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
897fefe7dea369540b92914d82dd69bfd413b5ff41923eabe887805f357e141b
[ "arxiv", "semantic_scholar" ]
Improving Multi-candidate Speculative Decoding
Speculative Decoding (SD) is a technique to accelerate the inference of Large Language Models (LLMs) by using a lower complexity draft model to propose candidate tokens verified by a larger target model. To further improve efficiency, Multi-Candidate Speculative Decoding (MCSD) improves upon this by sampling multiple c...
[ "Xiaofan Lu", "Yixiao Zeng", "Feiyang Ma", "Zixu Yu", "Marco Levorato" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-09-16T00:00:00
https://arxiv.org/abs/2409.10644
https://arxiv.org/pdf/2409.10644v3
2409.10644
10.48550/arXiv.2409.10644
6
0
false
null
null
0.2113
2b80eaed4e9a91740e4a9f421b6dbb92f9ba3dc97b89197ea2dba920a341dc19
[ "arxiv", "semantic_scholar" ]
Dynamic Depth Decoding: Faster Speculative Decoding for LLMs
The acceleration of Large Language Models (LLMs) with speculative decoding provides a significant runtime improvement without any loss of accuracy. Currently, EAGLE-2 is the state-of-the-art speculative decoding method, improving on EAGLE with a dynamic draft tree. We introduce Dynamic Depth Decoding (DDD), which optim...
[ "Oscar Brown", "Zhengjie Wang", "Andrea Do", "Nikhil Mathew", "Cheng Yu" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-08-30T00:00:00
https://arxiv.org/abs/2409.00142
https://arxiv.org/pdf/2409.00142v1
2409.00142
10.48550/arXiv.2409.00142
17
3
false
null
arXiv.org
0.3138
2da62c5a97f735d04c021fc624526aa374355314904ac96594a7e5c49cfb9afd
[ "arxiv", "semantic_scholar" ]
Boosting Lossless Speculative Decoding via Feature Sampling and Partial Alignment Distillation
Lossless speculative decoding accelerates target large language model (LLM) inference by employing a lightweight draft model for generating tree-structured candidates, which are subsequently verified in parallel by the target LLM. Currently, effective approaches leverage feature-level rather than token-level autoregres...
[ "Lujun Gui", "Bin Xiao", "Lei Su", "Weipeng Chen" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-08-28T00:00:00
https://arxiv.org/abs/2408.15562
https://arxiv.org/pdf/2408.15562v1
2408.15562
10.48550/arXiv.2408.15562
9
1
false
null
arXiv.org
0.25
c84e0ec48439b9449dc2da31ce84c2a16274440f66ea6ce54f4f0c72dec07993
[ "arxiv", "semantic_scholar" ]
Learning Harmonized Representations for Speculative Sampling
Speculative sampling is a promising approach to accelerate the decoding stage for Large Language Models (LLMs). Recent advancements that leverage target LLM's contextual information, such as hidden states and KV cache, have shown significant practical improvements. However, these approaches suffer from inconsistent con...
[ "Lefan Zhang", "Xiaodan Wang", "Yanhua Huang", "Ruiwen Xu" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2024-08-28T00:00:00
https://arxiv.org/abs/2408.15766
https://arxiv.org/pdf/2408.15766v3
2408.15766
null
67
5
true
https://github.com/HArmonizedSS/HASS
International Conference on Learning Representations
0.4581
ec9d4b656d8c4e193d8b5a0c921a6ead190da563d3d3da5211f26d9aa727fa16
[ "arxiv", "semantic_scholar" ]
Can Unconfident LLM Annotations Be Used for Confident Conclusions?
Large language models (LLMs) have shown high agreement with human raters across a variety of tasks, demonstrating potential to ease the challenges of human data collection. In computational social science (CSS), researchers are increasingly leveraging LLM annotations to complement slow and expensive human annotations. ...
[ "Kristina Gligorić", "Tijana Zrnic", "Cinoo Lee", "Emmanuel J. Candès", "Dan Jurafsky" ]
[ "cs.CL", "cs.AI", "cs.HC" ]
[ "Computer Science" ]
2024-08-27T00:00:00
https://arxiv.org/abs/2408.15204
https://arxiv.org/pdf/2408.15204v2
2408.15204
10.48550/arXiv.2408.15204
51
2
false
null
North American Chapter of the Association for Computational Linguistics
0.429
a29b63ac083c472e85fc6558020bb27a52751eb03841f0e1f70d97e041904732
[ "arxiv", "semantic_scholar" ]
The Mamba in the Llama: Distilling and Accelerating Hybrid Models
Linear RNN architectures, like Mamba, can be competitive with Transformer models in language modeling while having advantageous deployment characteristics. Given the focus on training large-scale Transformer models, we consider the challenge of converting these pretrained models for deployment. We demonstrate that it i...
[ "Junxiong Wang", "Daniele Paliotta", "Avner May", "Alexander M. Rush", "Tri Dao" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2024-08-27T00:00:00
https://arxiv.org/abs/2408.15237
https://arxiv.org/pdf/2408.15237v4
2408.15237
10.48550/arXiv.2408.15237
124
21
true
https://github.com/jxiw/MambaInLlama
Neural Information Processing Systems
0.6712
fe26000d46f6e01a99726870640909867c4242645cb720f7d388c008f34369b8
[ "arxiv", "semantic_scholar" ]
Systematic Evaluation of LLM-as-a-Judge in LLM Alignment Tasks: Explainable Metrics and Diverse Prompt Templates
LLM-as-a-Judge has been widely applied to evaluate and compare different LLM alignmnet approaches (e.g., RLHF and DPO). However, concerns regarding its reliability have emerged, due to LLM judges' biases and inconsistent decision-making. Previous research has developed evaluation frameworks to assess reliability of LLM...
[ "Hui Wei", "Shenghua He", "Tian Xia", "Fei Liu", "Andy Wong", "Jingyang Lin", "Mei Han" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-08-23T00:00:00
https://arxiv.org/abs/2408.13006
https://arxiv.org/pdf/2408.13006v2
2408.13006
10.48550/arXiv.2408.13006
81
6
true
null
arXiv.org
0.4785
5f75adcab36fdafaeffd2cadfdbad069f1caaf11b8e2d172026d1b921a2f6425
[ "arxiv", "semantic_scholar" ]
MagicDec: Breaking the Latency-Throughput Tradeoff for Long Context Generation with Speculative Decoding
Large Language Models (LLMs) have become more prevalent in long-context applications such as interactive chatbots, document analysis, and agent workflows, but it is challenging to serve long-context requests with low latency and high throughput. Speculative decoding (SD) is a widely used technique to reduce latency los...
[ "Ranajoy Sadhukhan", "Jian Chen", "Zhuoming Chen", "Vashisth Tiwari", "Ruihang Lai", "Jinyuan Shi", "Ian En-Hsu Yen", "Avner May", "Tianqi Chen", "Beidi Chen" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-08-20T00:00:00
https://arxiv.org/abs/2408.11049
https://arxiv.org/pdf/2408.11049v5
2408.11049
10.48550/arXiv.2408.11049
83
10
false
null
International Conference on Learning Representations
0.5207
064a394a74613620e565b507fd44c2eec7f1cf2e039d89799b1d5825dd9dbe87
[ "arxiv", "semantic_scholar" ]
Strategist: Self-improvement of LLM Decision Making via Bi-Level Tree Search
Traditional reinforcement learning and planning typically requires vast amounts of data and training to develop effective policies. In contrast, large language models (LLMs) exhibit strong generalization and zero-shot capabilities, but struggle with tasks that require detailed planning and decision-making in complex ac...
[ "Jonathan Light", "Min Cai", "Weiqin Chen", "Guanzhi Wang", "Xiusi Chen", "Wei Cheng", "Yisong Yue", "Ziniu Hu" ]
[ "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2024-08-20T00:00:00
https://arxiv.org/abs/2408.10635
https://arxiv.org/pdf/2408.10635v3
2408.10635
null
14
2
false
null
null
0.294
ed971c5dd045a30c17672f24c34a6829477feb4f08b770fbaf3144b2ae1256ab
[ "arxiv", "semantic_scholar" ]
KOALA: Enhancing Speculative Decoding for LLM via Multi-Layer Draft Heads with Adversarial Learning
Large Language Models (LLMs) exhibit high inference latency due to their autoregressive decoding nature. While the draft head in speculative decoding mitigates this issue, its full potential remains unexplored. In this paper, we introduce KOALA (K-layer Optimized Adversarial Learning Architecture), an orthogonal approa...
[ "Kaiqi Zhang", "Jing Zhao", "Rui Chen" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-08-15T00:00:00
https://arxiv.org/abs/2408.08146
https://arxiv.org/pdf/2408.08146v1
2408.08146
10.1109/CSCWD64889.2025.11033265
5
0
false
null
International Conference on Computer Supported Cooperative Work in Design
0.1945
1f7a4cd0af7d7b9ce1dcec72f540aeaef944b909c1210f7c005ab9a2853612f3
[ "arxiv", "semantic_scholar" ]
PEARL: Parallel Speculative Decoding with Adaptive Draft Length
Speculative decoding (SD), where an extra draft model is employed to provide multiple draft tokens first, and then the original target model verifies these tokens in parallel, has shown great power for LLM inference acceleration. However, existing SD methods suffer from the mutual waiting problem, i.e., the target mode...
[ "Tianyu Liu", "Yun Li", "Qitan Lv", "Kai Liu", "Jianchen Zhu", "Winston Hu", "Xiao Sun" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-08-13T00:00:00
https://arxiv.org/abs/2408.11850
https://arxiv.org/pdf/2408.11850v3
2408.11850
10.48550/arXiv.2408.11850
74
7
true
https://github.com/smart-lty/ParallelSpeculativeDecoding
International Conference on Learning Representations
0.4688
18c86cd7db58bc39263e38afc6c13f2693fc2e798c8050328a6fdc25ecf378df
[ "arxiv", "semantic_scholar" ]
Efficiency Unleashed: Inference Acceleration for LLM-based Recommender Systems with Speculative Decoding
The past few years have witnessed a growing interest in LLM-based recommender systems (RSs), although their industrial deployment remains in a preliminary stage. Most existing deployments leverage LLMs offline as feature enhancers, generating augmented knowledge for downstream tasks. However, in recommendation scenario...
[ "Yunjia Xi", "Hangyu Wang", "Bo Chen", "Jianghao Lin", "Menghui Zhu", "Weiwen Liu", "Ruiming Tang", "Zhewei Wei", "Weinan Zhang", "Yong Yu" ]
[ "cs.IR" ]
[ "Computer Science" ]
2024-08-11T00:00:00
https://arxiv.org/abs/2408.05676
https://arxiv.org/pdf/2408.05676v2
2408.05676
10.1145/3726302.3729961
14
0
true
https://github.com/YunjiaXi/LASER
Annual International ACM SIGIR Conference on Research and Development in Information Retrieval
0.294
7b64842ab54d8e6b89a05d0da83d12278a66987bc33f0344ae34e79d33a612be
[ "arxiv", "semantic_scholar" ]
Speculative Diffusion Decoding: Accelerating Language Generation through Diffusion
Speculative decoding has emerged as a widely adopted method to accelerate large language model inference without sacrificing the quality of the model outputs. While this technique has facilitated notable speed improvements by enabling parallel sequence verification, its efficiency remains inherently limited by the reli...
[ "Jacob K Christopher", "Brian R Bartoldson", "Tal Ben-Nun", "Michael Cardei", "Bhavya Kailkhura", "Ferdinando Fioretto" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-08-10T00:00:00
https://arxiv.org/abs/2408.05636
https://arxiv.org/pdf/2408.05636v4
2408.05636
10.48550/arXiv.2408.05636
47
2
false
null
North American Chapter of the Association for Computational Linguistics
0.4203
a90f81ad13efb3e6dc609d5b3bd739aca0301530705a00c1ff9fc9f6f82da3f0
[ "arxiv", "semantic_scholar" ]
CREST: Effectively Compacting a Datastore For Retrieval-Based Speculative Decoding
We present CREST (Compact Retrieval-Based Speculative Decoding), a redesign of REST that allows it to be effectively "compacted". REST is a drafting technique for speculative decoding based on retrieving exact n-gram matches of the most recent n tokens generated by the target LLM from a datastore. The key idea of CREST...
[ "Sophia Ho", "Jinsol Park", "Patrick Wang" ]
[ "cs.CL", "cs.AI", "cs.DB" ]
[ "Computer Science" ]
2024-08-08T00:00:00
https://arxiv.org/abs/2408.04678
https://arxiv.org/pdf/2408.04678v1
2408.04678
10.48550/arXiv.2408.04678
0
0
false
null
arXiv.org
0
ddb871574bb02d70447dd8b089e6c55acd743758688826735cdca4e2ef78326f
[ "arxiv", "semantic_scholar" ]
Clover-2: Accurate Inference for Regressive Lightweight Speculative Decoding
Large Language Models (LLMs) frequently suffer from inefficiencies, largely attributable to the discord between the requirements of auto-regressive decoding and the architecture of contemporary GPUs. Recently, regressive lightweight speculative decoding has garnered attention for its notable efficiency improvements in ...
[ "Bin Xiao", "Lujun Gui", "Lei Su", "Weipeng Chen" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-08-01T00:00:00
https://arxiv.org/abs/2408.00264
https://arxiv.org/pdf/2408.00264v1
2408.00264
10.48550/arXiv.2408.00264
5
0
true
null
arXiv.org
0.1945
428718b4c19974153cddfdd4f2beab5591a5f3bd22a531cdb196e0658f9d287a
[ "arxiv", "semantic_scholar" ]
Graph-Structured Speculative Decoding
Speculative decoding has emerged as a promising technique to accelerate the inference of Large Language Models (LLMs) by employing a small language model to draft a hypothesis sequence, which is then validated by the LLM. The effectiveness of this approach heavily relies on the balance between performance and efficienc...
[ "Zhuocheng Gong", "Jiahao Liu", "Ziyue Wang", "Pengfei Wu", "Jingang Wang", "Xunliang Cai", "Dongyan Zhao", "Rui Yan" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-07-23T00:00:00
https://arxiv.org/abs/2407.16207
https://arxiv.org/pdf/2407.16207v1
2407.16207
10.48550/arXiv.2407.16207
7
1
false
null
Annual Meeting of the Association for Computational Linguistics
0.2258
dadbb611d1d138a6f11d071145604220e650be30d6a7897cbe4583908502ed7c
[ "arxiv", "semantic_scholar" ]
PipeInfer: Accelerating LLM Inference using Asynchronous Pipelined Speculation
Inference of Large Language Models (LLMs) across computer clusters has become a focal point of research in recent times, with many acceleration techniques taking inspiration from CPU speculative execution. These techniques reduce bottlenecks associated with memory bandwidth, but also increase end-to-end latency per inf...
[ "Branden Butler", "Sixing Yu", "Arya Mazaheri", "Ali Jannesari" ]
[ "cs.CL", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2024-07-16T00:00:00
https://arxiv.org/abs/2407.11798
https://arxiv.org/pdf/2407.11798v2
2407.11798
10.1109/SC41406.2024.00046
27
2
false
null
International Conference for High Performance Computing, Networking, Storage and Analysis
0.3618
306d6d9f9fded19d1a3ca7281bec07eeac0aff5e30c21e5aa6d31c11b7dd5e77
[ "arxiv", "semantic_scholar" ]
Optimized Multi-Token Joint Decoding with Auxiliary Model for LLM Inference
Large language models (LLMs) have achieved remarkable success across diverse tasks, yet their inference processes are hindered by substantial time and energy demands due to single-token generation at each decoding step. While previous methods such as speculative decoding mitigate these inefficiencies by producing multi...
[ "Zongyue Qin", "Ziniu Hu", "Zifan He", "Neha Prakriya", "Jason Cong", "Yizhou Sun" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-07-12T00:00:00
https://arxiv.org/abs/2407.09722
https://arxiv.org/pdf/2407.09722v4
2407.09722
null
12
1
false
null
International Conference on Learning Representations
0.2785
1f783f2b3fd029ca1701073fd121b350dbc6477cc27e408a97abe79a291431ed
[ "arxiv", "semantic_scholar" ]
FBI-LLM: Scaling Up Fully Binarized LLMs from Scratch via Autoregressive Distillation
This work presents a Fully BInarized Large Language Model (FBI-LLM), demonstrating for the first time how to train a large-scale binary language model from scratch (not the partial binary or ternary LLM like BitNet b1.58) to match the performance of its full-precision counterparts (e.g., FP16 or BF16) in transformer-ba...
[ "Liqun Ma", "Mingjie Sun", "Zhiqiang Shen" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-07-09T00:00:00
https://arxiv.org/abs/2407.07093
https://arxiv.org/pdf/2407.07093v1
2407.07093
10.48550/arXiv.2407.07093
15
3
true
https://github.com/LiqunMa/FBI-LLM
arXiv.org
0.301
abe057f735ac280b7723f4fa0f7986ffa76606d2d551e291ba01f34cbdbb7427
[ "arxiv", "semantic_scholar" ]
S2D: Sorted Speculative Decoding For More Efficient Deployment of Nested Large Language Models
Deployment of autoregressive large language models (LLMs) is costly, and as these models increase in size, the associated costs will become even more considerable. Consequently, different methods have been proposed to accelerate the token generation process and reduce costs. Speculative decoding (SD) is among the most ...
[ "Parsa Kavehzadeh", "Mohammadreza Pourreza", "Mojtaba Valipour", "Tinashu Zhu", "Haoli Bai", "Ali Ghodsi", "Boxing Chen", "Mehdi Rezagholizadeh" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-07-02T00:00:00
https://arxiv.org/abs/2407.01955
https://arxiv.org/pdf/2407.01955v1
2407.01955
10.48550/arXiv.2407.01955
1
0
false
null
arXiv.org
0.0753
db3aa7aa458b66fd5617181280bdabef432e81fe16d346e8fe5d3503e195b44f
[ "arxiv", "semantic_scholar" ]
SEED: Accelerating Reasoning Tree Construction via Scheduled Speculative Decoding
Large Language Models (LLMs) demonstrate remarkable emergent abilities across various tasks, yet fall short of complex reasoning and planning tasks. The tree-search-based reasoning methods address this by surpassing the capabilities of chain-of-thought prompting, encouraging exploration of intermediate steps. However, ...
[ "Zhenglin Wang", "Jialong Wu", "Yilong Lai", "Congzhi Zhang", "Deyu Zhou" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-06-26T00:00:00
https://arxiv.org/abs/2406.18200
https://arxiv.org/pdf/2406.18200v2
2406.18200
10.48550/arXiv.2406.18200
13
1
false
null
International Conference on Computational Linguistics
0.2865
ce3dad317ad49b0f63949f786033f1df3dc130c3c458aae6a317a60d2cdf07de
[ "arxiv", "semantic_scholar" ]
OPT-Tree: Speculative Decoding with Adaptive Draft Tree Structure
Autoregressive language models demonstrate excellent performance in various scenarios. However, the inference efficiency is limited by its one-step-one-word generation mode, which has become a pressing problem recently as the models become increasingly larger. Speculative decoding employs a "draft and then verify" mech...
[ "Jikai Wang", "Yi Su", "Juntao Li", "Qingrong Xia", "Zi Ye", "Xinyu Duan", "Zhefeng Wang", "Min Zhang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-06-25T00:00:00
https://arxiv.org/abs/2406.17276
https://arxiv.org/pdf/2406.17276v4
2406.17276
10.48550/arXiv.2406.17276
56
6
true
https://github.com/Jikai0Wang/OPT-Tree
Transactions of the Association for Computational Linguistics
0.439
ef379ac790a45ee522ae1f4ab05fc54855c7cefd14099389190ee49b7bb256b3
[ "arxiv", "semantic_scholar" ]
Towards Fast Multilingual LLM Inference: Speculative Decoding and Specialized Drafters
Large language models (LLMs) have revolutionized natural language processing and broadened their applicability across diverse commercial applications. However, the deployment of these models is constrained by high inference time in multilingual settings. To mitigate this challenge, this paper explores a training recipe...
[ "Euiin Yi", "Taehyeon Kim", "Hongseok Jeung", "Du-Seong Chang", "Se-Young Yun" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-06-24T00:00:00
https://arxiv.org/abs/2406.16758
https://arxiv.org/pdf/2406.16758v2
2406.16758
10.48550/arXiv.2406.16758
13
1
false
null
Conference on Empirical Methods in Natural Language Processing
0.2865
8034281d98b6ffbc0501915288535e43ee0f5b8602f82c1b8241714cf633bec2
[ "arxiv", "semantic_scholar" ]
TurboSpec: Closed-loop Speculation Control System for Optimizing LLM Serving Goodput
Large Language Model (LLM) serving systems batch concurrent user requests to achieve efficient serving. However, in real-world deployments, such inter-request parallelism from batching is often limited by external factors such as low request rates or memory constraints. Recent works focus on intra-request parallelism f...
[ "Xiaoxuan Liu", "Jongseok Park", "Langxiang Hu", "Woosuk Kwon", "Zhuohan Li", "Chen Zhang", "Kuntai Du", "Xiangxi Mo", "Kaichao You", "Alvin Cheung", "Zhijie Deng", "Ion Stoica", "Hao Zhang" ]
[ "cs.AI", "cs.PF" ]
[ "Computer Science" ]
2024-06-20T00:00:00
https://arxiv.org/abs/2406.14066
https://arxiv.org/pdf/2406.14066v3
2406.14066
null
33
3
false
null
null
0.3829
22e0561929a8c2460264287c414e6f1894c51b738d452f52cd8381d1413867c9
[ "arxiv", "semantic_scholar" ]
Amphista: Bi-directional Multi-head Decoding for Accelerating LLM Inference
Large Language Models (LLMs) inherently use autoregressive decoding, which lacks parallelism in inference and results in significantly slow inference speed. While methods such as Medusa constructs parallelized heads, they lack adequate information interaction across different prediction positions. To overcome this limi...
[ "Zeping Li", "Xinlong Yang", "Ziheng Gao", "Ji Liu", "Guanchen Li", "Zhuang Liu", "Dong Li", "Jinzhang Peng", "Lu Tian", "Emad Barsoum" ]
[ "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2024-06-19T00:00:00
https://arxiv.org/abs/2406.13170
https://arxiv.org/pdf/2406.13170v2
2406.13170
10.18653/v1/2025.naacl-long.450
2
0
false
null
North American Chapter of the Association for Computational Linguistics
0.1193
785c8f41d6f86de01e8a77ecd1765a09b127639d7e6e945e404c5c79645ec71c
[ "arxiv", "semantic_scholar" ]
Open-LLM-Leaderboard: From Multi-choice to Open-style Questions for LLMs Evaluation, Benchmark, and Arena
Multiple-choice questions (MCQ) are frequently used to assess large language models (LLMs). Typically, an LLM is given a question and selects the answer deemed most probable after adjustments for factors like length. Unfortunately, LLMs may inherently favor certain answer choice IDs, such as A/B/C/D, due to inherent bi...
[ "Aidar Myrzakhan", "Sondos Mahmoud Bsharat", "Zhiqiang Shen" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-06-11T00:00:00
https://arxiv.org/abs/2406.07545
https://arxiv.org/pdf/2406.07545v1
2406.07545
10.48550/arXiv.2406.07545
86
2
true
https://github.com/VILA-Lab/Open-LLM-Leaderboard
arXiv.org
0.4849
5eb4eb7cd49f02bd42c0bf866f40f445d115c2fcad865fd5d686690df1124f9f
[ "arxiv", "semantic_scholar" ]
Speculative Decoding via Early-exiting for Faster LLM Inference with Thompson Sampling Control Mechanism
The recent advancements in large language models (LLMs) have been extraordinary, yet the escalating inference costs associated with them present challenges in real-world applications. To address these challenges, we propose a novel approach called Early-exiting Speculative Decoding (EESD) with lossless acceleration. Sp...
[ "Jiahao Liu", "Qifan Wang", "Jingang Wang", "Xunliang Cai" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-06-06T00:00:00
https://arxiv.org/abs/2406.03853
https://arxiv.org/pdf/2406.03853v1
2406.03853
10.48550/arXiv.2406.03853
37
3
false
null
Annual Meeting of the Association for Computational Linguistics
0.3949
ef5db1ddb7159d4aba37bcce453e447b78f8d5c9213b6a510357586cb1688d9d
[ "arxiv", "semantic_scholar" ]
SpecExec: Massively Parallel Speculative Decoding for Interactive LLM Inference on Consumer Devices
As large language models gain widespread adoption, running them efficiently becomes crucial. Recent works on LLM inference use speculative decoding to achieve extreme speedups. However, most of these works implicitly design their algorithms for high-end datacenter hardware. In this work, we ask the opposite question: h...
[ "Ruslan Svirschevski", "Avner May", "Zhuoming Chen", "Beidi Chen", "Zhihao Jia", "Max Ryabinin" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-06-04T00:00:00
https://arxiv.org/abs/2406.02532
https://arxiv.org/pdf/2406.02532v3
2406.02532
10.48550/arXiv.2406.02532
56
5
false
null
Neural Information Processing Systems
0.439
7abac9e41e114e05fbc1fbeb490ba5b3f3bf45535c9f583bbac3f5c19a5bfb5d
[ "arxiv", "semantic_scholar" ]
Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller
We propose SelfControl, an inference-time model control method utilizing gradients to control the behavior of large language models (LLMs) without explicit human annotations. Given a desired behavior expressed in a natural language suffix string concatenated to the input prompt, SelfControl computes gradients of the LL...
[ "Min Cai", "Yuchen Zhang", "Shichang Zhang", "Fan Yin", "Dan Zhang", "Difan Zou", "Yisong Yue", "Ziniu Hu" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-06-04T00:00:00
https://arxiv.org/abs/2406.02721
https://arxiv.org/pdf/2406.02721v3
2406.02721
10.48550/arXiv.2406.02721
5
0
false
null
arXiv.org
0.1945
5d510621df1af61993e486569d632a3169160ed273a382ffb3269724f1bcce10
[ "arxiv", "semantic_scholar" ]
Demystifying AI Platform Design for Distributed Inference of Next-Generation LLM models
Large language models (LLMs) have shown remarkable performance across a wide range of applications, often outperforming human experts. However, deploying these gigantic models efficiently for diverse inference use cases requires carefully designed hardware platforms with ample computing, memory, and network resources. ...
[ "Abhimanyu Bambhaniya", "Ritik Raj", "Geonhwa Jeong", "Souvik Kundu", "Sudarshan Srinivasan", "Suvinay Subramanian", "Midhilesh Elavazhagan", "Madhu Kumar", "Tushar Krishna" ]
[ "cs.AR", "cs.AI", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2024-06-03T00:00:00
https://arxiv.org/abs/2406.01698
https://arxiv.org/pdf/2406.01698v3
2406.01698
null
12
0
true
https://github.com/abhibambhaniya/GenZ-LLM-Analyzer
null
0.2785
e73f3105161668b763b4acd4585d5612ad49ff1af88863abc2bb8070eb9fb246
[ "arxiv", "semantic_scholar" ]
Snowflake: A Distributed Streaming Decoder
We design Snowflake, a quantum error correction decoder that, for the surface code under circuit-level noise, is roughly 25% more accurate than the Union-Find decoder, with a better mean runtime scaling: subquadratic as opposed to cubic in the code distance. Our decoder runs in a streaming fashion and has a distributed...
[ "Tim Chan" ]
[ "quant-ph" ]
[ "Physics", "Computer Science" ]
2024-06-03T00:00:00
https://arxiv.org/abs/2406.01701
https://arxiv.org/pdf/2406.01701v3
2406.01701
10.22331/q-2026-03-20-2033
4
0
false
null
Quantum
0.1747
935b4da18d05ec11f70059c8a9b4e7a8ebca75c36f59561389ef81144faac39e
[ "arxiv", "semantic_scholar" ]
S3D: A Simple and Cost-Effective Self-Speculative Decoding Scheme for Low-Memory GPUs
Speculative decoding (SD) has attracted a significant amount of research attention due to the substantial speedup it can achieve for LLM inference. However, despite the high speedups they offer, speculative decoding methods often achieve optimal performance on high-end devices or with a substantial GPU memory overhead....
[ "Wei Zhong", "Manasa Bharadwaj" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-05-30T00:00:00
https://arxiv.org/abs/2405.20314
https://arxiv.org/pdf/2405.20314v2
2405.20314
10.48550/arXiv.2405.20314
10
0
true
null
arXiv.org
0.2603
356d9d45e8d7e97dd240cd314d1c3c84b5ec0bd11169ac96e41b4e0b6a67260e
[ "arxiv", "semantic_scholar" ]
SpecDec++: Boosting Speculative Decoding via Adaptive Candidate Lengths
Speculative decoding reduces the inference latency of a target large language model via utilizing a smaller and faster draft model. Its performance depends on a hyperparameter K -- the candidate length, i.e., the number of candidate tokens for the target model to verify in each round. However, previous methods often us...
[ "Kaixuan Huang", "Xudong Guo", "Mengdi Wang" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-05-30T00:00:00
https://arxiv.org/abs/2405.19715
https://arxiv.org/pdf/2405.19715v3
2405.19715
10.48550/arXiv.2405.19715
56
3
true
https://github.com/Kaffaljidhmah2/SpecDec_pp
arXiv.org
0.439
5fc24943753a416bc170e82fa881004a68ec663c429acf2880295913352ba6ab
[ "arxiv", "semantic_scholar" ]
Nearest Neighbor Speculative Decoding for LLM Generation and Attribution
Large language models (LLMs) often hallucinate and lack the ability to provide attribution for their generations. Semi-parametric LMs, such as kNN-LM, approach these limitations by refining the output of an LM for a given prompt using its nearest neighbor matches in a non-parametric data store. However, these models of...
[ "Minghan Li", "Xilun Chen", "Ari Holtzman", "Beidi Chen", "Jimmy Lin", "Wen-tau Yih", "Xi Victoria Lin" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-05-29T00:00:00
https://arxiv.org/abs/2405.19325
https://arxiv.org/pdf/2405.19325v3
2405.19325
10.48550/arXiv.2405.19325
26
3
true
https://github.com/facebookresearch/NEST/tree/main
Neural Information Processing Systems
0.3578
6acae4981190d44b8a78f2cf32317774fcb65daa0c990c40314f76b15bcd050b
[ "arxiv", "semantic_scholar" ]
Faster Cascades via Speculative Decoding
Cascades and speculative decoding are two common approaches to improving language models' inference efficiency. Both approaches involve interleaving models of different sizes, but via fundamentally distinct mechanisms: cascades employ a deferral rule that invokes the larger model only for "hard" inputs, while speculati...
[ "Harikrishna Narasimhan", "Wittawat Jitkrittum", "Ankit Singh Rawat", "Seungyeon Kim", "Neha Gupta", "Aditya Krishna Menon", "Sanjiv Kumar" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-05-29T00:00:00
https://arxiv.org/abs/2405.19261
https://arxiv.org/pdf/2405.19261v2
2405.19261
10.48550/arXiv.2405.19261
35
2
false
null
International Conference on Learning Representations
0.3891
53ffe10d24c9bd1a3d69f53a2fe98e676e2c5cad7420b803975f92bab5f3f9a2
[ "arxiv", "semantic_scholar" ]
Hardware-Aware Parallel Prompt Decoding for Memory-Efficient Acceleration of LLM Inference
The auto-regressive decoding of Large Language Models (LLMs) results in significant overheads in their hardware performance. While recent research has investigated various speculative decoding techniques for multi-token generation, these efforts have primarily focused on improving processing speed such as throughput. C...
[ "Hao Mark Chen", "Wayne Luk", "Ka Fai Cedric Yiu", "Rui Li", "Konstantin Mishchenko", "Stylianos I. Venieris", "Hongxiang Fan" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2024-05-28T00:00:00
https://arxiv.org/abs/2405.18628
https://arxiv.org/pdf/2405.18628v3
2405.18628
10.48550/arXiv.2405.18628
18
2
true
https://github.com/hmarkc/parallel-prompt-decoding
Conference on Empirical Methods in Natural Language Processing
0.3197
5bbe9626cc1cf15881a604c0999978a672132bba1c5e3e2521bf65a7433d5549
[ "arxiv", "semantic_scholar" ]
Distributed Speculative Inference (DSI): Speculation Parallelism for Provably Faster Lossless Language Model Inference
This paper introduces distributed speculative inference (DSI), a novel inference algorithm that is provably faster than speculative inference (SI) [leviathan2023, chen2023, miao2024, sun2025, timor2025] and standard autoregressive inference (non-SI). Like other SI algorithms, DSI operates on frozen language models (LMs...
[ "Nadav Timor", "Jonathan Mamou", "Daniel Korat", "Moshe Berchansky", "Oren Pereg", "Moshe Wasserblat", "Tomer Galanti", "Michal Gordon", "David Harel" ]
[ "cs.DC", "cs.AI", "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-05-23T00:00:00
https://arxiv.org/abs/2405.14105
https://arxiv.org/pdf/2405.14105v5
2405.14105
null
9
0
true
null
International Conference on Learning Representations
0.25
3ae87cc8276a115349987a9b658e679ebc310bb5807fb517372f2c4b2bd93424
[ "arxiv", "semantic_scholar" ]
EMS-SD: Efficient Multi-sample Speculative Decoding for Accelerating Large Language Models
Speculative decoding emerges as a pivotal technique for enhancing the inference speed of Large Language Models (LLMs). Despite recent research aiming to improve prediction efficiency, multi-sample speculative decoding has been overlooked due to varying numbers of accepted tokens within a batch in the verification phase...
[ "Yunsheng Ni", "Chuanjian Liu", "Yehui Tang", "Kai Han", "Yunhe Wang" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-05-13T00:00:00
https://arxiv.org/abs/2405.07542
https://arxiv.org/pdf/2405.07542v2
2405.07542
10.48550/arXiv.2405.07542
2
0
true
https://github.com/niyunsheng/EMS-SD
North American Chapter of the Association for Computational Linguistics
0.1193
9c044b20b3f753e2057a7cf0bfbcf3b4942ed264594acd96b0fb0ab7a6da68b5
[ "arxiv", "semantic_scholar" ]
Dynamic Speculation Lookahead Accelerates Speculative Decoding of Large Language Models
Speculative decoding is commonly used for reducing the inference latency of large language models. Its effectiveness depends highly on the speculation lookahead (SL)-the number of tokens generated by the draft model at each iteration. In this work we show that the common practice of using the same SL for all iterations...
[ "Jonathan Mamou", "Oren Pereg", "Daniel Korat", "Moshe Berchansky", "Nadav Timor", "Moshe Wasserblat", "Roy Schwartz" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-05-07T00:00:00
https://arxiv.org/abs/2405.04304
https://arxiv.org/pdf/2405.04304v5
2405.04304
null
18
1
false
null
null
0.3197
6e6a31fb0f13afbc376c2f0d03feb04b6190ed5fe638c0456f835644fc662c42
[ "arxiv", "semantic_scholar" ]
Clover: Regressive Lightweight Speculative Decoding with Sequential Knowledge
Large language models (LLMs) suffer from low efficiency as the mismatch between the requirement of auto-regressive decoding and the design of most contemporary GPUs. Specifically, billions to trillions of parameters must be loaded to the GPU cache through its limited memory bandwidth for computation, but only a small b...
[ "Bin Xiao", "Chunan Shi", "Xiaonan Nie", "Fan Yang", "Xiangwei Deng", "Lei Su", "Weipeng Chen", "Bin Cui" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-05-01T00:00:00
https://arxiv.org/abs/2405.00263
https://arxiv.org/pdf/2405.00263v1
2405.00263
10.48550/arXiv.2405.00263
11
0
false
null
arXiv.org
0.2698
988206e09b0b62e53fdb357e771c69307ba14973e9457e00ac4eb041a9eaf696
[ "arxiv", "semantic_scholar" ]
Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting
Speculative decoding has demonstrated its effectiveness in accelerating the inference of large language models while maintaining a consistent sampling distribution. However, the conventional approach of training a separate draft model to achieve a satisfactory token acceptance rate can be costly. Drawing inspiration fr...
[ "Fangcheng Liu", "Yehui Tang", "Zhenhua Liu", "Yunsheng Ni", "Kai Han", "Yunhe Wang" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-04-29T00:00:00
https://arxiv.org/abs/2404.18911
https://arxiv.org/pdf/2404.18911v1
2404.18911
10.48550/arXiv.2404.18911
57
6
true
https://github.com/Equationliu/Kangaroo
arXiv.org
0.4409
681e54debd120618f1188d9050fd375d4cc2aba9011575ae6de3f87b025444b7
[ "arxiv", "semantic_scholar" ]
LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding
We present LayerSkip, an end-to-end solution to speed-up inference of large language models (LLMs). First, during training we apply layer dropout, with low dropout rates for earlier layers and higher dropout rates for later layers, and an early exit loss where all transformer layers share the same exit. Second, during ...
[ "Mostafa Elhoushi", "Akshat Shrivastava", "Diana Liskovich", "Basil Hosmer", "Bram Wasti", "Liangzhen Lai", "Anas Mahmoud", "Bilge Acun", "Saurabh Agarwal", "Ahmed Roman", "Ahmed A Aly", "Beidi Chen", "Carole-Jean Wu" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-04-25T00:00:00
https://arxiv.org/abs/2404.16710
https://arxiv.org/pdf/2404.16710v4
2404.16710
10.18653/v1/2024.acl-long.681
269
19
true
https://github.com/facebookresearch/LayerSkip
Annual Meeting of the Association for Computational Linguistics
0.6505
0611f3f579c30fc8e483c693d7710d3991274d1c259a9f7a0ad60cea8101ae8c
[ "arxiv", "semantic_scholar" ]
Beyond the Speculative Game: A Survey of Speculative Execution in Large Language Models
With the increasingly giant scales of (causal) large language models (LLMs), the inference efficiency comes as one of the core concerns along the improved performance. In contrast to the memory footprint, the latency bottleneck seems to be of greater importance as there can be billions of requests to a LLM (e.g., GPT-4...
[ "Chen Zhang", "Zhuorui Liu", "Dawei Song" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-04-23T00:00:00
https://arxiv.org/abs/2404.14897
https://arxiv.org/pdf/2404.14897v1
2404.14897
10.48550/arXiv.2404.14897
10
1
false
null
arXiv.org
0.2603
486f06f046f7434f0b6a3518eab5a304f6d019bfc36589f3a92c7eb39bd7e757
[ "arxiv", "semantic_scholar" ]
TriForce: Lossless Acceleration of Long Sequence Generation with Hierarchical Speculative Decoding
With large language models (LLMs) widely deployed in long content generation recently, there has emerged an increasing demand for efficient long-sequence inference support. However, key-value (KV) cache, which is stored to avoid re-computation, has emerged as a critical bottleneck by growing linearly in size with the s...
[ "Hanshi Sun", "Zhuoming Chen", "Xinyu Yang", "Yuandong Tian", "Beidi Chen" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-04-18T00:00:00
https://arxiv.org/abs/2404.11912
https://arxiv.org/pdf/2404.11912v3
2404.11912
10.48550/arXiv.2404.11912
105
5
true
https://github.com/Infini-AI-Lab/TriForce
arXiv.org
0.5063
38f2329b1ef98400df42f77b25c01a1f73c62d4d5dedc31c9f84aae3b52a2877
[ "arxiv", "semantic_scholar" ]
On Speculative Decoding for Multimodal Large Language Models
Inference with Multimodal Large Language Models (MLLMs) is slow due to their large-language-model backbone which suffers from memory bandwidth bottleneck and generates tokens auto-regressively. In this paper, we explore the application of speculative decoding to enhance the inference efficiency of MLLMs, specifically t...
[ "Mukul Gagrani", "Raghavv Goel", "Wonseok Jeon", "Junyoung Park", "Mingu Lee", "Christopher Lott" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-04-13T00:00:00
https://arxiv.org/abs/2404.08856
https://arxiv.org/pdf/2404.08856v1
2404.08856
10.48550/arXiv.2404.08856
32
4
false
null
arXiv.org
0.3796
b0e014a46b5ee6e40ba56180a410ddb821c9b41f5eeb39921eb6ecb4583342b8
[ "arxiv", "semantic_scholar" ]
DeFT: Decoding with Flash Tree-attention for Efficient Tree-structured LLM Inference
Large language models (LLMs) are increasingly employed for complex tasks that process multiple generation calls in a tree structure with shared prefixes of tokens, including few-shot prompting, multi-step reasoning, speculative decoding, etc. However, existing inference systems for tree-based applications are inefficie...
[ "Jinwei Yao", "Kaiqi Chen", "Kexun Zhang", "Jiaxuan You", "Binhang Yuan", "Zeke Wang", "Tao Lin" ]
[ "cs.CL", "cs.AI" ]
[ "Computer Science" ]
2024-03-30T00:00:00
https://arxiv.org/abs/2404.00242
https://arxiv.org/pdf/2404.00242v4
2404.00242
null
14
2
true
https://github.com/LINs-lab/DeFT
International Conference on Learning Representations
0.294
1d4fbe25611f739b2a42a7212b17961281f0d81a44127644286252d422782029
[ "arxiv", "semantic_scholar" ]
SDSAT: Accelerating LLM Inference through Speculative Decoding with Semantic Adaptive Tokens
We propose an acceleration scheme for large language models (LLMs) through Speculative Decoding with Semantic Adaptive Tokens (SDSAT). The primary objective of this design is to enhance the LLM model's ability to generate draft tokens more accurately without compromising the model's accuracy. The core strategies involv...
[ "Chengbo Liu", "Yong Zhu" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-03-27T00:00:00
https://arxiv.org/abs/2403.18647
https://arxiv.org/pdf/2403.18647v2
2403.18647
10.48550/arXiv.2403.18647
2
1
true
https://github.com/hasuoshenyun/SDSAT
arXiv.org
0.1505
27202f7f1e8f0f0b1f7f473545aaed629420bfc2110e52590e1798b175574b0b
[ "arxiv", "semantic_scholar" ]
Block Verification Accelerates Speculative Decoding
Speculative decoding is an effective method for lossless acceleration of large language models during inference. It uses a fast model to draft a block of tokens which are then verified in parallel by the target model, and provides a guarantee that the output is distributed identically to a sample from the target model....
[ "Ziteng Sun", "Uri Mendlovic", "Yaniv Leviathan", "Asaf Aharoni", "Jae Hun Ro", "Ahmad Beirami", "Ananda Theertha Suresh" ]
[ "cs.LG", "cs.CL", "cs.DS", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2024-03-15T00:00:00
https://arxiv.org/abs/2403.10444
https://arxiv.org/pdf/2403.10444v3
2403.10444
null
25
4
false
null
International Conference on Learning Representations
0.3537
3d23fefddad90bbce7ee7a548a2a9970ea0da484b3adffaf6c1b83167c1162b1
[ "arxiv", "semantic_scholar" ]
Recurrent Drafter for Fast Speculative Decoding in Large Language Models
We present Recurrent Drafter (ReDrafter), an advanced speculative decoding approach that achieves state-of-the-art speedup for large language models (LLMs) inference. The performance gains are driven by three key aspects: (1) leveraging a recurrent neural network (RNN) as the draft model conditioning on LLM's hidden st...
[ "Yunfei Cheng", "Aonan Zhang", "Xuanyu Zhang", "Chong Wang", "Yi Wang" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2024-03-14T00:00:00
https://arxiv.org/abs/2403.09919
https://arxiv.org/pdf/2403.09919v5
2403.09919
10.48550/arXiv.2403.09919
33
6
false
null
arXiv.org
0.4225
d09913b1792ce60e9925a92c4df817dbde664fd5342f45bee4e811c0af2ac801
[ "arxiv", "semantic_scholar" ]
Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs
Text generation with Large Language Models (LLMs) is known to be memory bound due to the combination of their auto-regressive nature, huge parameter counts, and limited memory bandwidths, often resulting in low token rates. Speculative decoding has been proposed as a solution for LLM inference acceleration. However, si...
[ "Raghavv Goel", "Mukul Gagrani", "Wonseok Jeon", "Junyoung Park", "Mingu Lee", "Christopher Lott" ]
[ "cs.LG", "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2024-02-29T00:00:00
https://arxiv.org/abs/2403.00858
https://arxiv.org/pdf/2403.00858v4
2403.00858
10.48550/arXiv.2403.00858
15
1
true
null
arXiv.org
0.301
c4e2410b3603dd01c8663b5c3afc115ea094d3fd9080050889f87fb49129bfdf
[ "arxiv", "semantic_scholar" ]
Minions: Accelerating Large Language Model Inference with Aggregated Speculative Execution
Large language models (LLM) have recently attracted surging interest due to their outstanding capabilities across various domains. However, enabling efficient LLM inference is challenging due to its autoregressive decoding that generates tokens only one at a time. Although research works apply pruning or quantization t...
[ "Siqi Wang", "Hailong Yang", "Xuezhu Wang", "Tongxuan Liu", "Pengbo Wang", "Xuning Liang", "Kejie Ma", "Tianyu Feng", "Xin You", "Yongjun Bao", "Yi Liu", "Zhongzhi Luan", "Depei Qian" ]
[ "cs.DC" ]
[ "Computer Science" ]
2024-02-24T00:00:00
https://arxiv.org/abs/2402.15678
https://arxiv.org/pdf/2402.15678v2
2402.15678
null
3
0
false
null
null
0.1505
41151d071dbc3846638ecebb019e55afca3581d45e84ea734cb43fd8179f5011
[ "arxiv" ]
Seeing is Believing: Mitigating Hallucination in Large Vision-Language Models via CLIP-Guided Decoding
Large Vision-Language Models (LVLMs) are susceptible to object hallucinations, an issue in which their generated text contains non-existent objects, greatly limiting their reliability and practicality. Current approaches often rely on the model's token likelihoods or other internal information, instruction tuning on ad...
[ "Ailin Deng", "Zhirui Chen", "Bryan Hooi" ]
[ "cs.CV", "cs.AI", "cs.CL", "cs.LG", "cs.MM" ]
[]
2024-02-23T00:00:00
https://arxiv.org/abs/2402.15300
https://arxiv.org/pdf/2402.15300v2
2402.15300
null
0
0
true
https://github.com/d-ailin/CLIP-Guided-Decoding
null
0
d1dacab66163f4ef1e9d64269e5ab32b32fd603165eca5fc5693a79fe426aa68
[ "arxiv", "semantic_scholar" ]
Recursive Speculative Decoding: Accelerating LLM Inference via Sampling Without Replacement
Speculative decoding is an inference-acceleration method for large language models (LLMs) where a small language model generates a draft-token sequence which is further verified by the target LLM in parallel. Recent works have advanced this method by establishing a draft-token tree, achieving superior performance over ...
[ "Wonseok Jeon", "Mukul Gagrani", "Raghavv Goel", "Junyoung Park", "Mingu Lee", "Christopher Lott" ]
[ "cs.LG", "cs.AI" ]
[ "Computer Science" ]
2024-02-21T00:00:00
https://arxiv.org/abs/2402.14160
https://arxiv.org/pdf/2402.14160v2
2402.14160
10.48550/arXiv.2402.14160
30
4
false
null
arXiv.org
0.3728
bd4d8d65a9924c0431db4e9c75daac6f0b52302d226efbf600978c094845c269
[ "arxiv", "semantic_scholar" ]
Ouroboros: Generating Longer Drafts Phrase by Phrase for Faster Speculative Decoding
Speculative decoding is a widely used method that accelerates the generation process of large language models (LLMs) with no compromise in model performance. It achieves this goal by using an existing smaller model for drafting and then employing the target LLM to verify the draft in a low-cost parallel manner. Under s...
[ "Weilin Zhao", "Yuxiang Huang", "Xu Han", "Wang Xu", "Chaojun Xiao", "Xinrong Zhang", "Yewei Fang", "Kaihuo Zhang", "Zhiyuan Liu", "Maosong Sun" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-02-21T00:00:00
https://arxiv.org/abs/2402.13720
https://arxiv.org/pdf/2402.13720v3
2402.13720
10.18653/v1/2024.emnlp-main.742
27
5
true
https://github.com/thunlp/Ouroboros
Conference on Empirical Methods in Natural Language Processing
0.3891
d47bafaa2e9fc61df6f7e08a7ef3dcf7be4f6986318aa9bab2d15f95e9df8cf9
[ "arxiv", "semantic_scholar" ]
Distillation Contrastive Decoding: Improving LLMs Reasoning with Contrastive Decoding and Distillation
We propose a straightforward approach called Distillation Contrastive Decoding (DCD) to enhance the reasoning capabilities of Large Language Models (LLMs) during inference. In contrast to previous approaches that relied on smaller amateur models or analysis of hidden state differences, DCD employs Contrastive Chain-of-...
[ "Phuc Phan", "Hieu Tran", "Long Phan" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-02-21T00:00:00
https://arxiv.org/abs/2402.14874
https://arxiv.org/pdf/2402.14874v2
2402.14874
10.48550/arXiv.2402.14874
17
0
false
null
arXiv.org
0.3138
768003fca07b400151cd526e84da9febe2d2beb31fe9044a8be540b68a5ff7c4
[ "arxiv", "semantic_scholar" ]
Sequoia: Scalable, Robust, and Hardware-aware Speculative Decoding
As the usage of large language models (LLMs) grows, performing efficient inference with these models becomes increasingly important. While speculative decoding has recently emerged as a promising direction for speeding up inference, existing methods are limited in their ability to scale to larger speculation budgets, a...
[ "Zhuoming Chen", "Avner May", "Ruslan Svirschevski", "Yuhsun Huang", "Max Ryabinin", "Zhihao Jia", "Beidi Chen" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-02-19T00:00:00
https://arxiv.org/abs/2402.12374
https://arxiv.org/pdf/2402.12374v3
2402.12374
10.48550/arXiv.2402.12374
95
11
false
null
arXiv.org
0.5396
67ef487bb6d1c263e7e18ebf9b01a0015d7139fb1d21d27bee057a62d5181125
[ "arxiv", "semantic_scholar" ]
Generation Meets Verification: Accelerating Large Language Model Inference with Smart Parallel Auto-Correct Decoding
This research aims to accelerate the inference speed of large language models (LLMs) with billions of parameters. We propose \textbf{S}mart \textbf{P}arallel \textbf{A}uto-\textbf{C}orrect d\textbf{E}coding (SPACE), an innovative approach designed for achieving lossless acceleration of LLMs. By integrating semi-autoreg...
[ "Hanling Yi", "Feng Lin", "Hongbin Li", "Peiyang Ning", "Xiaotian Yu", "Rong Xiao" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-02-19T00:00:00
https://arxiv.org/abs/2402.11809
https://arxiv.org/pdf/2402.11809v3
2402.11809
10.48550/arXiv.2402.11809
26
1
false
null
Annual Meeting of the Association for Computational Linguistics
0.3578
f7bfd7fd916df004684bfad76c94639ef631a600067f0f3b0a59341d5db7264c
[ "arxiv", "semantic_scholar" ]
Speculative Streaming: Fast LLM Inference without Auxiliary Models
Speculative decoding is a prominent technique to speed up the inference of a large target language model based on predictions of an auxiliary draft model. While effective, in application-specific settings, it often involves fine-tuning both draft and target models to achieve high acceptance rates. As the number of down...
[ "Nikhil Bhendawade", "Irina Belousova", "Qichen Fu", "Henry Mason", "Mohammad Rastegari", "Mahyar Najibi" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2024-02-16T00:00:00
https://arxiv.org/abs/2402.11131
https://arxiv.org/pdf/2402.11131v1
2402.11131
10.48550/arXiv.2402.11131
44
2
false
null
null
0.4133
dff0644adcdb708365e373b19ad2f33f82ce9be491fe7db1c504f6221ba5e466
[ "arxiv", "semantic_scholar" ]
Any-Precision LLM: Low-Cost Deployment of Multiple, Different-Sized LLMs
Recently, considerable efforts have been directed towards compressing Large Language Models (LLMs), which showcase groundbreaking capabilities across diverse applications but entail significant deployment costs due to their large sizes. Meanwhile, much less attention has been given to mitigating the costs associated wi...
[ "Yeonhong Park", "Jake Hyun", "SangLyul Cho", "Bonggeun Sim", "Jae W. Lee" ]
[ "cs.LG" ]
[ "Computer Science" ]
2024-02-16T00:00:00
https://arxiv.org/abs/2402.10517
https://arxiv.org/pdf/2402.10517v4
2402.10517
10.48550/arXiv.2402.10517
54
11
true
https://github.com/SNU-ARC/any-precision-llm
International Conference on Machine Learning
0.5396
546d766e1e4387ef530e3151eb52bd42a1c79f069847f3f24de216c474c5d889
[ "arxiv", "semantic_scholar" ]
GliDe with a CaPE: A Low-Hassle Method to Accelerate Speculative Decoding
Speculative decoding is a relatively new decoding framework that leverages small and efficient draft models to reduce the latency of LLMs. In this study, we introduce GliDe and CaPE, two low-hassle modifications to vanilla speculative decoding to further improve the decoding speed of a frozen LLM. Specifically, GliDe i...
[ "Cunxiao Du", "Jing Jiang", "Xu Yuanchen", "Jiawei Wu", "Sicheng Yu", "Yongqi Li", "Shenggui Li", "Kai Xu", "Liqiang Nie", "Zhaopeng Tu", "Yang You" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-02-03T00:00:00
https://arxiv.org/abs/2402.02082
https://arxiv.org/pdf/2402.02082v1
2402.02082
10.48550/arXiv.2402.02082
78
4
false
null
International Conference on Machine Learning
0.4744
5afeb01a5e46fba807f459ae1e691d411dd2127eb024b6e49a25152ef04caefe
[ "arxiv", "semantic_scholar" ]
Break the Sequential Dependency of LLM Inference Using Lookahead Decoding
Autoregressive decoding of large language models (LLMs) is memory bandwidth bounded, resulting in high latency and significant wastes of the parallel processing power of modern accelerators. Existing methods for accelerating LLM decoding often require a draft model (e.g., speculative decoding), which is nontrivial to o...
[ "Yichao Fu", "Peter Bailis", "Ion Stoica", "Hao Zhang" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2024-02-03T00:00:00
https://arxiv.org/abs/2402.02057
https://arxiv.org/pdf/2402.02057v1
2402.02057
10.48550/arXiv.2402.02057
315
35
true
https://github.com/hao-ai-lab/LookaheadDecoding
International Conference on Machine Learning
0.7782
4d4e11f693ae1f37f62f30b2b926422543110fa1dd5f341bd0bff2b0728d2534
[ "arxiv", "semantic_scholar" ]
Decoding Speculative Decoding
Speculative Decoding is a widely used technique to speed up inference for Large Language Models (LLMs) without sacrificing quality. When performing inference, speculative decoding uses a smaller draft model to generate speculative tokens and then uses the target LLM to verify those draft tokens. The speedup provided by...
[ "Minghao Yan", "Saurabh Agarwal", "Shivaram Venkataraman" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2024-02-02T00:00:00
https://arxiv.org/abs/2402.01528
https://arxiv.org/pdf/2402.01528v4
2402.01528
10.48550/arXiv.2402.01528
39
2
false
null
North American Chapter of the Association for Computational Linguistics
0.4005
dfd9d429a28a236017b071e728afd8feee8abc64132fd4bd0b116dc6c0b17581
[ "arxiv", "semantic_scholar" ]
Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads
Large Language Models (LLMs) employ auto-regressive decoding that requires sequential computation, with each step reliant on the previous one's output. This creates a bottleneck as each step necessitates moving the full model parameters from High-Bandwidth Memory (HBM) to the accelerator's cache. While methods such as ...
[ "Tianle Cai", "Yuhong Li", "Zhengyang Geng", "Hongwu Peng", "Jason D. Lee", "Deming Chen", "Tri Dao" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2024-01-19T00:00:00
https://arxiv.org/abs/2401.10774
https://arxiv.org/pdf/2401.10774v3
2401.10774
10.48550/arXiv.2401.10774
753
147
true
https://github.com/FasterDecoding/Medusa
International Conference on Machine Learning
1
3739a5bf6bc169148adb1c108e3f1338f2c82f67c08799ecc6cef741383ee896
[ "arxiv", "semantic_scholar" ]
Unlocking Efficiency in Large Language Model Inference: A Comprehensive Survey of Speculative Decoding
To mitigate the high inference latency stemming from autoregressive decoding in Large Language Models (LLMs), Speculative Decoding has emerged as a novel decoding paradigm for LLM inference. In each decoding step, this method first drafts several future tokens efficiently and then verifies them in parallel. Unlike auto...
[ "Heming Xia", "Zhe Yang", "Qingxiu Dong", "Peiyi Wang", "Yongqi Li", "Tao Ge", "Tianyu Liu", "Wenjie Li", "Zhifang Sui" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-01-15T00:00:00
https://arxiv.org/abs/2401.07851
https://arxiv.org/pdf/2401.07851v3
2401.07851
10.48550/arXiv.2401.07851
283
50
false
null
Annual Meeting of the Association for Computational Linguistics
0.8538
3b5e717fc383731e127e738c5842430f05d1b04a76cc8beb48d2e9c575923e4d
[ "arxiv", "semantic_scholar" ]
Small LLMs Are Weak Tool Learners: A Multi-LLM Agent
Large Language Model (LLM) agents significantly extend the capabilities of standalone LLMs, empowering them to interact with external tools (e.g., APIs, functions) and complete various tasks in a self-directed fashion. The challenge of tool use demands that LLMs not only understand user queries and generate answers acc...
[ "Weizhou Shen", "Chenliang Li", "Hongzhan Chen", "Ming Yan", "Xiaojun Quan", "Hehong Chen", "Ji Zhang", "Fei Huang" ]
[ "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2024-01-14T00:00:00
https://arxiv.org/abs/2401.07324
https://arxiv.org/pdf/2401.07324v3
2401.07324
10.48550/arXiv.2401.07324
124
6
true
https://github.com/X-PLUG/Multi-LLM-Agent
Conference on Empirical Methods in Natural Language Processing
0.5242
9289d4a06c883c5d803b3c4a71e69c87d4c36977648eb360689f1b92910b96af
[ "arxiv", "semantic_scholar" ]
Multi-Candidate Speculative Decoding
Large language models have shown impressive capabilities across a variety of NLP tasks, yet their generating text autoregressively is time-consuming. One way to speed them up is speculative decoding, which generates candidate segments (a sequence of tokens) from a fast draft model that is then verified in parallel by t...
[ "Sen Yang", "Shujian Huang", "Xinyu Dai", "Jiajun Chen" ]
[ "cs.CL" ]
[ "Computer Science" ]
2024-01-12T00:00:00
https://arxiv.org/abs/2401.06706
https://arxiv.org/pdf/2401.06706v1
2401.06706
10.48550/arXiv.2401.06706
37
3
false
null
Natural Language Processing and Chinese Computing
0.3949
172cdfcaed73f43ec06d4e8d3885d211bc109844b8346580d8b9003583c9170a
[ "arxiv", "semantic_scholar" ]
Cascade Speculative Drafting for Even Faster LLM Inference
Introduced to enhance the efficiency of large language model (LLM) inference, speculative decoding operates by having a smaller model generate a draft. A larger target model then reviews this draft to align with its output, and any acceptance by the target model results in a reduction of the number of the target model ...
[ "Ziyi Chen", "Xiaocong Yang", "Jiacheng Lin", "Chenkai Sun", "Kevin Chen-Chuan Chang", "Jie Huang" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2023-12-18T00:00:00
https://arxiv.org/abs/2312.11462
https://arxiv.org/pdf/2312.11462v5
2312.11462
10.48550/arXiv.2312.11462
96
8
false
null
Neural Information Processing Systems
0.4967
418682902e811217ef1402456ecf3ad7ba498f0194e64f4f76d6592a981b8d61
[ "arxiv", "semantic_scholar" ]
Speculative Contrastive Decoding
Large language models~(LLMs) exhibit exceptional performance in language tasks, yet their auto-regressive inference is limited due to high computational requirements and is sub-optimal due to the exposure bias. Inspired by speculative decoding and contrastive decoding, we introduce Speculative Contrastive Decoding~(SCD...
[ "Hongyi Yuan", "Keming Lu", "Fei Huang", "Zheng Yuan", "Chang Zhou" ]
[ "cs.CL" ]
[ "Computer Science" ]
2023-11-15T00:00:00
https://arxiv.org/abs/2311.08981
https://arxiv.org/pdf/2311.08981v2
2311.08981
10.48550/arXiv.2311.08981
8
1
false
null
Annual Meeting of the Association for Computational Linguistics
0.2386
2d0641d2877fb244ffa00863f126db4b56cb63a10035476a394e51ee41f0d0e3
[ "arxiv", "semantic_scholar" ]
REST: Retrieval-Based Speculative Decoding
We introduce Retrieval-Based Speculative Decoding (REST), a novel algorithm designed to speed up language model generation. The key insight driving the development of REST is the observation that the process of text generation often includes certain common phases and patterns. Unlike previous methods that rely on a dra...
[ "Zhenyu He", "Zexuan Zhong", "Tianle Cai", "Jason D. Lee", "Di He" ]
[ "cs.CL", "cs.AI", "cs.IR", "cs.LG" ]
[ "Computer Science" ]
2023-11-14T00:00:00
https://arxiv.org/abs/2311.08252
https://arxiv.org/pdf/2311.08252v2
2311.08252
10.48550/arXiv.2311.08252
162
16
true
https://github.com/FasterDecoding/REST
North American Chapter of the Association for Computational Linguistics
0.6152
005924600d1d561a41e58f127eb55b2a4c77af234b6386ae573eba5a025ee598
[ "arxiv", "semantic_scholar" ]
Leveraging Speculative Sampling and KV-Cache Optimizations Together for Generative AI using OpenVINO
Inference optimizations are critical for improving user experience and reducing infrastructure costs and power consumption. In this article, we illustrate a form of dynamic execution known as speculative sampling to reduce the overall latency of text generation and compare it with standard autoregressive sampling. This...
[ "Haim Barad", "Ekaterina Aidova", "Yury Gorbachev" ]
[ "cs.LG", "cs.AI", "cs.PF" ]
[ "Computer Science" ]
2023-11-08T00:00:00
https://arxiv.org/abs/2311.04951
https://arxiv.org/pdf/2311.04951v2
2311.04951
10.48550/arXiv.2311.04951
1
0
true
https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/speculative-sampling
arXiv.org
0.0753
59426672d48f139b9674b376fb424828577342832d3a069035d365dd21f86da9
[ "arxiv", "semantic_scholar" ]
The Synergy of Speculative Decoding and Batching in Serving Large Language Models
Large Language Models (LLMs) like GPT are state-of-the-art text generation models that provide significant assistance in daily routines. However, LLM execution is inherently sequential, since they only produce one token at a time, thus incurring low hardware utilization on modern GPUs. Batching and speculative decoding...
[ "Qidong Su", "Christina Giannoula", "Gennady Pekhimenko" ]
[ "cs.LG", "cs.DC" ]
[ "Computer Science" ]
2023-10-28T00:00:00
https://arxiv.org/abs/2310.18813
https://arxiv.org/pdf/2310.18813v1
2310.18813
10.48550/arXiv.2310.18813
21
2
false
null
arXiv.org
0.3356
c7a03c3f5d42878235e16256009225ab25138f3e54853397f3fae81b3a993f5b
[ "arxiv", "semantic_scholar" ]
SpecTr: Fast Speculative Decoding via Optimal Transport
Autoregressive sampling from large language models has led to state-of-the-art results in several natural language tasks. However, autoregressive sampling generates tokens one at a time making it slow, and even prohibitive in certain tasks. One way to speed up sampling is $\textit{speculative decoding}$: use a small mo...
[ "Ziteng Sun", "Ananda Theertha Suresh", "Jae Hun Ro", "Ahmad Beirami", "Himanshu Jain", "Felix Yu" ]
[ "cs.LG", "cs.CL", "cs.DS", "cs.IT" ]
[ "Computer Science", "Mathematics" ]
2023-10-23T00:00:00
https://arxiv.org/abs/2310.15141
https://arxiv.org/pdf/2310.15141v2
2310.15141
10.48550/arXiv.2310.15141
151
18
false
null
Neural Information Processing Systems
0.6394
4a5e48eaf98f79409c10cbbc00f3eba9bf1e1fe42857d9571ff0f2454c36c96b
[ "arxiv", "semantic_scholar" ]
SPEED: Speculative Pipelined Execution for Efficient Decoding
Generative Large Language Models (LLMs) based on the Transformer architecture have recently emerged as a dominant foundation model for a wide range of Natural Language Processing tasks. Nevertheless, their application in real-time scenarios has been highly restricted due to the significant inference latency associated ...
[ "Coleman Hooper", "Sehoon Kim", "Hiva Mohammadzadeh", "Hasan Genc", "Kurt Keutzer", "Amir Gholami", "Sophia Shao" ]
[ "cs.CL" ]
[ "Computer Science" ]
2023-10-18T00:00:00
https://arxiv.org/abs/2310.12072
https://arxiv.org/pdf/2310.12072v2
2310.12072
10.48550/arXiv.2310.12072
52
6
false
null
arXiv.org
0.4311
0a0048d6d75dc0f0af4f2cac013af9142883a3c23daf8f01dd01ea43ca72e28f
[ "arxiv", "semantic_scholar" ]
DistillSpec: Improving Speculative Decoding via Knowledge Distillation
Speculative decoding (SD) accelerates large language model inference by employing a faster draft model for generating multiple tokens, which are then verified in parallel by the larger target model, resulting in the text generated according to the target model distribution. However, identifying a compact draft model th...
[ "Yongchao Zhou", "Kaifeng Lyu", "Ankit Singh Rawat", "Aditya Krishna Menon", "Afshin Rostamizadeh", "Sanjiv Kumar", "Jean-François Kagy", "Rishabh Agarwal" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
[ "Computer Science" ]
2023-10-12T00:00:00
https://arxiv.org/abs/2310.08461
https://arxiv.org/pdf/2310.08461v2
2310.08461
10.48550/arXiv.2310.08461
159
12
false
null
International Conference on Learning Representations
0.557
8f8b1b6229e18341df51a3b2a856ff108a87eba39bf21ebd3324ce8f0c01549a
[ "arxiv", "semantic_scholar" ]
Online Speculative Decoding
Speculative decoding is a pivotal technique to accelerate the inference of large language models (LLMs) by employing a smaller draft model to predict the target model's outputs. However, its efficacy can be limited due to the low predictive accuracy of the draft model, particularly when faced with diverse text inputs a...
[ "Xiaoxuan Liu", "Lanxiang Hu", "Peter Bailis", "Alvin Cheung", "Zhijie Deng", "Ion Stoica", "Hao Zhang" ]
[ "cs.AI", "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2023-10-11T00:00:00
https://arxiv.org/abs/2310.07177
https://arxiv.org/pdf/2310.07177v4
2310.07177
10.48550/arXiv.2310.07177
112
9
true
https://github.com/LiuXiaoxuanPKU/OSD
International Conference on Machine Learning
0.5133
80046499ae7fe34b7d3ceeeb93e6c84cb79fd7a37e04b7e1c0346ec84371dfe2
[ "arxiv" ]
Draft & Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding
We present a novel inference scheme, self-speculative decoding, for accelerating Large Language Models (LLMs) without the need for an auxiliary model. This approach is characterized by a two-stage process: drafting and verification. The drafting stage generates draft tokens at a slightly lower quality but more quickly,...
[ "Jun Zhang", "Jue Wang", "Huan Li", "Lidan Shou", "Ke Chen", "Gang Chen", "Sharad Mehrotra" ]
[ "cs.CL" ]
[]
2023-09-15T00:00:00
https://arxiv.org/abs/2309.08168
https://arxiv.org/pdf/2309.08168v2
2309.08168
null
0
0
false
null
null
0
49216a6b94fcca5887cbfa932cb17e60ca294b4a213b49b3e9b1a6269de650e5
[ "arxiv", "semantic_scholar" ]
SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills
Large Language Model (LLM) inference consists of two distinct phases - prefill phase which processes the input prompt and decode phase which generates output tokens autoregressively. While the prefill phase effectively saturates GPU compute at small batch sizes, the decode phase results in low compute utilization as it...
[ "Amey Agrawal", "Ashish Panwar", "Jayashree Mohan", "Nipun Kwatra", "Bhargav S. Gulavani", "Ramachandran Ramjee" ]
[ "cs.LG", "cs.DC" ]
[ "Computer Science" ]
2023-08-31T00:00:00
https://arxiv.org/abs/2308.16369
https://arxiv.org/pdf/2308.16369v1
2308.16369
10.48550/arXiv.2308.16369
231
19
false
null
arXiv.org
0.6505
8114250c7ba56938a075f69d04d50662d43f8a0a16cc9c40f32b7e0785db9223
[ "arxiv", "semantic_scholar" ]
Accelerating LLM Inference with Staged Speculative Decoding
Recent advances with large language models (LLM) illustrate their diverse capabilities. We propose a novel algorithm, staged speculative decoding, to accelerate LLM inference in small-batch, on-device scenarios. We address the low arithmetic intensity of small-batch inference by improving upon previous work in speculat...
[ "Benjamin Spector", "Chris Re" ]
[ "cs.AI", "cs.CL" ]
[ "Computer Science" ]
2023-08-08T00:00:00
https://arxiv.org/abs/2308.04623
https://arxiv.org/pdf/2308.04623v1
2308.04623
10.48550/arXiv.2308.04623
178
9
false
null
arXiv.org
0.5632
6e04394cce44d057e3bcbd2217a25675868387cf0c1bc36a6cd6b8435d406631
[ "arxiv", "semantic_scholar" ]
RETA-LLM: A Retrieval-Augmented Large Language Model Toolkit
Although Large Language Models (LLMs) have demonstrated extraordinary capabilities in many domains, they still have a tendency to hallucinate and generate fictitious responses to user requests. This problem can be alleviated by augmenting LLMs with information retrieval (IR) systems (also known as retrieval-augmented L...
[ "Jiongnan Liu", "Jiajie Jin", "Zihan Wang", "Jiehan Cheng", "Zhicheng Dou", "Ji-Rong Wen" ]
[ "cs.IR" ]
[ "Computer Science" ]
2023-06-08T00:00:00
https://arxiv.org/abs/2306.05212
https://arxiv.org/pdf/2306.05212v1
2306.05212
10.48550/arXiv.2306.05212
61
2
true
https://github.com/RUC-GSAI/YuLan-IR/tree/main/RETA-LLM
arXiv.org
0.4481
ae9d74b24d12158c81e74194281d8cce682e379fc765f2defb62ff138cab96a0
[ "arxiv", "semantic_scholar" ]
AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration
Large language models (LLMs) have transformed numerous AI applications. On-device LLM is becoming increasingly important: running LLMs locally on edge devices can reduce the cloud computing cost and protect users' privacy. However, the astronomical model size and the limited hardware resource pose significant deploymen...
[ "Ji Lin", "Jiaming Tang", "Haotian Tang", "Shang Yang", "Wei-Ming Chen", "Wei-Chen Wang", "Guangxuan Xiao", "Xingyu Dang", "Chuang Gan", "Song Han" ]
[ "cs.CL" ]
[ "Computer Science" ]
2023-06-01T00:00:00
https://arxiv.org/abs/2306.00978
https://arxiv.org/pdf/2306.00978v6
2306.00978
10.1145/3714983.3714987
1,454
217
true
https://github.com/mit-han-lab/llm-awq
Conference on Machine Learning and Systems
1
3e23452c9fc523b106c8710f61fc345800751af58ed7ce7f0fa2af1b0a718721
[ "arxiv", "semantic_scholar" ]
SpecInfer: Accelerating Generative Large Language Model Serving with Tree-based Speculative Inference and Verification
This paper introduces SpecInfer, a system that accelerates generative large language model (LLM) serving with tree-based speculative inference and verification. The key idea behind SpecInfer is leveraging small speculative models to predict the LLM's outputs; the predictions are organized as a token tree, whose nodes e...
[ "Xupeng Miao", "Gabriele Oliaro", "Zhihao Zhang", "Xinhao Cheng", "Zeyu Wang", "Zhengxin Zhang", "Rae Ying Yee Wong", "Alan Zhu", "Lijie Yang", "Xiaoxiang Shi", "Chunan Shi", "Zhuoming Chen", "Daiyaan Arfeen", "Reyna Abhyankar", "Zhihao Jia" ]
[ "cs.CL", "cs.DC", "cs.LG" ]
[ "Computer Science" ]
2023-05-16T00:00:00
https://arxiv.org/abs/2305.09781
https://arxiv.org/pdf/2305.09781v4
2305.09781
10.1145/3620666.3651335
368
55
true
https://github.com/flexflow/FlexFlow/
International Conference on Architectural Support for Programming Languages and Operating Systems
0.8741
fb6008aa313ec9941aa1e1a1e593086f2dccf2fcedde4e9dbee3ce4576e8af0b
[ "arxiv", "semantic_scholar" ]
Accelerating Large Language Model Decoding with Speculative Sampling
We present speculative sampling, an algorithm for accelerating transformer decoding by enabling the generation of multiple tokens from each transformer call. Our algorithm relies on the observation that the latency of parallel scoring of short continuations, generated by a faster but less powerful draft model, is compa...
[ "Charlie Chen", "Sebastian Borgeaud", "Geoffrey Irving", "Jean-Baptiste Lespiau", "Laurent Sifre", "John Jumper" ]
[ "cs.CL" ]
[ "Computer Science" ]
2023-02-02T00:00:00
https://arxiv.org/abs/2302.01318
https://arxiv.org/pdf/2302.01318v1
2302.01318
10.48550/arXiv.2302.01318
908
108
false
null
arXiv.org
1
1adec7fc5b299c3e07eba043ad02b9090d6e8249d04ad85781ab58da756bcca6
[ "arxiv", "semantic_scholar" ]
Fast Inference from Transformers via Speculative Decoding
Inference from large autoregressive models like Transformers is slow - decoding K tokens takes K serial runs of the model. In this work we introduce speculative decoding - an algorithm to sample from autoregressive models faster without any changes to the outputs, by computing several tokens in parallel. At the heart o...
[ "Yaniv Leviathan", "Matan Kalman", "Yossi Matias" ]
[ "cs.LG", "cs.CL" ]
[ "Computer Science" ]
2022-11-30T00:00:00
https://arxiv.org/abs/2211.17192
https://arxiv.org/pdf/2211.17192v2
2211.17192
10.48550/arXiv.2211.17192
1,646
279
false
null
International Conference on Machine Learning
1
da28afc96931be423a3c8200370b6484947e29f2a8f0b3fde8a1a79ee9ff0ace
[ "arxiv", "semantic_scholar" ]
Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation
We propose Speculative Decoding (SpecDec), for the first time ever, to formally study exploiting the idea of speculative execution to accelerate autoregressive (AR) decoding. Speculative Decoding has two innovations: Spec-Drafter -- an independent model specially optimized for efficient and accurate drafting -- and Spe...
[ "Heming Xia", "Tao Ge", "Peiyi Wang", "Si-Qing Chen", "Furu Wei", "Zhifang Sui" ]
[ "cs.CL", "cs.LG" ]
[ "Computer Science" ]
2022-03-30T00:00:00
https://arxiv.org/abs/2203.16487
https://arxiv.org/pdf/2203.16487v6
2203.16487
10.18653/v1/2023.findings-emnlp.257
183
12
true
https://github.com/hemingkx/SpecDec
Conference on Empirical Methods in Natural Language Processing
0.5662
e965fbcd433831dc0def0d0120e14b36cb2dd4d0b53cff938308d76ba597b58a
[ "arxiv", "semantic_scholar" ]
Causal Inference Using Tractable Circuits
The aim of this paper is to discuss a recent result which shows that probabilistic inference in the presence of (unknown) causal mechanisms can be tractable for models that have traditionally been viewed as intractable. This result was reported recently to facilitate model-based supervised learning but it can be interp...
[ "Adnan Darwiche" ]
[ "cs.AI", "cs.CC", "cs.LG", "cs.LO", "stat.ME" ]
[ "Computer Science", "Mathematics" ]
2022-02-07T00:00:00
https://arxiv.org/abs/2202.02891
https://arxiv.org/pdf/2202.02891v1
2202.02891
null
23
3
false
null
arXiv.org
0.3451