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
3b4607797a56c785e0912b27a890e8bf85f3f3e29811ba5de7c9a8196c6caa68
[ "arxiv", "semantic_scholar" ]
Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification
Recent works have revealed the great potential of speculative decoding in accelerating the autoregressive generation process of large language models. The success of these methods relies on the alignment between draft candidates and the sampled outputs of the target model. Existing methods mainly achieve draft-target a...
[ "Jikai Wang", "Zhenxu Tian", "Juntao Li", "Qingrong Xia", "Xinyu Duan", "Zhefeng Wang", "Baoxing Huai", "Min Zhang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-19T00:00:00
https://arxiv.org/abs/2505.13204
https://arxiv.org/pdf/2505.13204v2
2505.13204
10.48550/arXiv.2505.13204
8
1
false
null
Conference on Empirical Methods in Natural Language Processing
0.2386
4eeda6b53d1657b1cf7de184f5e1f06c5b24026fc731a4fa38b9fa0c60ed494a
[ "arxiv", "semantic_scholar" ]
Traversal Verification for Speculative Tree Decoding
Speculative decoding is a promising approach for accelerating large language models. The primary idea is to use a lightweight draft model to speculate the output of the target model for multiple subsequent timesteps, and then verify them in parallel to determine whether the drafted tokens should be accepted or rejected...
[ "Yepeng Weng", "Qiao Hu", "Xujie Chen", "Li Liu", "Dianwen Mei", "Huishi Qiu", "Jiang Tian", "Zhongchao Shi" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-05-18T00:00:00
https://arxiv.org/abs/2505.12398
https://arxiv.org/pdf/2505.12398v2
2505.12398
10.48550/arXiv.2505.12398
6
0
false
null
arXiv.org
0.2113
84de75f78a7bda16010ed8a9e7b3da7528cb019d3602670e25ee17ef2db45363
[ "arxiv", "semantic_scholar" ]
SpecMemo: Speculative Decoding is in Your Pocket
Recent advancements in speculative decoding have demonstrated considerable speedup across a wide array of large language model (LLM) tasks. Speculative decoding inherently relies on sacrificing extra memory allocations to generate several candidate tokens, of which acceptance rate drives the speedup. However, deploying...
[ "Selin Yildirim", "Deming Chen" ]
[ "cs.LG", "cs.AI", "cs.DC" ]
[ "Computer Science" ]
2025-05-16T00:00:00
https://arxiv.org/abs/2506.01986
https://arxiv.org/pdf/2506.01986v1
2506.01986
10.48550/arXiv.2506.01986
0
0
false
null
arXiv.org
0.1616
7d116ea662927547c9a1fdfe569b8c728f1dd63762080583f87a7283d7243c1e
[ "arxiv", "semantic_scholar" ]
SpecBranch: Speculative Decoding via Hybrid Drafting and Rollback-Aware Branch Parallelism
Recently, speculative decoding (SD) has emerged as a promising technique to accelerate LLM inference by employing a small draft model to propose draft tokens in advance, and validating them in parallel with the large target model. However, the existing SD methods still remain fundamentally constrained by their serializ...
[ "Yuhao Shen", "Junyi Shen", "Quan Kong", "Tianyu Liu", "Yao Lu", "Cong Wang" ]
[ "cs.DC", "cs.AI" ]
[ "Computer Science" ]
2025-05-16T00:00:00
https://arxiv.org/abs/2506.01979
https://arxiv.org/pdf/2506.01979v4
2506.01979
null
12
0
false
null
null
0.2785
0a32a41125d14672f832133dfa4090cc77dea1d1b6035f7e47382f9998d02a91
[ "arxiv", "semantic_scholar" ]
MASSV: Multimodal Adaptation and Self-Data Distillation for Speculative Decoding of Vision-Language Models
Speculative decoding significantly accelerates language model inference by enabling a lightweight draft model to propose multiple tokens that a larger target model verifies simultaneously. However, applying this technique to vision-language models (VLMs) presents two fundamental challenges: small language models that c...
[ "Mugilan Ganesan", "Shane Segal", "Ankur Aggarwal", "Nish Sinnadurai", "Sean Lie", "Vithursan Thangarasa" ]
[ "cs.LG", "cs.CL", "cs.CV" ]
[ "Computer Science" ]
2025-05-15T00:00:00
https://arxiv.org/abs/2505.10526
https://arxiv.org/pdf/2505.10526v2
2505.10526
10.48550/arXiv.2505.10526
0
0
false
null
Conference on Empirical Methods in Natural Language Processing
0.1604
5e6c3e7feb57d72657e9ec2437560c424ddd21c135dc1bce2937a2bb410cdbc2
[ "arxiv", "semantic_scholar" ]
SpecOffload: Unlocking Latent GPU Capacity for LLM Inference on Resource-Constrained Devices
Efficient LLM inference on resource-constrained devices presents significant challenges in compute and memory utilization. Due to limited GPU memory, existing systems offload model weights to CPU memory, incurring substantial I/O overhead between the CPU and GPU. This leads to two major inefficiencies: (1) GPU cores ar...
[ "Xiangwen Zhuge", "Xu Shen", "Zeyu Wang", "Fan Dang", "Xuan Ding", "Danyang Li", "Yahui Han", "Tianxiang Hao", "Zheng Yang" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-05-15T00:00:00
https://arxiv.org/abs/2505.10259
https://arxiv.org/pdf/2505.10259v3
2505.10259
10.48550/arXiv.2505.10259
1
1
true
https://github.com/MobiSense/SpecOffload-public
arXiv.org
0.2479
3128cc6a9ba2fa0fd63a91c1e8ff9eaa9673b7930f229f6ba2546c7587b6533b
[ "arxiv", "semantic_scholar" ]
ELIS: Efficient LLM Iterative Scheduling System with Response Length Predictor
We propose ELIS, a serving system for Large Language Models (LLMs) featuring an Iterative Shortest Remaining Time First (ISRTF) scheduler designed to efficiently manage inference tasks with the shortest remaining tokens. Current LLM serving systems often employ a first-come-first-served scheduling strategy, which can l...
[ "Seungbeom Choi", "Jeonghoe Goo", "Eunjoo Jeon", "Mingyu Yang", "Minsung Jang" ]
[ "cs.DC", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-05-14T00:00:00
https://arxiv.org/abs/2505.09142
https://arxiv.org/pdf/2505.09142v1
2505.09142
10.48550/arXiv.2505.09142
7
0
false
null
arXiv.org
0.2258
84dd8a389eaf47c74725765a7b041096988401415ebd69f559639b91f423d5f5
[ "arxiv", "semantic_scholar" ]
Automatic Task Detection and Heterogeneous LLM Speculative Decoding
Speculative decoding, which combines a draft model with a target model, has emerged as an effective approach to accelerate large language model (LLM) inference. However, existing methods often face a trade-off between the acceptance rate and decoding speed in downstream tasks due to the limited capacity of the draft mo...
[ "Danying Ge", "Jianhua Gao", "Qizhi Jiang", "Yifei Feng", "Weixing Ji" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-13T00:00:00
https://arxiv.org/abs/2505.08600
https://arxiv.org/pdf/2505.08600v1
2505.08600
10.48550/arXiv.2505.08600
0
0
false
null
arXiv.org
0.1581
af0e472eecc7b3b4416d6be55cbbe9eec7bb07a4cf3df5ee65cf7e2bdfb979dc
[ "arxiv", "semantic_scholar" ]
SpecRouter: Adaptive Routing for Multi-Level Speculative Decoding in Large Language Models
Large Language Models (LLMs) present a critical trade-off between inference quality and computational cost: larger models offer superior capabilities but incur significant latency, while smaller models are faster but less powerful. Existing serving strategies often employ fixed model scales or static two-stage speculat...
[ "Hang Wu", "Jianian Zhu", "Yinghui Li", "Haojie Wang", "Biao Hou", "Jidong Zhai" ]
[ "cs.LG", "cs.DC" ]
[ "Computer Science" ]
2025-05-12T00:00:00
https://arxiv.org/abs/2505.07680
https://arxiv.org/pdf/2505.07680v1
2505.07680
10.48550/arXiv.2505.07680
1
0
false
null
arXiv.org
0.157
ff5672cdcffa20de10c8d92ecac2d9842be0fad501fc4b43442304f0ec7a87e3
[ "arxiv", "semantic_scholar" ]
Scaling Laws for Speculative Decoding
The escalating demand for efficient decoding in large language models (LLMs) is particularly critical for reasoning-intensive architectures like OpenAI-o3 and DeepSeek-R1, which depend on extended chain-of-thought reasoning. This study investigates speculative decoding techniques through dense LLM architectures to esta...
[ "Siyuan Yan", "Mo Zhu", "Guo-qing Jiang", "Jianfei Wang", "Jiaxing Chen", "Wentai Zhang", "Xiang Liao", "Xiao Cui", "Chen Zhang", "Zhuoran Song", "Ran Zhu" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-05-08T00:00:00
https://arxiv.org/abs/2505.07858
https://arxiv.org/pdf/2505.07858v1
2505.07858
10.48550/arXiv.2505.07858
3
1
false
null
arXiv.org
0.1524
aa29d28a1f46f1a2de01f61cdd67faf40632c4aad43f6f425c9591867abada65
[ "arxiv", "semantic_scholar" ]
SOAEsV2-7B/72B: Full-Pipeline Optimization for State-Owned Enterprise LLMs via Continual Pre-Training, Domain-Progressive SFT and Distillation-Enhanced Speculative Decoding
This study addresses key challenges in developing domain-specific large language models (LLMs) for Chinese state-owned assets and enterprises (SOAEs), where current approaches face three limitations: 1) constrained model capacity that limits knowledge integration and cross-task adaptability; 2) excessive reliance on do...
[ "Jingyang Deng", "Ran Chen", "Jo-Ku Cheng", "Jinwen Ma" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-05-07T00:00:00
https://arxiv.org/abs/2505.04723
https://arxiv.org/pdf/2505.04723v1
2505.04723
10.48550/arXiv.2505.04723
1
0
false
null
arXiv.org
0.1513
20f8beb58b7db52d32910b1513853ab3014089ebbac64d56bafe19b8a3170e5f
[ "arxiv", "semantic_scholar" ]
PipeSpec: Breaking Stage Dependencies in Hierarchical LLM Decoding
Speculative decoding accelerates large language model inference by using smaller draft models to generate candidate tokens for parallel verification. However, current approaches are limited by sequential stage dependencies that prevent full hardware utilization. We present PipeSpec, a framework that generalizes specula...
[ "Bradley McDanel", "Sai Qian Zhang", "Yunhai Hu", "Zining Liu" ]
[ "cs.AI", "cs.DC" ]
[ "Computer Science" ]
2025-05-02T00:00:00
https://arxiv.org/abs/2505.01572
https://arxiv.org/pdf/2505.01572v1
2505.01572
10.48550/arXiv.2505.01572
5
0
false
null
Annual Meeting of the Association for Computational Linguistics
0.1945
9752dc211fc57f85d7976f78d0d672dca1ab107ba0d1e5e6aa7e933709c15bda
[ "arxiv", "semantic_scholar" ]
Speculative Sampling via Exponential Races
Speculative decoding accelerates large language model inference using a smaller draft model. In this paper, we establish a surprising connection between speculative decoding and channel simulation, which aims at simulating a noisy channel using as few bits as possible. This connection allows us to provide an informatio...
[ "Szymon Kobus", "Deniz Gündüz" ]
[ "cs.CL", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2025-04-21T00:00:00
https://arxiv.org/abs/2504.15475
https://arxiv.org/pdf/2504.15475v1
2504.15475
10.48550/arXiv.2504.15475
1
0
false
null
Annual Meeting of the Association for Computational Linguistics
0.1329
8095e57bc85578032b7dc5eb627c54186cb79f0999ad826555f43ab05efbe4b7
[ "arxiv", "semantic_scholar" ]
MIST: A Co-Design Framework for Heterogeneous, Multi-Stage LLM Inference
Modern LLM serving now spans multi-stage pipelines including RAG retrieval and KV cache reuse, each with distinct compute, memory, and latency demands. Inference engines expose a large configuration space with no systematic navigation methodology, and exhaustively benchmarking configurations can exceed 40K in cloud cos...
[ "Abhimanyu Rajeshkumar Bambhaniya", "Hanjiang Wu", "Suvinay Subramanian", "Sudarshan Srinivasan", "Souvik Kundu", "Amir Yazdanbakhsh", "Midhilesh Elavazhagan", "Madhu Kumar", "Minlan Yu", "Arijit Raychowdhury", "Tushar Krishna" ]
[ "cs.AR", "cs.AI", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2025-04-14T00:00:00
https://arxiv.org/abs/2504.09775
https://arxiv.org/pdf/2504.09775v6
2504.09775
null
2
0
false
null
null
0.1193
1792d10cbdf711c47d677c52532ffdf6e7b1c0f551972561156d287ecedf90b7
[ "arxiv", "semantic_scholar" ]
SpecReason: Fast and Accurate Inference-Time Compute via Speculative Reasoning
Recent advances in inference-time compute have significantly improved performance on complex tasks by generating long chains of thought (CoTs) using Large Reasoning Models (LRMs). However, this improved accuracy comes at the cost of high inference latency due to the length of generated reasoning sequences and the autor...
[ "Rui Pan", "Yinwei Dai", "Zhihao Zhang", "Gabriele Oliaro", "Zhihao Jia", "Ravi Netravali" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-04-10T00:00:00
https://arxiv.org/abs/2504.07891
https://arxiv.org/pdf/2504.07891v2
2504.07891
10.48550/arXiv.2504.07891
57
12
true
https://github.com/ruipeterpan/specreason
arXiv.org
0.557
915bb99440a5c9610ec1a3bea89bcf3481b8232a3aa737c429cab050c6b7dd56
[ "arxiv", "semantic_scholar" ]
SPIRe: Boosting LLM Inference Throughput with Speculative Decoding
Speculative decoding (SD) has been shown to reduce the latency of autoregressive decoding (AD) by 2-3x for small batch sizes. However, increasing throughput and therefore reducing the cost per token requires decoding with large batch sizes. Recent work shows that SD can accelerate decoding with large batch sizes too if...
[ "Sanjit Neelam", "Daniel Heinlein", "Vaclav Cvicek", "Akshay Mishra", "Reiner Pope" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-04-08T00:00:00
https://arxiv.org/abs/2504.06419
https://arxiv.org/pdf/2504.06419v1
2504.06419
10.48550/arXiv.2504.06419
0
0
false
null
arXiv.org
0.118
7433730f6f8e81b2e4c9b38b2e2ea86ef83a71281ae262eeab79901d5b6c3346
[ "arxiv", "semantic_scholar" ]
DEL: Context-Aware Dynamic Exit Layer for Efficient Self-Speculative Decoding
Speculative Decoding (SD) is a widely used approach to accelerate the inference of large language models (LLMs) without reducing generation quality. It operates by first using a compact model to draft multiple tokens efficiently, followed by parallel verification using the target LLM. This approach leads to faster infe...
[ "Hossein Entezari Zarch", "Lei Gao", "Chaoyi Jiang", "Murali Annavaram" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2025-04-08T00:00:00
https://arxiv.org/abs/2504.05598
https://arxiv.org/pdf/2504.05598v2
2504.05598
10.48550/arXiv.2504.05598
6
2
true
https://github.com/hoenza/DEL
arXiv.org
0.2386
ecb96502b71d3bbac2182da1478f26fd93d76133bc5925d09aecfd890f732b71
[ "arxiv", "semantic_scholar" ]
SpecPipe: Accelerating Pipeline Parallelism-based LLM Inference with Speculative Decoding
The demand for large language model inference is rapidly increasing. Pipeline parallelism offers a cost-effective deployment strategy for distributed inference but suffers from high service latency. While incorporating speculative decoding to pipeline parallelism improves performance, it still faces challenges of low h...
[ "Haofei Yin", "Mengbai Xiao", "Tinghong Li", "Xiao Zhang", "Dongxiao Yu", "Guanghui Zhang" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-04-05T00:00:00
https://arxiv.org/abs/2504.04104
https://arxiv.org/pdf/2504.04104v2
2504.04104
null
4
0
false
null
null
0.1747
02f2ea2ebb8810f78d0f4caa8723904d254e6c8c1448c4f66cab6a98dfac0760
[ "arxiv", "semantic_scholar" ]
Token-Driven GammaTune: Adaptive Calibration for Enhanced Speculative Decoding
Speculative decoding accelerates large language model (LLM) inference by using a smaller draft model to propose tokens, which are then verified by a larger target model. However, selecting an optimal speculation length is critical for maximizing speedup while minimizing wasted computation. We introduce \textit{GammaTun...
[ "Aayush Gautam", "Susav Shrestha", "Narasimha Reddy" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-03-28T00:00:00
https://arxiv.org/abs/2504.00030
https://arxiv.org/pdf/2504.00030v3
2504.00030
10.48550/arXiv.2504.00030
2
0
false
null
arXiv.org
0.1193
23e13f6df3b0fa061484c1c935a46ab6730e4b6af458f300a07d991f10293b61
[ "arxiv", "semantic_scholar" ]
Collab: Controlled Decoding using Mixture of Agents for LLM Alignment
Alignment of Large Language models (LLMs) is crucial for safe and trustworthy deployment in applications. Reinforcement learning from human feedback (RLHF) has emerged as an effective technique to align LLMs to human preferences and broader utilities, but it requires updating billions of model parameters, which is comp...
[ "Souradip Chakraborty", "Sujay Bhatt", "Udari Madhushani Sehwag", "Soumya Suvra Ghosal", "Jiahao Qiu", "Mengdi Wang", "Dinesh Manocha", "Furong Huang", "Alec Koppel", "Sumitra Ganesh" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-03-27T00:00:00
https://arxiv.org/abs/2503.21720
https://arxiv.org/pdf/2503.21720v1
2503.21720
10.48550/arXiv.2503.21720
19
1
true
null
International Conference on Learning Representations
0.3253
00471c250e77fe0be58a76d2b261c4723d1f5b7b308d115d15a14208d444eaeb
[ "arxiv", "semantic_scholar" ]
Accelerate Parallelizable Reasoning via Parallel Decoding within One Sequence
Recent advances in reasoning models have demonstrated significant improvements in accuracy by employing detailed and comprehensive reasoning processes. However, generating these lengthy reasoning sequences is computationally expensive and time-consuming. To address this inefficiency, we leverage the inherent paralleliz...
[ "Yijiong Yu" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-03-26T00:00:00
https://arxiv.org/abs/2503.20533
https://arxiv.org/pdf/2503.20533v4
2503.20533
10.48550/arXiv.2503.20533
4
0
true
https://github.com/yuyijiong/parallel-decoding-in-one-sequence
Conference on Empirical Methods in Natural Language Processing
0.1747
3875b7afdcb3d338913a88b0842ee39ee285a2dedcc979a1a7700f14ece8b39a
[ "arxiv", "semantic_scholar" ]
A Multi-Model Adaptation of Speculative Decoding for Classification
The current study introduces a novel adaptation of speculative decoding, repurposed from generation to classification tasks. We propose a multi-model framework employing up to three lightweight worker models and a single, more robust judge model analogous to draft models and target model, respectively, in speculative d...
[ "Somnath Roy", "Padharthi Sreekar", "Srivatsa Narasimha", "Anubhav Anand" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-03-23T00:00:00
https://arxiv.org/abs/2503.18076
https://arxiv.org/pdf/2503.18076v1
2503.18076
10.48550/arXiv.2503.18076
0
0
false
null
arXiv.org
0.0997
8499012a8b1c554a4dca5e69b2c021fc1a6ec1d255aa7fd91ea53a1c2aa8c289
[ "arxiv", "semantic_scholar" ]
SPIN: Accelerating Large Language Model Inference with Heterogeneous Speculative Models
Speculative decoding has been shown as an effective way to accelerate Large Language Model (LLM) inference by using a Small Speculative Model (SSM) to generate candidate tokens in a so-called speculation phase, which are subsequently verified by the LLM in a verification phase. However, current state-of-the-art specula...
[ "Fahao Chen", "Peng Li", "Tom H. Luan", "Zhou Su", "Jing Deng" ]
[ "cs.DC" ]
[ "Computer Science" ]
2025-03-20T00:00:00
https://arxiv.org/abs/2503.15921
https://arxiv.org/pdf/2503.15921v1
2503.15921
10.1109/INFOCOM55648.2025.11044522
14
0
false
null
IEEE Conference on Computer Communications
0.294
0c403e5639e006c5d8916f31b1e6f74a4fee97444e3a51081acf4cabb849927c
[ "arxiv", "semantic_scholar" ]
SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs
Transformer-based large language models (LLMs) have already achieved remarkable results on long-text tasks, but the limited GPU memory (VRAM) resources struggle to accommodate the linearly growing demand for key-value (KV) cache as the sequence length increases, which has become a bottleneck for the application of LLMs...
[ "Shibo Jie", "Yehui Tang", "Kai Han", "Zhi-Hong Deng", "Jing Han" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-03-20T00:00:00
https://arxiv.org/abs/2503.16163
https://arxiv.org/pdf/2503.16163v1
2503.16163
10.48550/arXiv.2503.16163
5
0
false
null
International Conference on Machine Learning
0.1945
138aa166ad0bd48ec771b6ccd2e0111183c5a0d27dc21568bb9146cb9db0f511
[ "arxiv", "semantic_scholar" ]
Speculative Decoding for Verilog: Speed and Quality, All in One
The rapid advancement of large language models (LLMs) has revolutionized code generation tasks across various programming languages. However, the unique characteristics of programming languages, particularly those like Verilog with specific syntax and lower representation in training datasets, pose significant challeng...
[ "Changran Xu", "Yi Liu", "Yunhao Zhou", "Shan Huang", "Ningyi Xu", "Qiang Xu" ]
[ "cs.LG", "cs.AR", "cs.CL" ]
[ "Computer Science" ]
2025-03-18T00:00:00
https://arxiv.org/abs/2503.14153
https://arxiv.org/pdf/2503.14153v1
2503.14153
10.1109/DAC63849.2025.11133030
2
0
false
null
Design Automation Conference
0.1193
4577ae26bc97202f5a95529643c096ad80ed4cb0c09395ca7d7c8d4d9071b16a
[ "arxiv", "semantic_scholar" ]
ML-SpecQD: Multi-Level Speculative Decoding with Quantized Drafts
Speculative decoding (SD) has emerged as a method to accelerate LLM inference without sacrificing any accuracy over the 16-bit model inference. In a typical SD setup, the idea is to use a full-precision, small, fast model as "draft" to generate the next few tokens and use the "target" large model to verify the draft-ge...
[ "Evangelos Georganas", "Dhiraj Kalamkar", "Alexander Kozlov", "Alexander Heinecke" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-03-17T00:00:00
https://arxiv.org/abs/2503.13565
https://arxiv.org/pdf/2503.13565v1
2503.13565
10.48550/arXiv.2503.13565
7
0
false
null
arXiv.org
0.2258
c6bb892be526c541d84bfbe0e37fe50302d8fc407564b41896bd22a5398705ec
[ "arxiv", "semantic_scholar" ]
Collaborative Speculative Inference for Efficient LLM Inference Serving
Speculative inference is a promising paradigm employing small speculative models (SSMs) as drafters to generate draft tokens, which are subsequently verified in parallel by the target large language model (LLM). This approach enhances the efficiency of inference serving by reducing LLM inference latency and costs while...
[ "Luyao Gao", "Jianchun Liu", "Hongli Xu", "Xichong Zhang", "Yunming Liao", "Liusheng Huang" ]
[ "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2025-03-13T00:00:00
https://arxiv.org/abs/2503.10325
https://arxiv.org/pdf/2503.10325v2
2503.10325
10.48550/arXiv.2503.10325
5
0
false
null
arXiv.org
0.1945
b8f12d91ebe3f0c05c86155a77f27bdcb3ddd320bd9279bcf454f80518e68ff0
[ "arxiv", "semantic_scholar" ]
Gumiho: A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding
Speculative decoding (SPD) aims to accelerate the auto-regressive token generation process of a target Large Language Model (LLM). Some approaches employ a draft model with multiple heads to predict a sequence of future tokens, where each head handles a token in the sequence. The target LLM verifies the predicted seque...
[ "Jinze Li", "Yixing Xu", "Haiduo Huang", "Xuanwu Yin", "Dong Li", "Edith C. H. Ngai", "Emad Barsoum" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-03-13T00:00:00
https://arxiv.org/abs/2503.10135
https://arxiv.org/pdf/2503.10135v2
2503.10135
10.48550/arXiv.2503.10135
8
1
true
https://github.com/AMD-AIG-AIMA/Gumiho
International Conference on Machine Learning
0.2386
f5fe03837757d5ebda52b54eb542048f91be0a069e3c40814d7a878c92c61d94
[ "arxiv", "semantic_scholar" ]
Training Domain Draft Models for Speculative Decoding: Best Practices and Insights
Speculative decoding is an effective method for accelerating inference of large language models (LLMs) by employing a small draft model to predict the output of a target model. However, when adapting speculative decoding to domain-specific target models, the acceptance rate of the generic draft model drops significantl...
[ "Fenglu Hong", "Ravi Raju", "Jonathan Lingjie Li", "Bo Li", "Urmish Thakker", "Avinash Ravichandran", "Swayambhoo Jain", "Changran Hu" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-03-10T00:00:00
https://arxiv.org/abs/2503.07807
https://arxiv.org/pdf/2503.07807v2
2503.07807
10.48550/arXiv.2503.07807
7
0
false
null
arXiv.org
0.2258
fce8f76ecb8379a2d747f62f47434378a61e60ac232ab53372dec2a588995a10
[ "arxiv", "semantic_scholar" ]
Speculative Decoding for Multi-Sample Inference
We propose a novel speculative decoding method tailored for multi-sample reasoning scenarios, such as self-consistency and Best-of-N sampling. Our method exploits the intrinsic consensus of parallel generation paths to synthesize high-quality draft tokens without requiring auxiliary models or external databases. By dyn...
[ "Yiwei Li", "Jiayi Shi", "Shaoxiong Feng", "Peiwen Yuan", "Xinglin Wang", "Yueqi Zhang", "Ji Zhang", "Chuyi Tan", "Boyuan Pan", "Yao Hu", "Kan Li" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-03-07T00:00:00
https://arxiv.org/abs/2503.05330
https://arxiv.org/pdf/2503.05330v1
2503.05330
10.48550/arXiv.2503.05330
4
0
false
null
Conference on Empirical Methods in Natural Language Processing
0.1747
071ec866b67bb04e3d9b888ea0938dd3edafeb2fef3f1a3589583ab5cae1e898
[ "arxiv", "semantic_scholar" ]
AdaSpec: Adaptive Speculative Decoding for Fast, SLO-Aware Large Language Model Serving
Cloud-based Large Language Model (LLM) services often face challenges in achieving low inference latency and meeting Service Level Objectives (SLOs) under dynamic request patterns. Speculative decoding, which exploits lightweight models for drafting and LLMs for verification, has emerged as a compelling technique to ac...
[ "Kaiyu Huang", "Hao Wu", "Zhubo Shi", "Han Zou", "Minchen Yu", "Qingjiang Shi" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-03-07T00:00:00
https://arxiv.org/abs/2503.05096
https://arxiv.org/pdf/2503.05096v2
2503.05096
10.1145/3772052.3772239
14
3
true
https://github.com/cerebellumking/AdaSpec
ACM Symposium on Cloud Computing
0.301
a35d039489f2b418130459207e92848aa90a3be33beefaf121caf3ab9a3fe7c7
[ "arxiv", "semantic_scholar" ]
RASD: Retrieval-Augmented Speculative Decoding
Speculative decoding accelerates inference in large language models (LLMs) by generating draft tokens for target model verification. Current approaches for obtaining draft tokens rely on lightweight draft models or additional model structures to generate draft tokens and retrieve context from databases. Due to the draf...
[ "Guofeng Quan", "Wenfeng Feng", "Chuzhan Hao", "Guochao Jiang", "Yuewei Zhang", "Hao Wang" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-03-05T00:00:00
https://arxiv.org/abs/2503.03434
https://arxiv.org/pdf/2503.03434v1
2503.03434
10.48550/arXiv.2503.03434
5
0
false
null
Annual Meeting of the Association for Computational Linguistics
0.1945
f302273702346aa266f44960e65f51a83479eda973110564d5417efef338cd80
[ "arxiv", "semantic_scholar" ]
DuoDecoding: Hardware-aware Heterogeneous Speculative Decoding with Dynamic Multi-Sequence Drafting
Large language models (LLMs) exhibit exceptional performance across a wide range of tasks; however, their token-by-token autoregressive generation process significantly hinders inference speed. Speculative decoding presents a promising draft-then-verify framework that reduces generation latency while maintaining output...
[ "Kai Lv", "Honglin Guo", "Qipeng Guo", "Xipeng Qiu" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-03-02T00:00:00
https://arxiv.org/abs/2503.00784
https://arxiv.org/pdf/2503.00784v1
2503.00784
10.48550/arXiv.2503.00784
5
0
true
https://github.com/KaiLv69/DuoDecoding
arXiv.org
0.1945
3fb7815a944b7f3265c6c6d00920b13f8266319f0f85a240b72f7e82e192cea9
[ "arxiv", "semantic_scholar" ]
Tutorial Proposal: Speculative Decoding for Efficient LLM Inference
This tutorial presents a comprehensive introduction to Speculative Decoding (SD), an advanced technique for LLM inference acceleration that has garnered significant research interest in recent years. SD is introduced as an innovative decoding paradigm to mitigate the high inference latency stemming from autoregressive ...
[ "Heming Xia", "Cunxiao Du", "Yongqi Li", "Qian Liu", "Wenjie Li" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-03-01T00:00:00
https://arxiv.org/abs/2503.00491
https://arxiv.org/pdf/2503.00491v1
2503.00491
10.48550/arXiv.2503.00491
4
1
false
null
arXiv.org
0.1747
074ce10412b8df85dd200d973989c39a57492c44232da19b87e15b61a50e8fe6
[ "arxiv", "semantic_scholar" ]
Fuzzy Speculative Decoding for a Tunable Accuracy-Runtime Tradeoff
Speculative Decoding (SD) enforces strict distributional equivalence to the target model when accepting candidate tokens. While it maintains the target model's generation quality, this strict equivalence limits the speedup achievable by SD and prevents users from trading deviations from the target distribution in excha...
[ "Maximilian Holsman", "Yukun Huang", "Bhuwan Dhingra" ]
[ "cs.AI" ]
[ "Computer Science" ]
2025-02-28T00:00:00
https://arxiv.org/abs/2502.20704
https://arxiv.org/pdf/2502.20704v4
2502.20704
10.48550/arXiv.2502.20704
7
2
false
null
Annual Meeting of the Association for Computational Linguistics
0.2386
be152fba1c0d31be3a4b83bf7df5e47c44487f8098c7f93a08c3b806e2a9deca
[ "arxiv", "semantic_scholar" ]
RAPID: Long-Context Inference with Retrieval-Augmented Speculative Decoding
The emergence of long-context large language models (LLMs) offers a promising alternative to traditional retrieval-augmented generation (RAG) for processing extensive documents. However, the computational overhead of long-context inference presents significant efficiency challenges. While Speculative Decoding (SD) trad...
[ "Guanzheng Chen", "Qilong Feng", "Jinjie Ni", "Xin Li", "Michael Qizhe Shieh" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-02-27T00:00:00
https://arxiv.org/abs/2502.20330
https://arxiv.org/pdf/2502.20330v2
2502.20330
10.48550/arXiv.2502.20330
9
0
false
null
International Conference on Machine Learning
0.25
89fa649a8661950248303467c7c21cfb8832cf12741d65607f39dff31a4813f7
[ "arxiv", "semantic_scholar" ]
Towards Optimal Multi-draft Speculative Decoding
Large Language Models (LLMs) have become an indispensable part of natural language processing tasks. However, autoregressive sampling has become an efficiency bottleneck. Multi-Draft Speculative Decoding (MDSD) is a recent approach where, when generating each token, a small draft model generates multiple drafts, and th...
[ "Zhengmian Hu", "Tong Zheng", "Vignesh Viswanathan", "Ziyi Chen", "Ryan A. Rossi", "Yihan Wu", "Dinesh Manocha", "Heng Huang" ]
[ "cs.CL", "cs.DS" ]
[ "Computer Science" ]
2025-02-26T00:00:00
https://arxiv.org/abs/2502.18779
https://arxiv.org/pdf/2502.18779v1
2502.18779
10.48550/arXiv.2502.18779
15
3
false
null
International Conference on Learning Representations
0.301
952dfb175e181df3a1732eab1cae9d6baa147bc74cbb829f9986464cf48caf89
[ "arxiv", "semantic_scholar" ]
Harnessing Multiple Large Language Models: A Survey on LLM Ensemble
LLM Ensemble -- which involves the comprehensive use of multiple large language models (LLMs), each aimed at handling user queries during downstream inference, to benefit from their individual strengths -- has gained substantial attention recently. The widespread availability of LLMs, coupled with their varying strengt...
[ "Zhijun Chen", "Xiaodong Lu", "Jingzheng Li", "Pengpeng Chen", "Zhuoran Li", "Kai Sun", "Yuankai Luo", "Qianren Mao", "Ming Li", "Likang Xiao", "Dingqi Yang", "Xiao Huang", "Yikun Ban", "Hailong Sun", "Philip S. Yu" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-02-25T00:00:00
https://arxiv.org/abs/2502.18036
https://arxiv.org/pdf/2502.18036v6
2502.18036
10.48550/arXiv.2502.18036
108
4
true
https://github.com/junchenzhi/Awesome-LLM-Ensemble
arXiv.org
0.5094
24eb4df97f0ffef16ad13e5d36c738bdc02c7baeedda2e94dfc3025d191acc8d
[ "arxiv", "semantic_scholar" ]
LongSpec: Long-Context Lossless Speculative Decoding with Efficient Drafting and Verification
As Large Language Models (LLMs) can now process extremely long contexts, efficient inference over these extended inputs has become increasingly important, especially for emerging applications like LLM agents that highly depend on this capability. Speculative decoding (SD) offers a promising lossless acceleration techni...
[ "Penghui Yang", "Cunxiao Du", "Fengzhuo Zhang", "Haonan Wang", "Tianyu Pang", "Chao Du", "Bo An" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-02-24T00:00:00
https://arxiv.org/abs/2502.17421
https://arxiv.org/pdf/2502.17421v4
2502.17421
null
11
1
true
https://github.com/sail-sg/LongSpec
null
0.2698
3c6d0366d983fe0aac9a6270707bf3d2f45a5d0a85d4e204b7bc6c7cae3ec2e2
[ "arxiv", "semantic_scholar" ]
Capability Instruction Tuning: A New Paradigm for Dynamic LLM Routing
Large Language Models (LLMs) have demonstrated human-like instruction-following abilities, particularly those exceeding 100 billion parameters. The combined capability of some smaller, resource-friendly LLMs can address most of the instructions that larger LLMs excel at. In this work, we explore how to route the best-p...
[ "Yi-Kai Zhang", "De-Chuan Zhan", "Han-Jia Ye" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-02-24T00:00:00
https://arxiv.org/abs/2502.17282
https://arxiv.org/pdf/2502.17282v1
2502.17282
10.48550/arXiv.2502.17282
24
3
true
https://github.com/Now-Join-Us/CIT-LLM-Routing
AAAI Conference on Artificial Intelligence
0.3495
b5c2fbb02569328ef6a10763215a3feb340618baa30203e05d26dcc177fdfa91
[ "arxiv", "semantic_scholar" ]
CORAL: Learning Consistent Representations across Multi-step Training with Lighter Speculative Drafter
Speculative decoding is a powerful technique that accelerates Large Language Model (LLM) inference by leveraging a lightweight speculative draft model. However, existing designs suffers in performance due to misalignment between training and inference. Recent methods have tried to solve this issue by adopting a multi-s...
[ "Yepeng Weng", "Dianwen Mei", "Huishi Qiu", "Xujie Chen", "Li Liu", "Jiang Tian", "Zhongchao Shi" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-02-24T00:00:00
https://arxiv.org/abs/2502.16880
https://arxiv.org/pdf/2502.16880v3
2502.16880
10.48550/arXiv.2502.16880
9
1
false
null
Annual Meeting of the Association for Computational Linguistics
0.25
a1bd5edafcca4a9bffffa182f00eb7f94731c91888e5fec7648d6a1eff63c632
[ "arxiv", "semantic_scholar" ]
TETRIS: Optimal Draft Token Selection for Batch Speculative Decoding
We propose TETRIS, a novel method that optimizes the total throughput of batch speculative decoding in multi-request settings. Unlike existing methods that optimize for a single request or a group of requests as a whole, TETRIS actively selects the most promising draft tokens (for every request in a batch) to be accept...
[ "Zhaoxuan Wu", "Zijian Zhou", "Arun Verma", "Alok Prakash", "Daniela Rus", "Bryan Kian Hsiang Low" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-02-21T00:00:00
https://arxiv.org/abs/2502.15197
https://arxiv.org/pdf/2502.15197v2
2502.15197
10.48550/arXiv.2502.15197
7
1
false
null
Annual Meeting of the Association for Computational Linguistics
0.2258
0d44ff6bbc01fae22dd8f5ed2037e065addd39824e739557cd5d076bf87f62a9
[ "arxiv", "semantic_scholar" ]
DReSD: Dense Retrieval for Speculative Decoding
Speculative decoding (SD) accelerates Large Language Model (LLM) generation by using an efficient draft model to propose the next few tokens, which are verified by the LLM in a single forward call, reducing latency while preserving its outputs. We focus on retrieval-based SD where the draft model retrieves the next tok...
[ "Milan Gritta", "Huiyin Xue", "Gerasimos Lampouras" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-02-21T00:00:00
https://arxiv.org/abs/2502.15572
https://arxiv.org/pdf/2502.15572v2
2502.15572
10.48550/arXiv.2502.15572
5
1
false
null
Annual Meeting of the Association for Computational Linguistics
0.1945
4eab278e79a4757612dc66da38cabda2616e826342f82a6b1e67e316181cac2c
[ "arxiv", "semantic_scholar" ]
BP-GPT: Auditory Neural Decoding Using fMRI-prompted LLM
Decoding language information from brain signals represents a vital research area within brain-computer interfaces, particularly in the context of deciphering the semantic information from the fMRI signal. Although existing work uses LLM to achieve this goal, their method does not use an end-to-end approach and avoids ...
[ "Xiaoyu Chen", "Changde Du", "Che Liu", "Yizhe Wang", "Huiguang He" ]
[ "cs.HC", "cs.CL" ]
[ "Computer Science" ]
2025-02-21T00:00:00
https://arxiv.org/abs/2502.15172
https://arxiv.org/pdf/2502.15172v1
2502.15172
10.1109/ICASSP49660.2025.10890142
6
0
true
https://github.com/1994cxy/BP-GPT
IEEE International Conference on Acoustics, Speech, and Signal Processing
0.2113
5295c397ed4254f2f9bd8107e17a57f01462fd06dc5a8df7127f2c4df53d4811
[ "arxiv", "semantic_scholar" ]
C2T: A Classifier-Based Tree Construction Method in Speculative Decoding
The growing scale of Large Language Models (LLMs) has exacerbated inference latency and computational costs. Speculative decoding methods, which aim to mitigate these issues, often face inefficiencies in the construction of token trees and the verification of candidate tokens. Existing strategies, including chain mode,...
[ "Feiye Huo", "Jianchao Tan", "Kefeng Zhang", "Xunliang Cai", "Shengli Sun" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-02-19T00:00:00
https://arxiv.org/abs/2502.13652
https://arxiv.org/pdf/2502.13652v1
2502.13652
10.48550/arXiv.2502.13652
7
0
false
null
arXiv.org
0.2258
6e8d4b4315c6cfdde1b0c0d38a09e5adff35cc7c7bf71e46e9b73b1dc7e7b110
[ "arxiv", "semantic_scholar" ]
GRIFFIN: Effective Token Alignment for Faster Speculative Decoding
Speculative decoding accelerates inference in large language models (LLMs) by generating multiple draft tokens simultaneously. However, existing methods often struggle with token misalignment between the training and decoding phases, limiting their performance. To address this, we propose GRIFFIN, a novel framework tha...
[ "Shijing Hu", "Jingyang Li", "Xingyu Xie", "Zhihui Lu", "Kim-Chuan Toh", "Pan Zhou" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-02-16T00:00:00
https://arxiv.org/abs/2502.11018
https://arxiv.org/pdf/2502.11018v3
2502.11018
10.48550/arXiv.2502.11018
10
2
true
https://github.com/hsj576/GRIFFIN
arXiv.org
0.2603
1f825cf4c9f844e424a1f12d1259bef23b2aeee8014e67ff86bb29117aa1ec6c
[ "arxiv", "semantic_scholar" ]
CopySpec: Accelerating LLMs with Speculative Copy-and-Paste Without Compromising Quality
We introduce CopySpec, a simple yet effective technique to tackle the inefficiencies LLMs face when generating responses that closely resemble previous outputs or responses that can be verbatim extracted from context. CopySpec identifies repeated sequences in the model's chat history or context and speculates that the ...
[ "Razvan-Gabriel Dumitru", "Minglai Yang", "Vikas Yadav", "Mihai Surdeanu" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-02-13T00:00:00
https://arxiv.org/abs/2502.08923
https://arxiv.org/pdf/2502.08923v2
2502.08923
10.48550/arXiv.2502.08923
5
1
true
https://github.com/RazvanDu/CopySpec
Conference on Empirical Methods in Natural Language Processing
0.1945
d206bd2fdaa2950f31fe3b71cf37e549f4e6405f816fb9733e04ef4e457536f0
[ "arxiv", "semantic_scholar" ]
Speculate, then Collaborate: Fusing Knowledge of Language Models during Decoding
Large Language Models (LLMs) often excel in specific domains but fall short in others due to the limitations of their training. Thus, enabling LLMs to solve problems collaboratively by integrating their complementary knowledge promises to improve their performance across domains. To realize this potential, we introduce...
[ "Ziyao Wang", "Muneeza Azmat", "Ang Li", "Raya Horesh", "Mikhail Yurochkin" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-02-11T00:00:00
https://arxiv.org/abs/2502.08020
https://arxiv.org/pdf/2502.08020v2
2502.08020
10.48550/arXiv.2502.08020
8
2
false
null
International Conference on Machine Learning
0.2386
b8317a46b5953116904f8671806482505b9654160e1b47e86ce3348b88a6f007
[ "arxiv", "semantic_scholar" ]
Jakiro: Boosting Speculative Decoding with Decoupled Multi-Head via MoE
Speculative decoding (SD) accelerates large language model inference by using a smaller draft model to predict multiple tokens, which are then verified in parallel by the larger target model. However, the limited capacity of the draft model often necessitates tree-based sampling to improve prediction accuracy, where mu...
[ "Haiduo Huang", "Fuwei Yang", "Zhenhua Liu", "Yixing Xu", "Jinze Li", "Yang Liu", "Xuanwu Yin", "Dong Li", "Pengju Ren", "Emad Barsoum" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-02-10T00:00:00
https://arxiv.org/abs/2502.06282
https://arxiv.org/pdf/2502.06282v1
2502.06282
10.48550/arXiv.2502.06282
4
0
true
https://github.com/haiduo/Jakiro
arXiv.org
0.1747
e463e13b91bcd7bae9f0b54e5d4544de938051f27ff8e0ad8bc6675480e8151a
[ "arxiv", "semantic_scholar" ]
LANTERN++: Enhancing Relaxed Speculative Decoding with Static Tree Drafting for Visual Auto-regressive Models
Speculative decoding has been widely used to accelerate auto-regressive (AR) text generation. However, its effectiveness for visual AR models remains limited due to token selection ambiguity, where multiple tokens share similarly low probabilities and thus reduce acceptance rates. Recently, relaxed speculative decoding...
[ "Sihwan Park", "Doohyuk Jang", "Sungyub Kim", "Souvik Kundu", "Eunho Yang" ]
[ "cs.CV" ]
[ "Computer Science" ]
2025-02-10T00:00:00
https://arxiv.org/abs/2502.06352
https://arxiv.org/pdf/2502.06352v2
2502.06352
10.48550/arXiv.2502.06352
11
3
true
https://github.com/jadohu/LANTERN
arXiv.org
0.301
368cc9d45aaa33fd610dceed97284283c25b3e056fbe0b321f57087891d5b94d
[ "arxiv", "semantic_scholar" ]
Acceleration Multiple Heads Decoding for LLM via Dynamic Tree Attention
Multiple heads decoding accelerates the inference of Large Language Models (LLMs) by predicting next several tokens simultaneously. It generates and verifies multiple candidate sequences in parallel via tree attention with a fixed structure. In this paper, we replace the fixed tree attention with dynamic tree attention...
[ "Zhendong Zhang" ]
[ "cs.CV", "cs.CL" ]
[ "Computer Science" ]
2025-02-09T00:00:00
https://arxiv.org/abs/2502.05947
https://arxiv.org/pdf/2502.05947v1
2502.05947
10.48550/arXiv.2502.05947
0
0
false
null
arXiv.org
0.0516
ae671a898c0c36ef5f1d48b51eb9b634636be5417f5db2525bb4f1635257ca2a
[ "arxiv", "semantic_scholar" ]
Lossless Acceleration of Large Language Models with Hierarchical Drafting based on Temporal Locality in Speculative Decoding
Accelerating inference in Large Language Models (LLMs) is critical for real-time interactions, as they have been widely incorporated into real-world services. Speculative decoding, a fully algorithmic solution, has gained attention for improving inference speed by drafting and verifying tokens, thereby generating multi...
[ "Sukmin Cho", "Sangjin Choi", "Taeho Hwang", "Jeongyeon Seo", "Soyeong Jeong", "Huije Lee", "Hoyun Song", "Jong C. Park", "Youngjin Kwon" ]
[ "cs.CL" ]
[ "Computer Science" ]
2025-02-08T00:00:00
https://arxiv.org/abs/2502.05609
https://arxiv.org/pdf/2502.05609v1
2502.05609
10.48550/arXiv.2502.05609
5
0
false
null
North American Chapter of the Association for Computational Linguistics
0.1945
c5e41008fc1f7af6788ed83e5d935319495ac0bb0073831f53148782bbb3f444
[ "arxiv", "semantic_scholar" ]
Speeding up Speculative Decoding via Sequential Approximate Verification
Speculative Decoding (SD) is a recently proposed technique for faster inference using Large Language Models (LLMs). SD operates by using a smaller draft LLM for autoregressively generating a sequence of tokens and a larger target LLM for parallel verification to ensure statistical consistency. However, periodic paralle...
[ "Meiyu Zhong", "Noel Teku", "Ravi Tandon" ]
[ "cs.LG", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2025-02-06T00:00:00
https://arxiv.org/abs/2502.04557
https://arxiv.org/pdf/2502.04557v3
2502.04557
null
10
0
false
null
null
0.2603
34945b27884ef4ef7fd735328da26b2598beb14bc81ec907b7f7ee3bc2eb7e15
[ "arxiv", "semantic_scholar" ]
QuantSpec: Self-Speculative Decoding with Hierarchical Quantized KV Cache
Large Language Models (LLMs) are increasingly being deployed on edge devices for long-context settings, creating a growing need for fast and efficient long-context inference. In these scenarios, the Key-Value (KV) cache is the primary bottleneck in terms of both GPU memory and latency, as the full KV cache must be load...
[ "Rishabh Tiwari", "Haocheng Xi", "Aditya Tomar", "Coleman Hooper", "Sehoon Kim", "Maxwell Horton", "Mahyar Najibi", "Michael W. Mahoney", "Kurt Keutzer", "Amir Gholami" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2025-02-05T00:00:00
https://arxiv.org/abs/2502.10424
https://arxiv.org/pdf/2502.10424v1
2502.10424
10.48550/arXiv.2502.10424
18
2
false
null
International Conference on Machine Learning
0.3197
cdf165dd9931532c8527c1cab3f803fef55d8156e8835fd9dd97ead7701960fc
[ "arxiv", "semantic_scholar" ]
EasySpec: Layer-Parallel Speculative Decoding for Efficient Multi-GPU Utilization
Speculative decoding is an effective and lossless method for Large Language Model (LLM) inference acceleration. It employs a smaller model to generate a draft token sequence, which is then verified by the original base model. In multi-GPU systems, inference latency can be further reduced through tensor parallelism (TP)...
[ "Yize Wu", "Ke Gao", "Ling Li", "Yanjun Wu" ]
[ "cs.LG" ]
[ "Computer Science" ]
2025-02-04T00:00:00
https://arxiv.org/abs/2502.02493
https://arxiv.org/pdf/2502.02493v2
2502.02493
10.48550/arXiv.2502.02493
1
0
true
https://github.com/Yize-Wu/EasySpec
arXiv.org
0.0753
bce356f4f90edfc82727db7e21d64032fbfdc3d7dc82ee7c19b2e7d27ebdf109
[ "arxiv", "semantic_scholar" ]
Fast Large Language Model Collaborative Decoding via Speculation
Large Language Model (LLM) collaborative decoding techniques improve output quality by combining the outputs of multiple models at each generation step, but they incur high computational costs. In this paper, we introduce Collaborative decoding via Speculation (CoS), a novel framework that accelerates collaborative dec...
[ "Jiale Fu", "Yuchu Jiang", "Junkai Chen", "Jiaming Fan", "Xin Geng", "Xu Yang" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-02-01T00:00:00
https://arxiv.org/abs/2502.01662
https://arxiv.org/pdf/2502.01662v2
2502.01662
null
10
0
true
https://github.com/Kamichanw/CoS/
International Conference on Machine Learning
0.2603
fd660812c5e37daa5a00032cac9b1e01b6689591a9ba340f221157a890b61d41
[ "arxiv", "semantic_scholar" ]
Accelerating LLM Inference with Lossless Speculative Decoding Algorithms for Heterogeneous Vocabularies
Accelerating the inference of large language models (LLMs) is a critical challenge in generative AI. Speculative decoding (SD) methods offer substantial efficiency gains by generating multiple tokens using a single target forward pass. However, existing SD approaches require the drafter and target models to share the s...
[ "Nadav Timor", "Jonathan Mamou", "Daniel Korat", "Moshe Berchansky", "Gaurav Jain", "Oren Pereg", "Moshe Wasserblat", "David Harel" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2025-01-31T00:00:00
https://arxiv.org/abs/2502.05202
https://arxiv.org/pdf/2502.05202v3
2502.05202
10.48550/arXiv.2502.05202
19
3
false
null
International Conference on Machine Learning
0.3253
08d779b15d93e309f2d6cc1c82db177f07921093606b6c76aec4e15081bf2309
[ "arxiv", "semantic_scholar" ]
Reward-Guided Speculative Decoding for Efficient LLM Reasoning
We introduce Reward-Guided Speculative Decoding (RSD), a novel framework aimed at improving the efficiency of inference in large language models (LLMs). RSD synergistically combines a lightweight draft model with a more powerful target model, incorporating a controlled bias to prioritize high-reward outputs, in contras...
[ "Baohao Liao", "Yuhui Xu", "Hanze Dong", "Junnan Li", "Christof Monz", "Silvio Savarese", "Doyen Sahoo", "Caiming Xiong" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2025-01-31T00:00:00
https://arxiv.org/abs/2501.19324
https://arxiv.org/pdf/2501.19324v3
2501.19324
10.48550/arXiv.2501.19324
104
11
true
https://github.com/BaohaoLiao/RSD
International Conference on Machine Learning
0.5396
16026c717e4c81cffc61c1aa5397b09f339e633bf13be978b11ea7361c0c5935
[ "arxiv", "semantic_scholar" ]
Judge Decoding: Faster Speculative Sampling Requires Going Beyond Model Alignment
The performance of large language models (LLMs) is closely linked to their underlying size, leading to ever-growing networks and hence slower inference. Speculative decoding has been proposed as a technique to accelerate autoregressive generation, leveraging a fast draft model to propose candidate tokens, which are the...
[ "Gregor Bachmann", "Sotiris Anagnostidis", "Albert Pumarola", "Markos Georgopoulos", "Artsiom Sanakoyeu", "Yuming Du", "Edgar Schönfeld", "Ali Thabet", "Jonas Kohler" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2025-01-31T00:00:00
https://arxiv.org/abs/2501.19309
https://arxiv.org/pdf/2501.19309v1
2501.19309
10.48550/arXiv.2501.19309
53
11
false
null
International Conference on Learning Representations
0.5396
e9771fbdedaec9c4b0c9315a5c9380800e1deb8888e3c395a65ff585128aaae0
[ "arxiv", "semantic_scholar" ]
Probing LLM World Models: Enhancing Guesstimation with Wisdom of Crowds Decoding
Guesstimation -- the task of making approximate quantitative estimates about objects or events -- is a common real-world skill, yet remains underexplored in large language model (LLM) research. We introduce three guesstimation datasets: MARBLES, FUTURE, and ELECPRED, spanning physical estimation (e.g., how many marbles...
[ "Yun-Shiuan Chuang", "Sameer Narendran", "Nikunj Harlalka", "Alexander Cheung", "Sizhe Gao", "Siddharth Suresh", "Junjie Hu", "Timothy T. Rogers" ]
[ "cs.AI", "cs.HC" ]
[ "Computer Science" ]
2025-01-28T00:00:00
https://arxiv.org/abs/2501.17310
https://arxiv.org/pdf/2501.17310v4
2501.17310
10.48550/arXiv.2501.17310
1
0
false
null
Conference on Empirical Methods in Natural Language Processing
0.0753
1a68a08a344317440f59905cacc3416d121fa2e1f55fdef9124ee2e249ec03a4
[ "arxiv", "semantic_scholar" ]
AdaServe: Accelerating Multi-SLO LLM Serving with SLO-Customized Speculative Decoding
Modern large language model (LLM) applications exhibit diverse service-level objectives (SLOs), from low-latency requirements in interactive coding assistants to more relaxed constraints in data wrangling tasks. Existing LLM serving systems, which rely on uniform batching and scheduling strategies, often fail to meet t...
[ "Zikun Li", "Zhuofu Chen", "Remi Delacourt", "Gabriele Oliaro", "Zeyu Wang", "Qinghan Chen", "Shuhuai Lin", "April Yang", "Zhihao Zhang", "Zhuoming Chen", "Sean Lai", "Xinhao Cheng", "Xupeng Miao", "Zhihao Jia" ]
[ "cs.CL", "cs.AI", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2025-01-21T00:00:00
https://arxiv.org/abs/2501.12162
https://arxiv.org/pdf/2501.12162v2
2501.12162
null
0
0
false
null
null
0.019
7d8ad91c4e6b98ddaf46fdc740b4a11992dbcd603685e2efb9a6f5287bc35116
[ "arxiv", "semantic_scholar" ]
HADES: Hardware Accelerated Decoding for Efficient Speculation in Large Language Models
Large Language Models (LLMs) have revolutionized natural language processing by understanding and generating human-like text. However, the increasing demand for more sophisticated LLMs presents significant computational challenges due to their scale and complexity. This paper introduces Hardware Accelerated Decoding (H...
[ "Ze Yang", "Yihong Jin", "Xinhe Xu" ]
[ "cs.CL", "cs.AI", "cs.AR" ]
[ "Computer Science" ]
2024-12-27T00:00:00
https://arxiv.org/abs/2412.19925
https://arxiv.org/pdf/2412.19925v2
2412.19925
10.1109/ICCEA65460.2025.11103323
17
0
false
null
International Conference Civil Engineering and Architecture
0.3138
ec419d860fedf52cfc59d1ace891358321c8f606121f74a22921619e54d752b8
[ "arxiv", "semantic_scholar" ]
Dovetail: A CPU/GPU Heterogeneous Speculative Decoding for LLM inference
With the continuous advancement in the performance of large language models (LLMs), their demand for computational resources and memory has significantly increased, which poses major challenges for efficient inference on consumer-grade devices and legacy servers. These devices typically feature relatively weaker GPUs a...
[ "Libo Zhang", "Zhaoning Zhang", "Baizhou Xu", "Rui Li", "Zhiliang Tian", "Songzhu Mei", "Dongsheng Li" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-12-25T00:00:00
https://arxiv.org/abs/2412.18934
https://arxiv.org/pdf/2412.18934v2
2412.18934
10.48550/arXiv.2412.18934
18
1
false
null
Conference on Empirical Methods in Natural Language Processing
0.3197
e41a6cc89acea55efca9236afb2a0e711941c20257be26f3917dd1699486306b
[ "arxiv", "semantic_scholar" ]
AdaEAGLE: Optimizing Speculative Decoding via Explicit Modeling of Adaptive Draft Structures
Speculative Decoding (SD) is a popular lossless technique for accelerating the inference of Large Language Models (LLMs). We show that the decoding speed of SD frameworks with static draft structures can be significantly improved by incorporating context-aware adaptive draft structures. However, current studies on adap...
[ "Situo Zhang", "Hankun Wang", "Da Ma", "Zichen Zhu", "Lu Chen", "Kunyao Lan", "Kai Yu" ]
[ "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2024-12-25T00:00:00
https://arxiv.org/abs/2412.18910
https://arxiv.org/pdf/2412.18910v1
2412.18910
10.48550/arXiv.2412.18910
11
0
false
null
arXiv.org
0.2698
3f51517c30d349c5df2d4080244480584389de5be71577e8b31691a88a807915
[ "arxiv", "semantic_scholar" ]
Falcon: Faster and Parallel Inference of Large Language Models through Enhanced Semi-Autoregressive Drafting and Custom-Designed Decoding Tree
Striking an optimal balance between minimal drafting latency and high speculation accuracy to enhance the inference speed of Large Language Models remains a significant challenge in speculative decoding. In this paper, we introduce Falcon, an innovative semi-autoregressive speculative decoding framework fashioned to au...
[ "Xiangxiang Gao", "Weisheng Xie", "Yiwei Xiang", "Feng Ji" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-12-17T00:00:00
https://arxiv.org/abs/2412.12639
https://arxiv.org/pdf/2412.12639v3
2412.12639
10.48550/arXiv.2412.12639
24
3
false
null
AAAI Conference on Artificial Intelligence
0.3495
868ba97a8771ccf32395279cee99ea19ddc23f32f04ccec37bdeec3db00b72f5
[ "arxiv", "semantic_scholar" ]
Constrained Decoding with Speculative Lookaheads
Constrained decoding with lookahead heuristics (CDLH) is a highly effective method for aligning LLM generations to human preferences. However, the extensive lookahead roll-out operations for each generated token makes CDLH prohibitively expensive, resulting in low adoption in practice. In contrast, common decoding stra...
[ "Nishanth Nakshatri", "Shamik Roy", "Rajarshi Das", "Suthee Chaidaroon", "Leonid Boytsov", "Rashmi Gangadharaiah" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-12-09T00:00:00
https://arxiv.org/abs/2412.10418
https://arxiv.org/pdf/2412.10418v2
2412.10418
10.48550/arXiv.2412.10418
8
0
false
null
North American Chapter of the Association for Computational Linguistics
0.2386
099276887181b59b379f2937f19e48d648120e383f9ba00fd89f52aeb4b1194c
[ "arxiv", "semantic_scholar" ]
PLD+: Accelerating LLM inference by leveraging Language Model Artifacts
To reduce the latency associated with autoretrogressive LLM inference, speculative decoding has emerged as a novel decoding paradigm, where future tokens are drafted and verified in parallel. However, the practical deployment of speculative decoding is hindered by its requirements for additional computational resources...
[ "Shwetha Somasundaram", "Anirudh Phukan", "Apoorv Saxena" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-12-02T00:00:00
https://arxiv.org/abs/2412.01447
https://arxiv.org/pdf/2412.01447v1
2412.01447
10.48550/arXiv.2412.01447
13
2
false
null
North American Chapter of the Association for Computational Linguistics
0.2865
e39f7f27fca82d58bcd40d064439cde5481be1150c513714f9a42a8781beff56
[ "arxiv", "semantic_scholar" ]
Draft Model Knows When to Stop: Self-Verification Speculative Decoding for Long-Form Generation
Conventional speculative decoding (SD) methods utilize a predefined length policy for proposing drafts, which implies the premise that the target model smoothly accepts the proposed draft tokens. However, reality deviates from this assumption: the oracle draft length varies significantly, and the fixed-length policy ha...
[ "Ziyin Zhang", "Jiahao Xu", "Tian Liang", "Xingyu Chen", "Zhiwei He", "Rui Wang", "Zhaopeng Tu" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-11-27T00:00:00
https://arxiv.org/abs/2411.18462
https://arxiv.org/pdf/2411.18462v2
2411.18462
10.18653/v1/2025.emnlp-main.844
9
1
false
null
Conference on Empirical Methods in Natural Language Processing
0.25
4e5018ee695bce5c2c8fb3d32c5d822a16c798090e593c4817e3f0888b6f0751
[ "arxiv", "semantic_scholar" ]
Speculative Decoding with CTC-based Draft Model for LLM Inference Acceleration
Inference acceleration of large language models (LLMs) has been put forward in many application scenarios and speculative decoding has shown its advantage in addressing inference acceleration. Speculative decoding usually introduces a draft model to assist the base LLM where the draft model produces drafts and the base...
[ "Zhuofan Wen", "Shangtong Gui", "Yang Feng" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2024-11-25T00:00:00
https://arxiv.org/abs/2412.00061
https://arxiv.org/pdf/2412.00061v1
2412.00061
10.48550/arXiv.2412.00061
17
1
false
null
Neural Information Processing Systems
0.3138
84cb1691e5cd343e4452297706935345d82e1de5146b800f02374f6f5ba11ebf
[ "arxiv", "semantic_scholar" ]
A Survey on LLM-as-a-Judge
Accurate and consistent evaluation is crucial for decision-making across numerous fields, yet it remains a challenging task due to inherent subjectivity, variability, and scale. Large Language Models (LLMs) have achieved remarkable success across diverse domains, leading to the emergence of "LLM-as-a-Judge," where LLMs...
[ "Jiawei Gu", "Xuhui Jiang", "Zhichao Shi", "Hexiang Tan", "Xuehao Zhai", "Chengjin Xu", "Wei Li", "Yinghan Shen", "Shengjie Ma", "Honghao Liu", "Saizhuo Wang", "Kun Zhang", "Yuanzhuo Wang", "Wen Gao", "Lionel Ni", "Jian Guo" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-11-23T00:00:00
https://arxiv.org/abs/2411.15594
https://arxiv.org/pdf/2411.15594v6
2411.15594
10.48550/arXiv.2411.15594
1,447
120
false
null
arXiv.org
1
9a4115a62b6d3a37d0ae39dde6a6c75a7d15ca2c2e69f9a891dcbe0ee5e28770
[ "arxiv", "semantic_scholar" ]
Closer Look at Efficient Inference Methods: A Survey of Speculative Decoding
Efficient inference in large language models (LLMs) has become a critical focus as their scale and complexity grow. Traditional autoregressive decoding, while effective, suffers from computational inefficiencies due to its sequential token generation process. Speculative decoding addresses this bottleneck by introducin...
[ "Hyun Ryu", "Eric Kim" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-11-20T00:00:00
https://arxiv.org/abs/2411.13157
https://arxiv.org/pdf/2411.13157v2
2411.13157
10.48550/arXiv.2411.13157
7
0
false
null
arXiv.org
0.2258
03004d57fd097f502c45b6259e86a65a231638b602dfc21282405a7422117ec0
[ "arxiv", "semantic_scholar" ]
Continuous Speculative Decoding for Autoregressive Image Generation
Continuous visual autoregressive (AR) models have demonstrated promising performance in image generation. However, the heavy autoregressive inference burden imposes significant overhead. In Large Language Models (LLMs), speculative decoding has effectively accelerated discrete autoregressive inference. However, the abs...
[ "Zili Wang", "Robert Zhang", "Kun Ding", "Qi Yang", "Fei Li", "Shiming Xiang" ]
[ "cs.CV" ]
[ "Computer Science" ]
2024-11-18T00:00:00
https://arxiv.org/abs/2411.11925
https://arxiv.org/pdf/2411.11925v2
2411.11925
10.48550/arXiv.2411.11925
17
0
true
https://github.com/MarkXCloud/CSpD
arXiv.org
0.3138
03fd5745fca7988259118fd262f85a6300cdd2469df9e9929383c9de76df5fdf
[ "arxiv", "semantic_scholar" ]
SAM Decoding: Speculative Decoding via Suffix Automaton
Speculative decoding (SD) has been demonstrated as an effective technique for lossless LLM inference acceleration. Retrieval-based SD methods, one kind of model-free method, have yielded promising speedup, but they often rely on incomplete retrieval resources, inefficient retrieval methods, and are constrained to certa...
[ "Yuxuan Hu", "Ke Wang", "Xiaokang Zhang", "Fanjin Zhang", "Cuiping Li", "Hong Chen", "Jing Zhang" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-11-16T00:00:00
https://arxiv.org/abs/2411.10666
https://arxiv.org/pdf/2411.10666v3
2411.10666
10.48550/arXiv.2411.10666
24
4
true
https://github.com/hyx1999/SAM-Decoding}{repository}
Annual Meeting of the Association for Computational Linguistics
0.3495
cce3697d6a635c2423d92c52ce5bb28bae823675a2dd2919cc409f2203cee5b4
[ "arxiv", "semantic_scholar" ]
SSSD: Simply-Scalable Speculative Decoding
Speculative Decoding has emerged as a popular technique for accelerating inference in Large Language Models. However, most existing approaches yield only modest improvements in production serving systems. Methods that achieve substantial speedups typically rely on an additional trained draft model or auxiliary model co...
[ "Michele Marzollo", "Jiawei Zhuang", "Niklas Roemer", "Niklas Zwingenberger", "Lorenz K. Müller", "Lukas Cavigelli" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-11-08T00:00:00
https://arxiv.org/abs/2411.05894
https://arxiv.org/pdf/2411.05894v3
2411.05894
10.48550/arXiv.2411.05894
2
0
false
null
arXiv.org
0.1193
3181ff1c3c23fbd3c9cef004c0c3999c575448479bd8b3dad0e4e1dc4601ef63
[ "arxiv", "semantic_scholar" ]
SpecHub: Provable Acceleration to Multi-Draft Speculative Decoding
Large Language Models (LLMs) have become essential in advancing natural language processing (NLP) tasks, but their sequential token generation limits inference speed. Multi-Draft Speculative Decoding (MDSD) offers a promising solution by using a smaller draft model to generate multiple token sequences, which the target...
[ "Ryan Sun", "Tianyi Zhou", "Xun Chen", "Lichao Sun" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-11-08T00:00:00
https://arxiv.org/abs/2411.05289
https://arxiv.org/pdf/2411.05289v1
2411.05289
10.18653/v1/2024.emnlp-main.1148
10
2
true
https://github.com/MasterGodzilla/Speculative_decoding_OT}
Conference on Empirical Methods in Natural Language Processing
0.2603
a8369e2dda58650b4b7c97d55e1a9edee70952783152b14d1ddca3ab717599a6
[ "arxiv", "semantic_scholar" ]
SuffixDecoding: Extreme Speculative Decoding for Emerging AI Applications
Speculative decoding is widely adopted to reduce latency in large language model (LLM) inference by leveraging smaller draft models capable of handling diverse user tasks. However, emerging AI applications, such as LLM-based agents, present unique workload characteristics: instead of diverse independent requests, agent...
[ "Gabriele Oliaro", "Zhihao Jia", "Daniel Campos", "Aurick Qiao" ]
[ "cs.CL", "cs.AI", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2024-11-07T00:00:00
https://arxiv.org/abs/2411.04975
https://arxiv.org/pdf/2411.04975v3
2411.04975
null
17
3
true
https://github.com/snowflakedb/ArcticInference
null
0.3138
5dcabae329dec8df1448551fa0449aee3a85fbe2a7d0b8052411e16e36ee3dd5
[ "arxiv", "semantic_scholar" ]
When Speculation Spills Secrets: Side Channels via Speculative Decoding In LLMs
Deployed large language models (LLMs) often rely on speculative decoding, a technique that generates and verifies multiple candidate tokens in parallel, to improve throughput and latency. In this work, we reveal a new side-channel whereby input-dependent patterns of correct and incorrect speculations can be inferred by...
[ "Jiankun Wei", "Abdulrahman Abdulrazzag", "Tianchen Zhang", "Adel Muursepp", "Gururaj Saileshwar" ]
[ "cs.CL", "cs.AI", "cs.CR", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2024-11-01T00:00:00
https://arxiv.org/abs/2411.01076
https://arxiv.org/pdf/2411.01076v4
2411.01076
null
4
0
false
null
null
0.1747
c832770bf60a9e05bc3e1036528826a28be5210da647c61f8c42de8218fd3ae0
[ "arxiv", "semantic_scholar" ]
A Theoretical Perspective for Speculative Decoding Algorithm
Transformer-based autoregressive sampling has been the major bottleneck for slowing down large language model inferences. One effective way to accelerate inference is \emph{Speculative Decoding}, which employs a small model to sample a sequence of draft tokens and a large model to validate. Given its empirical effectiv...
[ "Ming Yin", "Minshuo Chen", "Kaixuan Huang", "Mengdi Wang" ]
[ "cs.LG", "cs.AI", "cs.CL", "stat.ML" ]
[ "Computer Science", "Mathematics" ]
2024-10-30T00:00:00
https://arxiv.org/abs/2411.00841
https://arxiv.org/pdf/2411.00841v1
2411.00841
10.48550/arXiv.2411.00841
27
3
false
null
Neural Information Processing Systems
0.3618
2efc7e9d844e87ea5ca0bfda6514a36b1445a67848e8900e0a7a573946edbfe1
[ "arxiv", "semantic_scholar" ]
Fast and High-Quality Auto-Regressive Speech Synthesis via Speculative Decoding
The auto-regressive architecture, like GPTs, is widely used in modern Text-to-Speech (TTS) systems. However, it incurs substantial inference time, particularly due to the challenges in the next-token prediction posed by lengthy sequences of speech tokens. In this work, we introduce VADUSA, one of the first approaches t...
[ "Bohan Li", "Hankun Wang", "Situo Zhang", "Yiwei Guo", "Kai Yu" ]
[ "eess.AS", "cs.AI", "cs.SD" ]
[ "Engineering", "Computer Science" ]
2024-10-29T00:00:00
https://arxiv.org/abs/2410.21951
https://arxiv.org/pdf/2410.21951v2
2410.21951
10.1109/ICASSP49660.2025.10888194
22
0
false
null
IEEE International Conference on Acoustics, Speech, and Signal Processing
0.3404
6b9c34c19b56104c44ca32d9f768aa298195c84c7434e25ba1aaafec7c180d70
[ "arxiv", "semantic_scholar" ]
Fast Best-of-N Decoding via Speculative Rejection
The safe and effective deployment of Large Language Models (LLMs) involves a critical step called alignment, which ensures that the model's responses are in accordance with human preferences. Prevalent alignment techniques, such as DPO, PPO and their variants, align LLMs by changing the pre-trained model weights during...
[ "Hanshi Sun", "Momin Haider", "Ruiqi Zhang", "Huitao Yang", "Jiahao Qiu", "Ming Yin", "Mengdi Wang", "Peter Bartlett", "Andrea Zanette" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-10-26T00:00:00
https://arxiv.org/abs/2410.20290
https://arxiv.org/pdf/2410.20290v2
2410.20290
10.48550/arXiv.2410.20290
131
15
false
null
Neural Information Processing Systems
0.6021
61a1b36da45fdb72b3ad3ee3ec0fa977dff1a17f5e592b7a532c366666f7c642
[ "arxiv", "semantic_scholar" ]
AdaEDL: Early Draft Stopping for Speculative Decoding of Large Language Models via an Entropy-based Lower Bound on Token Acceptance Probability
Speculative decoding is a powerful technique that attempts to circumvent the autoregressive constraint of modern Large Language Models (LLMs). The aim of speculative decoding techniques is to improve the average inference time of a large, target model without sacrificing its accuracy, by using a more efficient draft mo...
[ "Sudhanshu Agrawal", "Wonseok Jeon", "Mingu Lee" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-10-24T00:00:00
https://arxiv.org/abs/2410.18351
https://arxiv.org/pdf/2410.18351v1
2410.18351
10.48550/arXiv.2410.18351
17
2
false
null
null
0.3138
22c84d0293840fef668552d3801312ef6db48915c69e8e77f0c7fa9fb0d1dbd9
[ "arxiv", "semantic_scholar" ]
POD-Attention: Unlocking Full Prefill-Decode Overlap for Faster LLM Inference
Each request in LLM inference goes through two phases: compute-bound prefill and memory-bandwidth-bound decode. To improve GPU utilization, recent systems use hybrid batching that combines the prefill and decode phases of different requests into the same batch. This approach optimizes linear operations but remains inef...
[ "Aditya K Kamath", "Ramya Prabhu", "Jayashree Mohan", "Simon Peter", "Ramachandran Ramjee", "Ashish Panwar" ]
[ "cs.LG", "cs.DC" ]
[ "Computer Science" ]
2024-10-23T00:00:00
https://arxiv.org/abs/2410.18038
https://arxiv.org/pdf/2410.18038v2
2410.18038
10.1145/3676641.3715996
76
7
false
null
International Conference on Architectural Support for Programming Languages and Operating Systems
0.4716
1c093c880586c630d3eec1fa16739838534ec398c013a14f92c03fdadfeebb17
[ "arxiv", "semantic_scholar" ]
AMUSD: Asynchronous Multi-Device Speculative Decoding for LLM Acceleration
Large language models typically generate tokens autoregressively, using each token as input for the next. Recent work on Speculative Decoding has sought to accelerate this process by employing a smaller, faster draft model to more quickly generate candidate tokens. These candidates are then verified in parallel by the ...
[ "Bradley McDanel" ]
[ "cs.CL", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2024-10-22T00:00:00
https://arxiv.org/abs/2410.17375
https://arxiv.org/pdf/2410.17375v1
2410.17375
10.1109/ISCAS56072.2025.11043575
13
3
true
https://github.com/BradMcDanel/AMUSD/
International Symposium on Circuits and Systems
0.301
91445b29ab819d595956f9e9fb5a20fda991cb16c9770b1296bdfee695d6a795
[ "arxiv", "semantic_scholar" ]
Progressive Mixed-Precision Decoding for Efficient LLM Inference
In spite of the great potential of large language models (LLMs) across various tasks, their deployment on resource-constrained devices remains challenging due to their excessive computational and memory demands. Quantization has emerged as an effective solution by storing weights in reduced precision. However, utilizin...
[ "Hao Mark Chen", "Fuwen Tan", "Alexandros Kouris", "Royson Lee", "Hongxiang Fan", "Stylianos I. Venieris" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2024-10-17T00:00:00
https://arxiv.org/abs/2410.13461
https://arxiv.org/pdf/2410.13461v2
2410.13461
10.48550/arXiv.2410.13461
14
2
false
null
International Conference on Learning Representations
0.294
f02aa6149c1fda7bee6cf7225cbe8e955277d0d744ec66648ed5b88853f8fd1c
[ "arxiv", "semantic_scholar" ]
Accelerating Codec-based Speech Synthesis with Multi-Token Prediction and Speculative Decoding
The goal of this paper is to accelerate codec-based speech synthesis systems with minimum sacrifice to speech quality. We propose an enhanced inference method that allows for flexible trade-offs between speed and quality during inference without requiring additional training. Our core idea is to predict multiple tokens...
[ "Tan Dat Nguyen", "Ji-Hoon Kim", "Jeongsoo Choi", "Shukjae Choi", "Jinseok Park", "Younglo Lee", "Joon Son Chung" ]
[ "cs.SD", "cs.AI", "eess.AS" ]
[ "Computer Science", "Engineering" ]
2024-10-17T00:00:00
https://arxiv.org/abs/2410.13839
https://arxiv.org/pdf/2410.13839v1
2410.13839
10.1109/ICASSP49660.2025.10887855
14
0
false
null
IEEE International Conference on Acoustics, Speech, and Signal Processing
0.294
17965272fb0d48fc110b7c1084ccad857bacd8754c3f53f85e5a9f4c56bed98b
[ "arxiv", "semantic_scholar" ]
Cerberus: Efficient Inference with Adaptive Parallel Decoding and Sequential Knowledge Enhancement
Large language models (LLMs) often face a bottleneck in inference speed due to their reliance on auto-regressive decoding. Recently, parallel decoding has shown significant promise in enhancing inference efficiency. However, we have identified two key issues with existing parallel decoding frameworks: (1) decoding head...
[ "Yuxuan Liu", "Wenyuan Li", "Laizhong Cui", "Hailiang Yang" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-10-17T00:00:00
https://arxiv.org/abs/2410.13344
https://arxiv.org/pdf/2410.13344v1
2410.13344
10.48550/arXiv.2410.13344
1
0
false
null
arXiv.org
0.0753
ab684e7d67d4a3784a2dbf32fedc845f9ad7e01e8df98fd80be66a3c208ee77b
[ "arxiv", "semantic_scholar" ]
QSpec: Speculative Decoding with Complementary Quantization Schemes
Quantization is widely adopted to accelerate inference and reduce memory consumption in large language models (LLMs). While activation-weight joint quantization enables efficient low-precision decoding, it suffers from substantial performance degradation on multi-step reasoning tasks. We propose QSpec, a novel quantiza...
[ "Juntao Zhao", "Wenhao Lu", "Sheng Wang", "Lingpeng Kong", "Chuan Wu" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2024-10-15T00:00:00
https://arxiv.org/abs/2410.11305
https://arxiv.org/pdf/2410.11305v3
2410.11305
10.48550/arXiv.2410.11305
14
3
true
https://github.com/hku-netexplo-lab/QSpec
Conference on Empirical Methods in Natural Language Processing
0.301
a93b5f3e1e4bd6a3848a063174fa5d114516887d7df869f91610601233788675
[ "arxiv", "semantic_scholar" ]
DySpec: Faster Speculative Decoding with Dynamic Token Tree Structure
While speculative decoding has recently appeared as a promising direction for accelerating the inference of large language models (LLMs), the speedup and scalability are strongly bounded by the token acceptance rate. Prevalent methods usually organize predicted tokens as independent chains or fixed token trees, which f...
[ "Yunfan Xiong", "Ruoyu Zhang", "Yanzeng Li", "Tianhao Wu", "Lei Zou" ]
[ "cs.LG" ]
[ "Computer Science" ]
2024-10-15T00:00:00
https://arxiv.org/abs/2410.11744
https://arxiv.org/pdf/2410.11744v1
2410.11744
10.1007/s11280-025-01344-0
19
2
false
null
World wide web (Bussum)
0.3253
4b5c1b6f8c51771d1edff1a9435c1d958151e5c62a17aebbdc0e586f96dda66c
[ "arxiv", "semantic_scholar" ]
Temperature-Centric Investigation of Speculative Decoding with Knowledge Distillation
Speculative decoding stands as a pivotal technique to expedite inference in autoregressive (large) language models. This method employs a smaller draft model to speculate a block of tokens, which the target model then evaluates for acceptance. Despite a wealth of studies aimed at increasing the efficiency of speculativ...
[ "Siru Ouyang", "Shuohang Wang", "Minhao Jiang", "Ming Zhong", "Donghan Yu", "Jiawei Han", "Yelong Shen" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-10-14T00:00:00
https://arxiv.org/abs/2410.10141
https://arxiv.org/pdf/2410.10141v1
2410.10141
10.48550/arXiv.2410.10141
8
1
true
https://github.com/ozyyshr/TempSpec
Conference on Empirical Methods in Natural Language Processing
0.2386
54100a3ee502ecb7f548f7cc5df9ea300ed1e391e306f486483f7be776f8e0fa
[ "arxiv", "semantic_scholar" ]
Local and Global Decoding in Text Generation
Text generation, a key component in applications such as dialogue systems, relies on decoding algorithms that sample strings from a language model distribution. Traditional methods, such as top-$k$ and top-$π$, apply local normalisation to the model's output distribution, which can distort it. In this paper, we investi...
[ "Daniel Gareev", "Thomas Hofmann", "Ezhilmathi Krishnasamy", "Tiago Pimentel" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-10-14T00:00:00
https://arxiv.org/abs/2410.10810
https://arxiv.org/pdf/2410.10810v1
2410.10810
10.48550/arXiv.2410.10810
3
1
true
https://github.com/lowlypalace/global-decoding
Conference on Empirical Methods in Natural Language Processing
0.1505
aeefa6380c28437d1b69e53bc3c92f98d3d5f6ea0e7a3be725a2448026dfd541
[ "arxiv", "semantic_scholar" ]
Nudging: Inference-time Alignment of LLMs via Guided Decoding
Large language models (LLMs) require alignment to effectively and safely follow user instructions. This process necessitates training an aligned version for every base model, resulting in significant computational overhead. In this work, we propose NUDGING, a simple, training-free algorithm that aligns any base model a...
[ "Yu Fei", "Yasaman Razeghi", "Sameer Singh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-10-11T00:00:00
https://arxiv.org/abs/2410.09300
https://arxiv.org/pdf/2410.09300v4
2410.09300
10.18653/v1/2025.acl-long.623
23
7
false
null
Annual Meeting of the Association for Computational Linguistics
0.4515
faf30e0935df27e700a08ef92f72991789e531a3d14b7b2b7135627f520a9ae8
[ "arxiv", "semantic_scholar" ]
SWIFT: On-the-Fly Self-Speculative Decoding for LLM Inference Acceleration
Speculative decoding (SD) has emerged as a widely used paradigm to accelerate LLM inference without compromising quality. It works by first employing a compact model to draft multiple tokens efficiently and then using the target LLM to verify them in parallel. While this technique has achieved notable speedups, most ex...
[ "Heming Xia", "Yongqi Li", "Jun Zhang", "Cunxiao Du", "Wenjie Li" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-10-09T00:00:00
https://arxiv.org/abs/2410.06916
https://arxiv.org/pdf/2410.06916v2
2410.06916
10.48550/arXiv.2410.06916
63
7
true
https://github.com/hemingkx/SWIFT
International Conference on Learning Representations
0.4515
5d2f3d3194aba121f17d6303b217f950bea95cee26872108081b1f39ffc1255d
[ "arxiv", "semantic_scholar" ]
Root Defence Strategies: Ensuring Safety of LLM at the Decoding Level
Large language models (LLMs) have demonstrated immense utility across various industries. However, as LLMs advance, the risk of harmful outputs increases due to incorrect or malicious instruction prompts. While current methods effectively address jailbreak risks, they share common limitations: 1) Judging harmful respon...
[ "Xinyi Zeng", "Yuying Shang", "Jiawei Chen", "Jingyuan Zhang", "Yu Tian" ]
[ "cs.CL", "cs.CR" ]
[ "Computer Science" ]
2024-10-09T00:00:00
https://arxiv.org/abs/2410.06809
https://arxiv.org/pdf/2410.06809v3
2410.06809
10.48550/arXiv.2410.06809
13
0
false
null
arXiv.org
0.2865
5323256de97396efd50cccfab12e97b271dc6ecb8e46fc91ff97c4441decc897
[ "arxiv", "semantic_scholar" ]
ParallelSpec: Parallel Drafter for Efficient Speculative Decoding
Speculative decoding has proven to be an efficient solution to large language model (LLM) inference, where the small drafter predicts future tokens at a low cost, and the target model is leveraged to verify them in parallel. However, most existing works still draft tokens auto-regressively to maintain sequential depend...
[ "Zilin Xiao", "Hongming Zhang", "Tao Ge", "Siru Ouyang", "Vicente Ordonez", "Dong Yu" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-10-08T00:00:00
https://arxiv.org/abs/2410.05589
https://arxiv.org/pdf/2410.05589v1
2410.05589
10.48550/arXiv.2410.05589
21
2
false
null
arXiv.org
0.3356
0e153fc8151a5930e39e2f20ea16a4e9e93dcc12cd84fb140d4edcb9a0228ad2
[ "arxiv", "semantic_scholar" ]
PAD: Personalized Alignment of LLMs at Decoding-Time
Aligning with personalized preferences, which vary significantly across cultural, educational, and political differences, poses a significant challenge due to the computational costs and data demands of traditional alignment methods. In response, this paper presents Personalized Alignment at Decoding-time (PAD), a nove...
[ "Ruizhe Chen", "Xiaotian Zhang", "Meng Luo", "Wenhao Chai", "Zuozhu Liu" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-10-05T00:00:00
https://arxiv.org/abs/2410.04070
https://arxiv.org/pdf/2410.04070v7
2410.04070
null
55
6
false
null
International Conference on Learning Representations
0.437
75d6d2345b8da07774dfd290af494015f86565b95bc3828a29668718ac3d9319
[ "arxiv", "semantic_scholar" ]
LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding
Auto-Regressive (AR) models have recently gained prominence in image generation, often matching or even surpassing the performance of diffusion models. However, one major limitation of AR models is their sequential nature, which processes tokens one at a time, slowing down generation compared to models like GANs or dif...
[ "Doohyuk Jang", "Sihwan Park", "June Yong Yang", "Yeonsung Jung", "Jihun Yun", "Souvik Kundu", "Sung-Yub Kim", "Eunho Yang" ]
[ "cs.CV", "cs.AI" ]
[ "Computer Science" ]
2024-10-04T00:00:00
https://arxiv.org/abs/2410.03355
https://arxiv.org/pdf/2410.03355v3
2410.03355
10.48550/arXiv.2410.03355
48
10
true
https://github.com/jadohu/LANTERN
International Conference on Learning Representations
0.5207
aa0047df85dde3f7bcf7a7a607ec1c0e1ed0f3543f58cc0322c86cd4e4f31d1c
[ "arxiv", "semantic_scholar" ]
Mixture of Attentions For Speculative Decoding
The growth in the number of parameters of Large Language Models (LLMs) has led to a significant surge in computational requirements, making them challenging and costly to deploy. Speculative decoding (SD) leverages smaller models to efficiently propose future tokens, which are then verified by the LLM in parallel. Smal...
[ "Matthieu Zimmer", "Milan Gritta", "Gerasimos Lampouras", "Haitham Bou Ammar", "Jun Wang" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-10-04T00:00:00
https://arxiv.org/abs/2410.03804
https://arxiv.org/pdf/2410.03804v2
2410.03804
10.48550/arXiv.2410.03804
17
1
false
null
International Conference on Learning Representations
0.3138
7c7483264deeca5cbb6cdf8ceb8ebf881691fbc8d81f8866467895f5f8a54e7c
[ "arxiv", "semantic_scholar" ]
Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding
The current large auto-regressive models can generate high-quality, high-resolution images, but these models require hundreds or even thousands of steps of next-token prediction during inference, resulting in substantial time consumption. In existing studies, Jacobi decoding, an iterative parallel decoding algorithm, h...
[ "Yao Teng", "Han Shi", "Xian Liu", "Xuefei Ning", "Guohao Dai", "Yu Wang", "Zhenguo Li", "Xihui Liu" ]
[ "cs.CV" ]
[ "Computer Science" ]
2024-10-02T00:00:00
https://arxiv.org/abs/2410.01699
https://arxiv.org/pdf/2410.01699v2
2410.01699
10.48550/arXiv.2410.01699
61
11
true
https://github.com/tyshiwo1/Accelerating-T2I-AR-with-SJD/
International Conference on Learning Representations
0.5396
d7f2a64117ff70b7c5f75112b9810d531001e70ae10f8ba868bff7617a8becb0
[ "arxiv", "semantic_scholar" ]
Draft on the Fly: Adaptive Self-Speculative Decoding using Cosine Similarity
We present a simple on the fly method for faster inference of large language models. Unlike other (self-)speculative decoding techniques, our method does not require fine-tuning or black-box optimization to generate a fixed draft model, relying instead on simple rules to generate varying draft models adapted to the inp...
[ "Michael R. Metel", "Peng Lu", "Boxing Chen", "Mehdi Rezagholizadeh", "Ivan Kobyzev" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-10-01T00:00:00
https://arxiv.org/abs/2410.01028
https://arxiv.org/pdf/2410.01028v1
2410.01028
10.48550/arXiv.2410.01028
12
1
false
null
Conference on Empirical Methods in Natural Language Processing
0.2785
57666f2b5c6effd4ffad19c2ac5a8eb9baa6c8cdbb45bfefc8a4794f643d1fa0
[ "arxiv", "semantic_scholar" ]
Dynamic-Width Speculative Beam Decoding for Efficient LLM Inference
Large language models (LLMs) have shown outstanding performance across numerous real-world tasks. However, the autoregressive nature of these models makes the inference process slow and costly. Speculative decoding has emerged as a promising solution, leveraging a smaller auxiliary model to draft future tokens, which a...
[ "Zongyue Qin", "Zifan He", "Neha Prakriya", "Jason Cong", "Yizhou Sun" ]
[ "cs.AI" ]
[ "Computer Science" ]
2024-09-25T00:00:00
https://arxiv.org/abs/2409.16560
https://arxiv.org/pdf/2409.16560v2
2409.16560
10.48550/arXiv.2409.16560
7
2
false
null
arXiv.org
0.2386