cs.SD", "cs.AI", "eess.AS" ]
[ "Computer Science", "Engineering" ]
2025-05-21T00:00:00
https://arxiv.org/abs/2505.15380
https://arxiv.org/pdf/2505.15380v2
2505.15380
10.48550/arXiv.2505.15380
6
1
false
null
Interspeech
0.2113
349a19a188c37991f76e5642cbccf771803718665627aca56ea602949140ba22
[ "arxiv", "semantic_scholar" ]
Semi-Clairvoyant Scheduling of Speculative Decoding Requests to Minimize LLM Inference Latency
Speculative decoding accelerates Large Language Model (LLM) inference by employing a small speculative model (SSM) to generate multiple candidate tokens and verify them using the LLM in parallel. This technique has been widely integrated into LLM inference serving systems. However, inference requests typically exhibit ...
[ "Ruixiao Li", "Fahao Chen", "Peng Li" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-05-20T00:00:00
https://arxiv.org/abs/2505.17074
https://arxiv.org/pdf/2505.17074v1
2505.17074
10.48550/arXiv.2505.17074
0
0
false
null
International Joint Conference on Artificial Intelligence
0.1661
7804e2f5a3739bf3c46ca4cefd0542741de31932d3b973ee9fe8a66393d63819
[ "arxiv", "semantic_scholar" ]
STree: Speculative Tree Decoding for Hybrid State-Space Models
Speculative decoding is a technique to leverage hardware concurrency in order to enable multiple steps of token generation in a single forward pass, thus improving the efficiency of large-scale autoregressive (AR) Transformer models. State-space models (SSMs) are already more efficient than AR Transformers, since their...
[ "Yangchao Wu", "Zongyue Qin", "Alex Wong", "Stefano Soatto" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-05-20T00:00:00
https://arxiv.org/abs/2505.14969
https://arxiv.org/pdf/2505.14969v2
2505.14969
10.48550/arXiv.2505.14969
4
0
true
https://github.com/wyc1997/stree
arXiv.org
0.2568
e6f0c3891a75ddbcfb95e415fa33476df62d6648ec5f3722fe7c605b7ff2d87d
[ "arxiv", "semantic_scholar" ]
Speculative Decoding Reimagined for Multimodal Large Language Models
This paper introduces Multimodal Speculative Decoding (MSD) to accelerate Multimodal Large Language Models (MLLMs) inference. Speculative decoding has been shown to accelerate Large Language Models (LLMs) without sacrificing accuracy. However, current speculative decoding methods for MLLMs fail to achieve the same spee...
[ "Luxi Lin", "Zhihang Lin", "Zhanpeng Zeng", "Rongrong Ji" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2025-05-20T00:00:00
https://arxiv.org/abs/2505.14260
https://arxiv.org/pdf/2505.14260v1
2505.14260
10.48550/arXiv.2505.14260
6
3
true
https://github.com/Lyn-Lucy/MSD
arXiv.org
0.301
452cfdf70ebd5ed96dfd486408b4a99b4f8c5cc487ccce6a53b5f6a69b93045c
[ "arxiv", "semantic_scholar" ]
HeteroSpec: Leveraging Contextual Heterogeneity for Efficient Speculative Decoding
Autoregressive decoding inherently limits the inference throughput of Large Language Model (LLM) due to its sequential dependency. Speculative decoding mitigates this by verifying multiple predicted tokens in parallel, but its efficiency remains constrained by what we identify as verification heterogeneity -- the uneve...
[ "Siran Liu", "Yang Ye", "Qianchao Zhu", "Zane Cao", "Yongchao He" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-19T00:00:00
https://arxiv.org/abs/2505.13254
https://arxiv.org/pdf/2505.13254v2
2505.13254
10.48550/arXiv.2505.13254
1
0
false
null
arXiv.org
0.165
c4e3c1a68326dbb905e7c9aa82d85f85a3689de6da87a3d1365623758b352577
[ "arxiv", "semantic_scholar" ]
SpecFLASH: A Latent-Guided Semi-autoregressive Speculative Decoding Framework for Efficient Multimodal Generation
Large language models and large multimodal models (LLMs and LMMs) deliver strong generative performance but suffer from slow decoding, a problem that becomes more severe when handling visual inputs, whose sequences typically contain many more tokens with lower information density than text. Speculative decoding acceler...
[ "Zihua Wang", "Ruibo Li", "Haozhe Du", "Joey Tianyi Zhou", "Yu Zhang", "Xu Yang" ]
[ "cs.CV", "cs.MM" ]
[ "Computer Science" ]
2025-05-19T00:00:00
https://arxiv.org/abs/2505.12728
https://arxiv.org/pdf/2505.12728v3
2505.12728
null
1
0
true
https://github.com/ZihuaEvan/FlashSD/
null
0.195