{"id": "989fef5fc253b0613a59b01cec35c0cc584fc413ce901e3b5192ea8bd456c69e", "sources": ["arxiv"], "title": "SSD: Spatially Speculative Decoding Accelerates Autoregressive Image Generation", "abstract": "Autoregressive models excel in visual generation by treating images as 1D sequences of discrete tokens, mirroring language modeling. However, this flattening discards the intrinsic 2D spatial locality of visual signals, creating severe computational bottlenecks during inference. We introduce Spatially Speculative Decoding (SSD), a framework that aligns the predictive objective with the natural geometry of images. Rather than predicting only the immediate next token in a 1D sequence, our model simultaneously predicts the adjacent horizontal token and the token directly below it. By capitalizing on this 2D spatial correlation, spatially speculative decoding overcomes the memory wall in visual inference. Our approach accelerates autoregressive image generation by up to 13.3x while maintaining high fidelity on DPG-Bench and GenEval. Our results suggest that respecting the underlying geometry of vision unlocks massive computational efficiencies, paving the way for real-time, high-resolution autoregressive generative models.", "authors": ["Shilong Xiang", "Zirui Zhang", "Lijun Yu", "Chengzhi Mao"], "categories": ["cs.CV"], "fields_of_study": [], "published_date": "2026-06-18", "url": "https://arxiv.org/abs/2606.20543", "pdf_url": "https://arxiv.org/pdf/2606.20543v1", "arxiv_id": "2606.20543", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "d6953c2d8a4c8fbbfaf4feaf13b58c3bca43fa126b3c3ad1191d86913ae7b0e1", "sources": ["arxiv"], "title": "EfficientRollout: System-Aware Self-Speculative Decoding for RL Rollouts", "abstract": "Reinforcement learning (RL) has become a representative post-training paradigm for LLMs, enabling strong reasoning and agentic capabilities. However, rollout generation remains a dominant latency bottleneck because autoregressive sampling decodes responses sequentially and a small number of long-tailed generations often determine completion time. Speculative decoding (SD) offers a natural way to address this bottleneck, as it is a well-established technique for serving fixed LLMs that reduces latency by rapidly drafting tokens and accepting them through parallel verification while preserving the target-model distribution. However, its practical speedups do not directly carry over to RL rollouts: (i) the evolving target policy makes any fixed drafter increasingly mismatched with the policy's output distribution; and (ii) active batch sizes shrink throughout rollout decoding, shifting decoding from compute-bound to memory-bound regimes where parallel verification can exploit underutilized compute. Therefore, accelerating RL rollouts requires both a drafter that remains effective under long, high-temperature generations from an evolving policy and system-aware use of SD that avoids compute-bound regimes. We present EfficientRollout, a system-aware self-SD framework designed to address this gap for RL rollouts. EfficientRollout induces a quantized drafter from the target model (i.e. self-speculative decoding), keeping it coupled to the evolving policy without separate drafter pretraining or online adaptation. It further coordinates a system-aware SD toggle policy with acceptance-aware draft-length adaptation, enabling speculation only in beneficial regimes while matching the drafting budget to evolving drafter quality. EfficientRollout reduces rollout and end-to-end latency by up to 19.6% and 12.7%, respectively, over an accelerated AR rollout baseline, while preserving final model quality.", "authors": ["Minseo Kim", "Minjae Lee", "Seunghyuk Oh", "Kevin Galim", "Donghoon Kim", "Coleman Hooper", "Harman Singh", "Amir Gholami", "Hyung Il Koo", "Wonjun Kang"], "categories": ["cs.LG"], "fields_of_study": [], "published_date": "2026-06-17", "url": "https://arxiv.org/abs/2606.18967", "pdf_url": "https://arxiv.org/pdf/2606.18967v1", "arxiv_id": "2606.18967", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/furiosa-ai/EfficientRollout", "venue": null, "quality_score": 0.65}
{"id": "e6b1bda0dd2caa8573a59a60e8a011d3ac98fec3b34a3f58fa4f3c9d97fba360", "sources": ["arxiv"], "title": "JetFlow: Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting", "abstract": "Speculative decoding (SD) accelerates autoregressive Large Language Models (LLMs) by drafting multiple tokens and verifying them in parallel, but it faces a scaling limitation: increasing the draft budget improves speed only when acceptance remains high and drafting overhead stays low. This ceiling has been difficult to break because prior head-based SD methods face a causality-efficiency dilemma. Autoregressive drafters produce path-conditioned candidates that are effective for tree speculative decoding with higher acceptance length, but their drafting cost grows with tree depth. Bidirectional block-diffusion drafters generate all positions in one pass, but their branch-agnostic marginals can form individually plausible yet mutually inconsistent trees, wasting budget and reducing acceptance. We propose JetFlow, a head-based SD framework that combines one-forward drafting efficiency with branch-wise causal conditioning. JetFlow trains a causal parallel draft head over fused hidden states from the frozen target model, producing candidate trees whose scores align with the target model's autoregressive factorization. This enables JetFlow to convert larger draft budgets into longer accepted prefixes and higher end-to-end speedup. Across math, coding, and chat benchmarks on dense and MoE Qwen3 models, JetFlow consistently outperforms bidirectional-head and tree-based SD baselines. On H100 GPUs, JetFlow achieves up to 9.64x speedup on MATH-500 and 4.58x on open-ended conversational workloads, with further latency gains demonstrated through vLLM integration under realistic serving loads. Our code and models are available at https://github.com/hao-ai-lab/JetFlow.", "authors": ["Lanxiang Hu", "Zhaoxiang Feng", "Yulun Wu", "Haoran Yuan", "Yujie Zhao", "Yu-Yang Qian", "Bojun Wang", "Daxin Jiang", "Yibo Zhu", "Tajana Rosing", "Hao Zhang"], "categories": ["cs.CL"], "fields_of_study": [], "published_date": "2026-06-16", "url": "https://arxiv.org/abs/2606.18394", "pdf_url": "https://arxiv.org/pdf/2606.18394v1", "arxiv_id": "2606.18394", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/hao-ai-lab/JetFlow", "venue": null, "quality_score": 0.65}
{"id": "e7d194915f2cea21b85eb67bb8325e9557465399c70a6708c0c19dfea7a2b985", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Speculative Diffusions via Block Verification", "abstract": "Speculative decoding speeds up LLM inference by using a draft model to generate tokens, with an acceptance-rejection scheme that ensures that the output matches the target distribution. Adapting this to continuous diffusions is difficult because speculative sampling requires drawing from a residual distribution. While straightforward in discrete spaces, efficiently sampling this residual in continuous space is non-trivial. Consequently, existing diffusion adaptations either use computationally inefficient sampling techniques or rely on an alternative scheme. In this work, we introduce a novel scheme that efficiently implements the original speculative sampling mechanism for diffusion models. Our approach offers a critical advantage over current methods: it enables us to adapt block verification from LLMs to diffusions -- which provably improves the acceptance rate of drafts. Furthermore, we formalize and analyze the Free Drafter, a heuristic self-speculative drafter for diffusions that requires no training. By enabling block verification, our Free Drafter yields up to a 6.3% speedup over existing speculative methods with no additional training and negligible overhead beyond the existing parallel verification pass.", "authors": ["Alexander Soen", "Hisham Husain", "Valentin De Bortoli", "Arnaud Doucet"], "categories": ["cs.LG", "stat.ML"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-06-11", "url": "https://arxiv.org/abs/2606.13426", "pdf_url": "https://arxiv.org/pdf/2606.13426v1", "arxiv_id": "2606.13426", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "9835e0ad1804d5b57d689525445d89ace6b367cb70c1fcd0dc5e0ca3918f2874", "sources": ["arxiv", "semantic_scholar"], "title": "VIA-SD: Verification via Intra-Model Routing for Speculative Decoding", "abstract": "Speculative decoding (SD) addresses the high inference costs of LLMs by having lightweight drafters generate candidates for large verifiers to validate in parallel. Existing draft-verify methods use binary decisions: accept or fully recompute. Yet we find that many rejected tokens can be verified correctly by a slim submodel derived from the full verifier via intra-model routing, instead of the full verifier. This motivates our slim-verifier to handle tokens requiring moderate verification resources, reducing expensive large-model calls. We propose Verification via Intra-Model Routing for Speculative Decoding (VIA-SD), a multi-tier framework using a routed slim-verifier. Draft tokens are processed hierarchically: direct acceptance for high-confidence cases, slim-verifier regeneration for medium-confidence cases, and full-model verification for uncertain cases. Across four representative tasks and multiple model families, VIA-SD reduces rejection rates by 0.10-0.22 and delivers 10-20% speedups over strong SD baselines, while achieving 2.5-3x acceleration over non-drafting decoding. Moreover, VIA-SD is compatible with existing SD frameworks without modifying their training procedures. Our results suggest multi-tier SD as a general paradigm for scalable and efficient LLM inference. Project page: https://zju-xyc.github.io/VIA-SD-Project-Page/", "authors": ["Yuchen Xian", "Yang He", "Yunqiu Xu", "Yi Yang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-10", "url": "https://arxiv.org/abs/2606.12243", "pdf_url": "https://arxiv.org/pdf/2606.12243v1", "arxiv_id": "2606.12243", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "0575977791d2291d5bbfd253a988f4ac64d1c8e3d8206b49af1ca202bb46201b", "sources": ["arxiv", "semantic_scholar"], "title": "PathRelax: Parallel-Path Relaxed Speculative Jacobi Decoding for Accelerating Auto-Regressive Text-to-Image Generation", "abstract": "The growing need for high-resolution image generation in autoregressive text-to-image models has resulted in extended token sequences, significantly increasing computational costs and inference times. However, existing state-of-the-art methods for accelerating autoregressive text-to-image models rely on chain-structured draft token sequences, leading to inefficient draft token search and limited acceptance lengths. To address this, we propose parallel-path cross-relaxed speculative Jacobi decoding (\\textbf{PathSpec}), a novel framework that enhances efficiency through a multi-sequence draft tree structure. Our parallel-path speculative Jacobi decoding (\\textbf{PathExplore}) expands the token search space, achieving a higher speedup ratio without sacrificing image quality. Additionally, we introduce cross-path relaxed verification (\\textbf{PathRelax}) that exploits semantic similarities across sequences to further boost token acceptance rates. Evaluated on the Parti-Prompts, MSCOCO2017, and T2ICompBench datasets, our method achieves a speedup ratio of 4.14 $\\times$, 3.95$\\times$, and 4.18$\\times$, respectively. Remarkably, PathExplore, without any relaxed sampling, outperforms relaxed sampling methods in the speedup ratio, such as GSD and LANTERN. Moreover, PathRelax's relaxation mechanism can be seamlessly integrated with other relaxation techniques, enabling further acceleration and providing an efficient solution for real-time text-to-image generation. Our code is available at https://github.com/Haodong-Lei-Ray/PathSpec.", "authors": ["Haodong Lei", "Hongsong Wang", "Bingxuan Dai", "Pan Zhou"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-09", "url": "https://arxiv.org/abs/2606.10492", "pdf_url": "https://arxiv.org/pdf/2606.10492v1", "arxiv_id": "2606.10492", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Haodong-Lei-Ray/PathSpec", "venue": null, "quality_score": 0.65}
{"id": "6ced2fae8c1c2631d3c6e33145f50e7d59272f75d55da532305c73c60ee63c94", "sources": ["arxiv", "semantic_scholar"], "title": "TRADE: Transducer-Augmented Decoder for Speech LLM", "abstract": "Speech Large Language Models (Speech LLMs) lack a principled mechanism for streaming inference: their label-synchronous generation has no acoustic-frame alignment, making real-time decoding and end-of-utterance detection difficult. We propose TRADE TRansducer-Augmented DEcoder, which augments a multimodal LLM with a transducer branch that shares the audio encoder and uses the LLM's hidden states directly as the prediction network -- coupling frame-synchronous acoustic alignment with the LLM's linguistic reasoning. Three design choices make the system accurate, streamable, and long-form capable: (1)Tightly coupled dual vocabularies -- a compact transducer vocabulary derived from the LLM vocabulary, enabling zero-cost score fusion; (2)Chunk-synchronized streaming training with gradient stopping, eliminating the train-inference mismatch at offline-equivalent memory cost; and (3)Localized Decoder Audio Attention (LDAA), a causal sliding window that caps KV-cache memory independently of utterance length. A single TRADE checkpoint supports offline and streaming decoding across a continuous range of latency operating points. TRADE achieves 6.71% average WER on the Open ASR Leaderboard, while the streaming recognition with 960ms chunk size reaches 8.40% from the same checkpoint. On long-form speech, it obtains 3.64% WER on TED-LIUM and 10.88% on Earnings-22 without external segmentation. TRADE provides sentence-end punctuation timestamps that, when combined with acoustic voice activity detection (VAD), improve end-of-utterance detection by +0.03 F_1 over acoustic VAD alone.", "authors": ["Yun Tang", "Shanil Puri", "Shinji Watanabe", "Subhabrata Mukherjee"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-07", "url": "https://arxiv.org/abs/2606.08486", "pdf_url": "https://arxiv.org/pdf/2606.08486v1", "arxiv_id": "2606.08486", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "99301d0dea92075618a6928d093fa57ef75c64b427ac22f38fce006a762521c4", "sources": ["arxiv", "semantic_scholar"], "title": "WhiFlash: Accelerating Speculative Decoding with Token-Level Cross-Paradigm Routing", "abstract": "The autoregressive nature of large language models (LLMs) remains a significant bottleneck for inference, particularly in complex agentic workloads. While speculative decoding (SD) accelerates inference, current approaches rely on static drafting paradigms, utilising either autoregressive drafting models for reasoning or diffusion-based parallel drafting models for structured outputs. We empirically find that drafting accuracy fluctuates dramatically within a single sequence, leaving significant performance unrealised by static paradigms and coarse-grained routing. To address this volatility, we introduce WhiFlash, the first cross-paradigm SD method that unifies autoregressive and diffusion-based parallel drafting under a single token-level controller. WhiFlash adopts a fine-grained routing mechanism that employs either a lightweight entropy-based or a learned neural policy, both parametrised to provide a tunable balance between expected token gain and latency. To make high-frequency switching computationally viable, we introduce novel cache-management optimisations, Lazy Catch-up and KV-only Prefill, reducing switching overhead to below 7% of per-round latency. By capitalising on the complementary strengths of fundamentally distinct drafting architectures, WhiFlash achieves significantly higher acceptance lengths, yielding category-specific throughput gains of up to 69.6% over the state-of-the-art autoregressive EAGLE-3 and 37.3% over the diffusion-based DFlash.", "authors": ["Young D. Kwon", "Miles Williams", "Rui Li", "Alexandros Kouris", "Stylianos I. Venieris"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-05", "url": "https://arxiv.org/abs/2606.07710", "pdf_url": "https://arxiv.org/pdf/2606.07710v1", "arxiv_id": "2606.07710", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "5ea3182923a447700ff7de9a75dc9d912a92dbc16a5d4fdefccb458d42d9e167", "sources": ["arxiv", "semantic_scholar"], "title": "YouZhi: Towards High-Concurrency Financial LLMs via Adaptive GQA-to-MLA Transition", "abstract": "Large language models (LLMs) drive significant financial innovations, yet their high-concurrency deployment is severely bottlenecked by KV cache memory overhead, which inflates infrastructure costs and throttles scalability. To address this, we propose YouZhi-LLM, a highly efficient financial LLM empowered by a comprehensive structural transition and training pipeline natively built on the Huawei Ascend ecosystem. At its algorithmic core, YouZhi-LLM features a layer-adaptive GQA-to-MLA transition framework that dynamically assigns per-layer FreqFold sizes, maximizing KV-cache compression while minimizing perplexity degradation. To recover representation capacity and inject domain expertise, the Ascend-based training pipeline seamlessly integrates generalized knowledge distillation with financial-specific supervised fine-tuning. Evaluations demonstrate the superiority of this systematic approach, with the adaptive transition reducing perplexity degradation by up to 35% over uniform baselines. Crucially, when evaluated on Ascend NPUs via vLLM-Ascend, the massive KV-cache reduction translates directly into deployment efficiency. Compared to their respective base models, YouZhi-7B yields a 12.3% improvement in average financial benchmark score alongside a 2.69$\\times$ increase in maximum concurrency; similarly, YouZhi-14B achieves a 7.0% accuracy gain and a 2.43$\\times$ concurrency boost, establishing a new paradigm for cost-effective, high-throughput financial inference.", "authors": [" PSBC LLM Team", " Huawei LLM Team", "Ruihan Long", "Junjie Wu", "Tianan Zhang", "Duo Zhang", "Yaozong Wu", "Jinbin Fu", "Chang Liu", "Zhentao Tang", "Wenshuang Yang", "Xin Wang", "Zhihao Song", "Ning Huang", "Wenjing Xu", "Shuai Zong", "Shupei Sun", "Sen Wang", "Jing Hu", "Bin Wang", "Xinyu Wang", "Junkui Ju", "Zequn Ding", "Jie Ran", "Man Luo", "Shixiong Kai", "Linkai Hou", "Kaichao Liang", "Hu Zhao", "Yang Zhao", "Shucheng Lin", "Wei Yu", "Chenghan Jiang", "Jingjing Ding", "Jiahui Zhang", "Tian Jin", "Yuhang Zhang", "Dong Guo", "Wei Sun", "Jun Xie", "Jianwei Li", "Lei Cao", "Pei Li", "Jiabin Li", "Jia Yuan", "Rui Yuan", "Jing Zhu", "Mingxuan Yuan", "Zhangcheng Lv", "Xin Jiang", "Xiuhong Fei", "Xiaozhe Ren", "Yulong Li", "Zhipeng Zhang", "Hang Wang", "Zhaohui Xu", "Rui Zhao", "Yibo He", "Xinzhuang Niu"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-04", "url": "https://arxiv.org/abs/2606.05868", "pdf_url": "https://arxiv.org/pdf/2606.05868v1", "arxiv_id": "2606.05868", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "0d4ec2a1e770ac73e7ea7b4d3e239d8bcf1fb1c16fc4ebfdbbffb26fc691afd3", "sources": ["arxiv", "semantic_scholar"], "title": "AdaPLD: Adaptive Retrieval and Reuse for Efficient Model-Free Speculative Decoding", "abstract": "Speculative decoding accelerates generation by verifying multiple drafted tokens in a single target-model forward pass, reducing sequential decoding iterations. Model-free variants avoid auxiliary draft models by reusing text and model states already available during generation, but their speedup depends on the reliability of the constructed drafts. We identify two limitations of existing reuse-based methods: lexically anchored retrieval has limited recall under surface-form variation, and deterministic span copying can be brittle when the retrieved context does not uniquely determine the continuation. We propose \\emph{AdaPLD}, a training-free method that adaptively improves both retrieval and draft construction. AdaPLD preserves high-precision lexical reuse while using semantic similarity to recover additional reuse opportunities when lexical matching fails. It further constructs branched reuse hypotheses to account for continuation uncertainty, rather than relying on a single copied span. Across diverse benchmarks, AdaPLD reduces target-model forward passes and achieves up to $3.10\\times$ decoding speedup.", "authors": ["Runheng Liu", "Jincheng Xie", "Wen Hu", "Xingchen Xiao", "Heyan Huang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-04", "url": "https://arxiv.org/abs/2606.05742", "pdf_url": "https://arxiv.org/pdf/2606.05742v2", "arxiv_id": "2606.05742", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "b59b2e4af5fd17912a3853612cec760e54c9bf310adf06402a54003a30186ded", "sources": ["arxiv", "semantic_scholar"], "title": "D^2SD: Accelerating Speculative Decoding with Dual Diffusion Draft Models", "abstract": "Speculative decoding accelerates autoregressive large language model inference by drafting multiple tokens and verifying them in a single target-model forward pass. Recent diffusion-based drafters generate an entire block of tokens in parallel but usually commit to a single draft sequence per verification: once the first mismatch occurs, all subsequent draft tokens are discarded, resulting in a limited acceptance rate. Naively batching more draft candidate sequences only introduces a marginal improvement, as redundant or poorly placed branches increase the cost of drafting and verification without proportionally increasing the number of accepted tokens. We propose D^2SD, a dual diffusion draft speculative decoding framework that organizes candidates into a confidence-guided prefix tree, where the first diffusion drafter generates a block along with per-position confidence scores that are used to identify the most likely rejection boundary and select the top-K prefix ranges for recovery; the second variable-prefix diffusion drafter re-anchors at each selected prefix and proposes alternative continuations in one batched pass; the resulting shared-prefix candidates are jointly verified via cascade attention. Empirically, D^2SD shows clear improvements over both the underlying diffusion approach and strong autoregressive speculative decoding baselines.", "authors": ["Liyuan Zhang", "Jiarui Zhang", "Jinwei Yao", "Ran Yan", "Yuchen Yang", "Jiahao Zhang", "Tongkai Yang", "Yi Wu", "Binhang Yuan"], "categories": ["cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-03", "url": "https://arxiv.org/abs/2606.04446", "pdf_url": "https://arxiv.org/pdf/2606.04446v1", "arxiv_id": "2606.04446", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "7bfd76541cf7094864c120ea7c6ab296bf68599cade499b5e32c163a38559bc0", "sources": ["arxiv", "semantic_scholar"], "title": "TreeFlash: Parallel AR-Approximation for Faster Speculative Decoding", "abstract": "One-shot block drafters for speculative decoding generate the full draft in a single forward pass, achieving strong throughput by eliminating sequential token generation. However, they predict each draft token conditioned only on the prefix context, with no dependence on previously drafted tokens. This non-autoregressive conditioning causes the drafter's distribution to diverge from the verifier's true autoregressive distribution as draft depth grows. This problem becomes more severe in tree-based drafting, where distinct branches are forced to share the same marginal distribution for subsequent tokens. We propose TreeFlash, which addresses this by incorporating an MLP layer conditioned on the drafter's hidden state and the previous token to approximate an autoregressive distribution. TreeFlash retains the $\\mathcal{O}(1)$ decoding time complexity of one-shot drafters by employing a two-stage approximation mechanism. TreeFlash achieves state-of-the-art performance across a variety of tasks and models, improving over marginal tree drafting by $12\\%$ higher block efficiency and $9\\%$ higher speedup.", "authors": ["Peer Rheinboldt", "Frédéric Berdoz", "Roger Wattenhofer"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-02", "url": "https://arxiv.org/abs/2606.03819", "pdf_url": "https://arxiv.org/pdf/2606.03819v1", "arxiv_id": "2606.03819", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "a4d7172a68c2513a2ea474b2fd98d98799697e94d04d32097615ad79a8bb5bff", "sources": ["arxiv", "semantic_scholar"], "title": "SimSD: Simple Speculative Decoding in Diffusion Language Models", "abstract": "Diffusion large language models (dLLMs) have recently emerged as a promising alternative to autoregressive (AR) LLMs, offering faster inference through parallel or blockwise decoding. However, their masked language modeling formulation remains incompatible with standard token-level speculative decoding, one of the most effective acceleration techniques for AR models. In AR decoding, the causal mask preserves temporally valid token-level contexts, enabling a target model to verify multiple drafted tokens in a single forward pass. In contrast, dLLMs rely on mask tokens and bidirectional attention, causing the effective context to change across denoising steps and preventing direct token-level speculative verification. To bridge this gap, we propose a simple but effective speculative decoding algorithm for diffusion language models, named SimSD, which mainly adopts a plug-and-play masking strategy that equips dLLMs with temporally valid token-level contexts for speculative decoding. Our method explicitly introduces reference tokens from draft-model predictions and designs an attention mask that regulates their interaction with current-step tokens, allowing dLLMs to compute valid logits for drafted tokens in a single forward pass. This restores the key verification ability provided by causal masking in AR models while preserving the parallel decoding advantages of dLLMs. The proposed method is training-free and can be flexibly integrated with other acceleration techniques such as KV cache and blockwise decoding. Experiments on SDAR-family dLLMs across four benchmarks show that our method achieves up to 7.46x higher decoding throughput while maintaining and even improving average generation quality.", "authors": ["Junxia Cui", "Haotian Ye", "Runchu Tian", "Hongcan Guo", "Jinya Jiang", "Haoru Li", "Chaojie Ren", "Yiming Huang", "Kaijie Zhu", "Zhongkai Yu", "Kun Zhou", "Jingbo Shang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-01", "url": "https://arxiv.org/abs/2606.02544", "pdf_url": "https://arxiv.org/pdf/2606.02544v1", "arxiv_id": "2606.02544", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/airevo2/SimSD-release", "venue": null, "quality_score": 0.65}
{"id": "ce69ae40ce9fed8ab2bc6e71c105119c207ec66e9a30dcb3c5bf93b5a3d25ccb", "sources": ["arxiv", "semantic_scholar"], "title": "Fast-dLLM++: Fréchet Profile Decoding for Faster Diffusion LLM Inference", "abstract": "Diffusion large language models promise parallel token generation, yet inference remains bottlenecked by deciding which masked tokens can be safely committed together. Fast-dLLM addressed this with KV caching and confidence-guided parallel decoding, but its decoding theory uses a homogeneous high-confidence assumption that effectively reduces each candidate set to its weakest selected token. We argue that this leaves speed on the table because real decoding steps exhibit heterogeneous confidence profiles. We propose \\textbf{Fast-dLLM++}, a training-free extension that introduces \\emph{Fréchet profile decoding}: selecting parallel commit sets from the full sorted confidence profile rather than a single worst-case confidence. The resulting rule is a heterogeneous-confidence generalization of Fast-dLLM's factor selector and it recovers the previous rule exactly in the equal-confidence case and adds a provable \\emph{heterogeneity bonus} when the selected tokens have uneven confidences. Fast-dLLM++ leaves the model, diffusion process, and cache implementation entirely unchanged, making it a drop-in replacement for existing Fast-dLLM decoding. Experiments on GSM8K, MATH, HumanEval, and MBPP with the LLaDA-8B model show that the theoretical improvement translates directly into empirical gains: profile-aware selection improves the accuracy--throughput frontier by exploiting safe parallelism that weakest-token rules miss, achieving up to 37\\% higher throughput at comparable accuracy. Our code release is at https://github.com/Ringo-Star/FastdLLM_plusplus.", "authors": ["Siva Rajesh Kasa", "Yasong Dai", "Sumit Negi", "Hongdong Li"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-01", "url": "https://arxiv.org/abs/2606.02955", "pdf_url": "https://arxiv.org/pdf/2606.02955v2", "arxiv_id": "2606.02955", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Ringo-Star/FastdLLM_plusplus", "venue": null, "quality_score": 0.65}
{"id": "ff9c8a7f73a37a9c55cb13592f6c9a8327401cb38f1c1d2844082cddc3ab203d", "sources": ["arxiv", "semantic_scholar"], "title": "DFlare: Scaling Up Draft Capacity for Block Diffusion Speculative Decoding", "abstract": "Block diffusion speculative decoding accelerates LLM inference by predicting all tokens within a block simultaneously for the target model to verify in parallel. Predicting an entire block at once requires a sufficiently capable draft model and effective utilization of the target model's internal knowledge. However, the state-of-the-art method DFlash constrains all draft layers to share a single fused representation derived from only a few target layers, limiting per-layer expressiveness and hindering further scaling of draft capacity. In this paper, we present \\modelname, which flares out the narrow conditioning bottleneck of DFlash through a lightweight layer-wise fusion mechanism: each draft layer attends to its own learnable combination of a broad set of target layers at negligible overhead, simultaneously injecting richer target knowledge and providing every draft layer with a distinct input. This enhanced per-layer expressiveness enables scaling the draft model to deeper architectures with consistent gains. We further scale training data from 800K to 2.4M samples to fully exploit the enlarged capacity. On six benchmarks spanning mathematical reasoning, code generation, and conversation, \\modelname attains average wall-clock speedups of 5.52x on Qwen3-4B, 5.46x on Qwen3-8B, and 3.91x on GPT-OSS-20B, improving over DFlash by roughly 11\\%, 8\\%, and 5\\% respectively. Our code is available at https://github.com/Tencent/AngelSlim.", "authors": ["Jiebin Zhang", "Zhenghan Yu", "Song Liu", "Eugene J. Yu", "Zheng Li", "Dawei Zhu", "Jiangshan Duo", "Weimin Xiong", "Yifan Song", "Guanghua Yu", "Jianchen Zhu", "Sujian Li"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-01", "url": "https://arxiv.org/abs/2606.02091", "pdf_url": "https://arxiv.org/pdf/2606.02091v2", "arxiv_id": "2606.02091", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Tencent/AngelSlim", "venue": null, "quality_score": 0.65}
{"id": "67cb850d1ae2bbad9e335769689abaedd7c3f713ce53eb560d69f14e04724a4c", "sources": ["arxiv", "semantic_scholar"], "title": "Cost-Aware Diffusion Draft Trees for Speculative Decoding", "abstract": "Speculative decoding accelerates inference by having a lightweight drafter propose tokens verified in parallel by the target language model. Block diffusion drafters such as DFlash generate an entire draft block in one pass, yielding per-position marginals; DDTree uses these to build a candidate tree that maximizes expected acceptance length under a fixed node budget. We observe, however, that acceptance length is non-decreasing in budget: it always favors larger trees regardless of verification cost, offering no principled basis for budget selection. We introduce \\textbf{CaDDTree} (Cost-aware Diffusion Draft Tree), a method that directly optimizes token throughput (expected tokens generated per unit time) by jointly selecting the tree structure and node budget. We model draft and verification latencies explicitly, show that the throughput objective decomposes into a per-round one-dimensional search over the budget, and prove that under a convex verification cost the throughput function is \\emph{unimodal}, enabling an efficient greedy stopping rule. CaDDTree requires no offline budget search, adapting the budget each round from the current per-position distributions and verification cost. Experiments on Qwen3-4B and Qwen3-8B across eight benchmarks spanning reasoning, coding, and instruction-following tasks show that \\caDDTree{} matches or surpasses DDTree with oracle budget selection on nearly all tasks.", "authors": ["Shuai Zhang", "Huachuan Qiu", "Hongliang He", "Yong Dai"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-06-01", "url": "https://arxiv.org/abs/2606.01813", "pdf_url": "https://arxiv.org/pdf/2606.01813v1", "arxiv_id": "2606.01813", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "9aaa22aa6ff1b729f41b82c2327c674ae94b8d7e4fa885340d0b12ae2283d19d", "sources": ["arxiv", "semantic_scholar"], "title": "Hybrid Verified Decoding: Learning to Allocate Verification in Speculative Decoding", "abstract": "Large Language Model (LLM) generation remains expensive because autoregressive decoding calls the model once for each new token. Speculative decoding reduces this cost by drafting multiple tokens and verifying them with the target model in one step, but its speedup depends on how many drafted tokens are accepted. Parameter-free draft sources can propose long continuations at low cost in structured and agentic workloads, yet a cache match that looks promising at one generation step may have low payoff at the next. We propose Hybrid Verified Decoding, which predicts the accepted length of a cache draft before verification and uses this payoff estimate to choose between cache verification and a model-based drafter. Across three LLMs and sixteen datasets, Hybrid Verified Decoding is especially effective on agentic workflows, where it outperforms EAGLE3 in every setting with a 2.73x average speedup. Our analysis shows how prompt structure creates cache opportunities, how high-payoff cache drafts concentrate in a small part of the draft space, and how payoff-guided selection reduces sequential decoding work, pointing to runtime draft selection as a promising direction for speculative decoding.", "authors": ["Xin Su", "Dawid Majchrowski", "Fangyuan Yu", "Vanshil Atul Shah", "Sebastian Rogawski", "Pawel Morkisz", "Anahita Bhiwandiwalla", "Phillip Howard"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-31", "url": "https://arxiv.org/abs/2606.01019", "pdf_url": "https://arxiv.org/pdf/2606.01019v1", "arxiv_id": "2606.01019", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "9dcd3f203797ef30a7139152faf9fdacb037877a4e76d5d63f75f843665a6110", "sources": ["arxiv", "semantic_scholar"], "title": "DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation", "abstract": "Speculative decoding (SD) has proven to be an effective technique for accelerating autoregressive generation in large language models (LLMs) however, its application to vision-language models (VLMs) remains relatively unexplored. We propose~\\textit{DREAM-S}, a novel SD framework designed specifically for fast and efficient decoding in VLMs. DREAM-S leverages a neural architecture search (NAS) framework with target-aware supernet training to automatically identify both the optimal interaction strategy between the draft and target models, and the most suitable draft model architecture for the underlying hardware implementation platform. DREAM-S additionally incorporates adaptive intermediate feature distillation, guided by attention entropy, to enable efficient draft training. Experiments on a range of well-established VLMs show that DREAM-S achieves up to a $3.85\\times$ speedup compared to standard decoding approaches and significantly outperforms existing SD baselines. The code is publicly available at: https://github.com/SAI-Lab-NYU/DREAM-S .", "authors": ["Zining Liu", "Yunhai Hu", "Tianhua Xia", "Bo Bao", "Eric Sather", "Vithursan Thangarasa", "Sai Qian Zhang"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-30", "url": "https://arxiv.org/abs/2606.00535", "pdf_url": "https://arxiv.org/pdf/2606.00535v1", "arxiv_id": "2606.00535", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/SAI-Lab-NYU/DREAM-S", "venue": null, "quality_score": 0.65}
{"id": "7b5bf44305ea75c39b6a396737e311e99fa30b92251fc32e114e68b2d9a10252", "sources": ["arxiv", "semantic_scholar"], "title": "Doing What They Say, Not What They Reason: Locating the Faithfulness Gap in LLM Agents", "abstract": "Do LLM agents act on the reasoning they state? This question of process fidelity is central to using LLMs in social simulation, yet it is hard to measure where no reference for correct behavior exists. We study it in acontrolled setting, a Texas Poker simulator with a verifiable reference action for every decision by decomposing the faithfulness gap into two steps: reasoning-conclusion and conclusion-action. The two steps behave oppositely.", "authors": ["Yufeng Wang"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-30", "url": "https://arxiv.org/abs/2606.00476", "pdf_url": "https://arxiv.org/pdf/2606.00476v1", "arxiv_id": "2606.00476", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "6cd545696b973a18c08ac72a12a584dc716d804210a521d7a889c32871f8e2a8", "sources": ["arxiv", "semantic_scholar"], "title": "TAPS: Target-Aware Prefix Tree Selection for Diffusion-Drafted Speculative Decoding", "abstract": "Using a diffusion model for parallel drafting is a promising approach for speculative decoding. By predicting tokens at multiple future positions in a single forward pass, diffusion drafters substantially reduce drafting latency. However, this shifts the bottleneck to verification: verifying a single sequence limits acceptance length, while verifying large draft trees incurs excessive target-model latency. We identify a key mismatch in existing draft-tree methods: existing diffusion-tree methods rank nodes by the marginal probability, ignoring that verification is prefix-conditioned. As a result, they may verify unreachable descendants of rejected prefixes, increasing latency with limited acceptance gains. To address this, we propose TAPS, a target-aware prefix selection method that turns diffusion marginals into path-conditioned acceptance estimates. TAPS then selects a compact prefix-closed subtree under a fixed verification budget, improving the acceptance-cost tradeoff rather than simply expanding the draft tree. Experiments across diverse datasets and model families demonstrate that TAPS achieves up to 7.9x lossless end-to-end speedup over vanilla autoregressive decoding, outperforming state-of-the-art DFlash and DDTree by 1.36x and 1.74x respectively. Our work is available at https://anonymous.4open.science/r/TAPS-EMNLP2026-53DD", "authors": ["Zhuoyu Wang", "Junnan Huang", "Xinyu Chen"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-30", "url": "https://arxiv.org/abs/2606.00487", "pdf_url": "https://arxiv.org/pdf/2606.00487v1", "arxiv_id": "2606.00487", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "9f4328520b446d19330d7c4dcbccaf6bd9dbbfc74e444eea7c61e2318d7fb401", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Pipeline Decoding: Higher-Accruacy and Zero-Bubble Speculation via Pipeline Parallelism", "abstract": "Speculative Decoding (SD) accelerates low-concurrency LLM inference by employing a draft-then-verify paradigm. However, mainstream methods typically rely on multi-token prediction, which introduces escalating prediction difficulty and serial drafting latency. To address these, we propose Speculative Pipeline Decoding (SPD), a groundbreaking framework that unlocks the true potential of pipeline parallelism. By partitioning the target LLM into $n$ pipeline stages, SPD allows LLM to process $n$ tokens in parallel to accelerate decoding. To continuous fill the pipeline in single sequence decoding, a speculation module aggregates intermediate features across different pipeline depths to predict the next token, executing strictly in parallel with the target model's pipeline step, to realize bounded difficulty, higher acceptance rates, and zero latency bubbles. Our experiments demonstrate that SPD achieves a significantly higher theoretical speedup compared to mainstream baselines, offering a highly scalable solution for LLM decoding acceleration. Our code is available at https://github.com/yuyijiong/speculative_pipeline_decoding", "authors": ["Yijiong Yu", "Huazheng Wang", "Shuai Yuan", "Ruilong Ren", "Ji Pei"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-29", "url": "https://arxiv.org/abs/2605.30852", "pdf_url": "https://arxiv.org/pdf/2605.30852v1", "arxiv_id": "2605.30852", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/yuyijiong/speculative_pipeline_decoding", "venue": null, "quality_score": 0.65}
{"id": "8f8355730b9437b1a8ad4f174b4109120e3f6c3a11321f06f206f5ee715c2ba2", "sources": ["arxiv", "semantic_scholar"], "title": "BudgetDraft: Acceptance-Aware Multi-View Training for Sparse-KV Speculative Decoding", "abstract": "Speculative decoding speeds up autoregressive decoding by using a drafter to propose multiple tokens that a verifier validates in parallel. In resource-constrained deployments, the drafter uses a sparse KV cache to limit peak GPU memory and end-to-end latency under a fixed KV budget, while the verifier keeps a full KV cache. Mid-to-long context inference (4K--16K context length) is common in real applications. However, naive sparse/full speculative decoding suffers from the sparse/full mismatch as context length grows, causing the acceptance rate to drop quickly. We propose BudgetDraft, a multi-view sparse training method for sparse drafting in mid-to-long inference. The drafter is exposed to multiple sampled KV budgets during training and learns to align each sparse view with one shared full-cache teacher target. BudgetDraft combines an acceptance-aware loss on a full-cache branch with a multi-view loss on a sparse-cache branch, producing a single budget-robust drafter that recovers acceptance across sparsity levels without extra inference-time components. Experimental results on PG-19, LongBench, and LWM show that BudgetDraft achieves up to 6.55x, 4.46x, and 2.10x end-to-end speedup vs AR at 4K, 8K, and 16K context lengths, while keeping the inference pipeline memory-friendly.", "authors": ["Liang He", "Jingbo Wen", "Qishi Zhan", "Yixiong Chen", "Kangning Cui", "Qizhen Lan", "Xilu Wang"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-29", "url": "https://arxiv.org/abs/2606.00144", "pdf_url": "https://arxiv.org/pdf/2606.00144v1", "arxiv_id": "2606.00144", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "6b7e8f9688dd54d8424545515380d722a0edb32b810ce0c2c9343e6299b7882f", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding Across Languages", "abstract": "Speculative decoding has become a crucial component of large language model (LLM) inference, enabling faster generation by drafting multiple tokens and verifying them in parallel. However, small draft models tend to suffer from disproportionately poor multilingual capabilities. Thus, when generating text in a non-English language, speculative decoding is far less effective. We compare three strategies to improve speculative decoding efficiency for eleven languages: finetuning the draft model on task-specific data (translation); finetuning the draft model on unlabeled monolingual corpora; and training simple n-gram draft models on the same monolingual corpora. We evaluate efficiency on translation (from English into the target language) and the held-out task of story generation. We find that while task-specific distillation can significantly improve efficiency, distilled models generalize poorly to a new task. Meanwhile, n-gram draft models, despite lower acceptance rates, consistently provide large speed-ups due to much faster draft generation.", "authors": ["Nirajan Paudel", "Michael Ginn", "Luc De Nardi", "Alexis Palmer"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-28", "url": "https://arxiv.org/abs/2605.30580", "pdf_url": "https://arxiv.org/pdf/2605.30580v1", "arxiv_id": "2605.30580", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "5e8bc4dba4e0946c2cf35ce10d096d6e2fdb902234a5ca8311d8059988d8e206", "sources": ["arxiv", "semantic_scholar"], "title": "Domino: Decoupling Causal Modeling from Autoregressive Drafting in Speculative Decoding", "abstract": "Speculative decoding accelerates LLM inference by drafting multiple tokens and verifying them in parallel with the target model. However, its practical speedup is constrained by the trade-off between draft quality and drafting cost: autoregressive drafters model causal dependencies among draft tokens but incur sequential overhead, while parallel drafters reduce drafting cost but weaken intra-block dependency modeling. In this paper, we propose Domino, a speculative decoding framework that decouples causal dependency modeling from expensive autoregressive draft execution. Domino first uses a parallel draft backbone to produce preliminary draft distributions for the entire block, and then applies a lightweight Domino head to refine them with prefix-dependent causal information. To stabilize teacher-forced causal encoding, we further introduce a base-anchored training curriculum that first strengthens the parallel backbone and then gradually shifts optimization toward the causally corrected final distribution. Experiments on Qwen3 models show that Domino achieves up to \\(5.49\\times\\) end-to-end speedup under the Transformers backend and up to \\(5.8\\times\\) throughput speedup under SGLang serving.", "authors": ["Jianuo Huang", "Yaojie Zhang", "Qituan Zhang", "Hao Lin", "Hanlin Xu", "Linfeng Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-28", "url": "https://arxiv.org/abs/2605.29707", "pdf_url": "https://arxiv.org/pdf/2605.29707v1", "arxiv_id": "2605.29707", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "147639c70b8115768cc73c5f632a6d7d2eb13266393202b58718d927c21f38c3", "sources": ["arxiv", "semantic_scholar"], "title": "Bastion: Budget-Aware Speculative Decoding with Tree-structured Block Diffusion Drafting", "abstract": "Block-diffusion drafters have recently emerged as a powerful alternative for speculative decoding by predicting multiple future-token distributions in a single parallel step. However, since these parallel predictions are sampled from position-wise marginals rather than fully conditioned sequences, committing to a single greedy path often fails to capture the target model's preferred trajectory. To address this, we propose BASTION, a budget-aware speculative decoding framework with tree-based diffusion drafting. Unlike existing methods that rely on static tree topologies, BASTION dynamically constructs query-dependent trees by balancing draft quality against hardware constraints. Our framework integrates three synergistic components: (1) an acceptance surrogate that estimates expected accepted length via path confidence, (2) an online latency estimator that calibrates a hardware-aware roofline model, and (3) an adaptive best-first expansion that grows the tree until marginal gains no longer justify incremental verification costs. BASTION is training-free, preserves the target model's distribution, and requires no per-setting tuning. Across diverse benchmarks and GPU architectures, BASTION achieves up to a 6.61x speedup over standard autoregressive decoding, outperforming state-of-the-art block-diffusion baselines by 39%.", "authors": ["Soowon Oh", "Nam Cao", "Yujin Kim", "Hojung Jung", "Huzama Ahmad", "Sangmin Bae", "Se-Young Yun"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-28", "url": "https://arxiv.org/abs/2605.29727", "pdf_url": "https://arxiv.org/pdf/2605.29727v1", "arxiv_id": "2605.29727", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "be70dc39d496949d095aaa790f0119c32b195a8e0989576371f7628e77995504", "sources": ["arxiv", "semantic_scholar"], "title": "Cassandra: Enabling Reasoning LLMs at Edge via Self-Speculative Decoding", "abstract": "Speculative decoding has emerged as a promising lossless approach for accelerating Large Language Models (LLMs). As reasoning LLMs increasingly suffer from decode-stage overhead and approximation-based methods degrade accuracy, lossless speculative decoding has become essential for efficient inference. However, existing methods still struggle to deliver strong low-batch performance without additional training, limiting practical deployment on consumer devices. To address this challenge, we propose Cassandra, an algorithm-hardware co-designed self-speculative decoding framework optimized for low-batch scenarios. Cassandra constructs a high-performance, training-free draft model through fine-grained data selection. Using optimized pruning and mantissa truncation, it identifies the most salient values in both model weights and the Key-Value (KV) cache, enabling rapid candidate token generation before full-precision parallel verification. Unlike prior self-speculative decoding methods based on layer skipping or structured KV compression, Cassandra achieves significantly higher efficiency. To further reduce the overhead of format conversion between Cassandra representations and standard floating-point formats, we also introduce a lightweight encoder-decoder hardware module designed for seamless integration with commercial GPUs and NPUs. Experimental results show that Cassandra achieves up to 2.41x speedup over the BF16 baseline without additional training. Furthermore, on Llama 3 8B running on an NVIDIA GeForce RTX 4090, Cassandra generates 1.81x more tokens under the same memory budget compared to Eagle-3, a state-of-the-art speculative decoding method.", "authors": ["Soongyu Choi", "Yuntae Kim", "Muyoung Son", "Joo-Young Kim"], "categories": ["cs.AR"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-26", "url": "https://arxiv.org/abs/2605.26558", "pdf_url": "https://arxiv.org/pdf/2605.26558v1", "arxiv_id": "2605.26558", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "416cf4347e2e0bec6775e27965ba73091905b9d5ef5ebdad4c9f6707001619a9", "sources": ["arxiv", "semantic_scholar"], "title": "Beyond the Target: From Imitation to Collaboration in Speculative Decoding", "abstract": "Speculative decoding (SPD) accelerates large language model (LLM) inference by letting a smaller draft model propose multiple future tokens that are verified in parallel by a larger target model. The dominant SPD paradigm treats the target model as the sole reliable teacher, accepting a draft token only when it exactly matches the target prediction. This design implicitly assumes that the target is always the better choice at every position. In practice, this assumption does not hold. Although the draft is the weaker model overall, it is not uniformly inferior at the token level. In a meaningful fraction of cases where draft and target disagree, the draft's choice is the one that leads to the correct final answer. Inspired by this, we introduce \\textbf{Collaborative Speculative Decoding (CoSpec)}, a generalization of SPD that no longer treats the target model as the sole token-level authority. CoSpec trains an arbitration policy via reinforcement learning to decide whether to accept tokens from the draft or target model, selectively accepting draft tokens at mismatches when doing so is likely to yield a correct final answer. Experimental results show that CoSpec maintains substantial speedups while surpassing target-only performance. By shifting the emphasis from imitation to collaboration, CoSpec suggests a new perspective on speculative decoding.", "authors": ["Jinze Li", "Yixing Xu", "Guanchen Li", "Jinfeng Xu", "Shuo Yang", "Yang Zhang", "Xuanwu Yin", "Dong Li", "Edith C. H. Ngai", "Emad Barsoum"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-24", "url": "https://arxiv.org/abs/2605.24793", "pdf_url": "https://arxiv.org/pdf/2605.24793v1", "arxiv_id": "2605.24793", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "0a2277aa69ce5c776cfaf8fbe3fc063c6e90bc08066b695bd32a8a5a3699b43e", "sources": ["arxiv", "semantic_scholar"], "title": "Optimus: Elastic Decoding for Efficient Diffusion LLM Serving", "abstract": "Large language model (LLM) serving is fundamentally limited by inefficient hardware utilization. Autoregressive (AR) decoding underutilizes GPUs due to its strictly sequential execution, while diffusion LLMs (DLLMs) improve throughput by decoding multiple tokens per iteration. However, fixed block-size diffusion decoding exhibits strong load sensitivity: large blocks exploit idle GPU resources under low load, but saturate early and incur substantial redundant computation under high load. As a result, throughput gains vanish beyond saturation, and no single decoding granularity performs well across dynamic serving workloads. We present Optimus, a serving system that enables elastic decoding for diffusion LLMs by dynamically adapting decoding granularity to runtime load. The key idea is to treat decoding granularity as a runtime control variable, balancing GPU utilization and token efficiency. Optimus combines chunked decoding, which enables fine-grained execution without retraining, with saturation-aware scheduling, a closed-loop mechanism that selects chunk sizes based on runtime conditions. Together with system-level optimizations and customized attention kernels, Optimus achieves significant performance improvements while preserving model accuracy. Experiments show that Optimus delivers up to 6.1x throughput improvement over AR decoding and 4.3x improvement over fixed-block diffusion LLM, while maintaining stable performance across diverse load regimes and improving end-to-end serving capacity under latency constraints. The source code is available at https://github.com/dubcyfor3/Optimus.", "authors": ["Chiyue Wei", "Cong Guo", "Bowen Duan", "Junyao Zhang", "Haoxuan Shan", "Yifei Wang", "Yangjie Zhou", "Hai \"Helen\" Li", "Danyang Zhuo", "Yiran Chen"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-24", "url": "https://arxiv.org/abs/2605.24832", "pdf_url": "https://arxiv.org/pdf/2605.24832v1", "arxiv_id": "2605.24832", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/dubcyfor3/Optimus", "venue": null, "quality_score": 0.65}
{"id": "098017fe4a68a6d123d033c48f94c0761c518f8591dda3385cd29a2948cd4038", "sources": ["arxiv", "semantic_scholar"], "title": "FlexDraft: Flexible Speculative Decoding via Attention Tuning and Bonus-Guided Calibration", "abstract": "Speculative decoding accelerates memory-bound LLM inference without quality degradation by using a fast drafter to propose multiple candidate tokens and the target model to verify them in parallel. However, conventional sequential speculative decoding suffers from mutual waiting between drafting and verification, and repeated exchange of intermediate states further increases memory access overhead. Parallel speculative decoding addresses this limitation by performing drafting and verification within a single target forward pass, allowing future drafts to be prepared while current candidates are being verified. Although effective at small batch sizes, existing parallel speculative decoding methods either require costly continual pretraining with quality degradation or suffer from low acceptance rates. More importantly, this paradigm inherently suffers from uncertainty in both the bonus token and the accepted length, leading to draft verification mismatch and causing throughput gains to collapse at large batch sizes. To address these limitations, we introduce FlexDraft, a lossless speculative decoding framework that flexibly adapts to varying batch sizes through three key designs. (1) Attention Tuning enables block diffusion drafting by tuning only the attention projectors of the final few layers on mask tokens, while keeping the autoregressive path frozen to preserve the target distribution and produce high quality drafts with minimal trainable parameters. (2) Bonus-guided Calibration uses a lightweight MLP conditioned on the resolved bonus token to calibrate draft logits, mitigating draft verification mismatch caused by bonus token uncertainty. (3) Flex Decoding dynamically switches between parallel draft and verify at small batch sizes and sequential draft then verify at large batch sizes, and adjusts verification length based on draft confidence to eliminate redundant computation.", "authors": ["Yaojie Zhang", "Jianuo Huang", "Junlong Ke", "Yuhang Han", "Yongji Long", "Tianchen Zhao", "Biqing Qi", "Linfeng Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-19", "url": "https://arxiv.org/abs/2605.20022", "pdf_url": "https://arxiv.org/pdf/2605.20022v1", "arxiv_id": "2605.20022", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "ebd62303ca3680e0604a8750d46aa645cf905886d42d9538aefdf0ed61443223", "sources": ["arxiv", "semantic_scholar"], "title": "SSV: Sparse Speculative Verification for Efficient LLM Inference", "abstract": "Speculative decoding and dynamic sparse attention are two complementary approaches for accelerating long-context LLM inference: the former amortizes target-model execution across multiple verifier queries, while the latter reduces each query's KV-cache working set. Directly combining them, however, exposes a structural mismatch: speculative verification relies on cross-query commonality, whereas dynamic sparse attention assigns query-specific sparse layouts. This mismatch limits KV-block reuse, amplifies NSA's branch-wise overheads, and makes verification strategy selection input- and regime-dependent. We present SSV, a sparse speculative-verification framework that turns dynamic sparse attention into a verification-oriented workload. SSV combines overlap-aware grouped-query execution, refresh/reuse-based NSA kernel fusion, and profile-guided prompt-adaptive orchestration to improve cross-query reuse, reduce selected-index and branch-fusion overheads, and select effective draft-verification strategies under user-specified precision classes. Experiments on NVIDIA H100 GPUs show that SSV achieves up to 3.49x end-to-end throughput over autoregressive NSA decoding and up to 6.86x kernel speedups for sparse speculative verification.", "authors": ["Zhibin Wang", "Ziyu Zhong", "Nuo Shen", "Yuhang Zhou", "Rong Gu", "Sheng Zhong"], "categories": ["cs.OS"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-19", "url": "https://arxiv.org/abs/2605.19893", "pdf_url": "https://arxiv.org/pdf/2605.19893v2", "arxiv_id": "2605.19893", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "c14a952a4a42efb330843fb7d5cdae09ff8aa24c1ce0c2ea16595a1712a5037c", "sources": ["arxiv", "semantic_scholar"], "title": "Mix-Quant: Quantized Prefilling, Precise Decoding for Agentic LLMs", "abstract": "LLM agents have recently emerged as a powerful paradigm for solving complex tasks through planning, tool use, memory retrieval, and multi-step interaction. However, these agentic workflows often introduce substantial input-side overhead, making the compute-intensive prefilling stage a key bottleneck in long-context, multi-turn inference. In this work, we propose Mix-Quant, a simple and effective phase-aware quantization framework for fast agentic inference. We first investigate FP4 quantization in agentic LLM workflows and observe that quantizing the entire inference process can incur significant performance degradation. In contrast, the prefilling stage exhibits substantial quantization redundancy and can therefore be quantized with minimal accuracy loss, despite being the dominant source of computation. Based on this insight, we apply high-throughput NVFP4 quantization to the prefilling phase while preserving BF16 precision for decoding. By decoupling prefilling acceleration from decoding quality, Mix-Quant combines phase-aware algorithmic quantization with hardware-efficient NVFP4 execution to alleviate the inference bottleneck in LLM agents. Extensive experiments across long-context and agentic benchmarks demonstrate that Mix-Quant largely preserves task performance while delivering significant efficiency improvements, achieving up to a 3x speedup during prefilling.", "authors": ["Haiquan Lu", "Zigeng Chen", "Gongfan Fang", "Xinyin Ma", "Xinchao Wang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-19", "url": "https://arxiv.org/abs/2605.20315", "pdf_url": "https://arxiv.org/pdf/2605.20315v1", "arxiv_id": "2605.20315", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "4be3d37b8d8128e8ae64c55f0f86437c437d7f2663f224e7048d640def9c6757", "sources": ["arxiv", "semantic_scholar"], "title": "Lever: Speculative LLM Inference on Smartphones", "abstract": "Large language models (LLMs) are increasingly needed for interactive mobile applications, but high-quality models exceed the limited DRAM available on smartphones. Flash storage can hold larger models, yet flash-backed inference is slow because autoregressive decoding repeatedly invokes the target model and incurs costly I/O. We observe that speculative decoding is a natural fit for this setting: a small draft model can remain in DRAM, while a larger flash-resident target model verifies multiple candidate tokens per invocation. However, existing methods assume server-class accelerators and fail to account for prolonged I/O latency, limited computation parallelism, and irregular speculation execution. We present Lever, an end-to-end system for efficient flash-backed LLM inference on smartphones. Lever jointly optimizes the three stages of speculative decoding under mobile constraints. For drafting, it builds token trees using an I/O- and compute-aware gain-cost objective. For verification, it prunes low-value branches through early-exit prediction to reduce target-model computation. For execution, it maps speculation efficiently across mobile CPU-NPU hardware to improve utilization. Comprehensive evaluations show that Lever reduces inference latency by an average of 2.93x over baseline flash-offloaded inference and 1.50x over conventional speculative decoding, narrowing the latency gap between flash-backed and memory-resident LLM inference.", "authors": ["Tuowei Wang", "Fengzu Li", "Yanfan Sun", "Wei Gao", "Ju Ren"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-16", "url": "https://arxiv.org/abs/2605.16786", "pdf_url": "https://arxiv.org/pdf/2605.16786v1", "arxiv_id": "2605.16786", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "e09688cdb4c0604a4d4ca4a4cdcdfb445f9347986546c5bd91ea9917c7b82b8d", "sources": ["arxiv", "semantic_scholar"], "title": "PSD: Pushing the Pareto Frontier of Diffusion LLMs via Parallel Speculative Decoding", "abstract": "Diffusion large language models (dLLMs) generate text by iteratively denoising masked token sequences. Although dLLMs can predict all masked positions in parallel within each step, the large number of denoising iterations still makes inference expensive. This cost can be reduced spatially by unmasking multiple tokens per step, or temporally by collapsing multiple denoising steps into one verification call. We propose Parallel Speculative Decoding (PSD), a training-free framework that jointly improves inference along both axes. Using the confidence scores from a single forward pass, PSD selects positions to unmask via a configurable, adaptive unmasking policy and constructs multi-depth speculative drafts without extra model calls. A final batched verification pass then applies hierarchical acceptance, keeping the deepest draft that remains consistent with the updated predictions. Experiments on three dLLMs across reasoning and code generation tasks show that PSD achieves favorable trade-offs between inference efficiency and generation quality, reaching up to $5.5\\times$ tokens per forward pass with accuracy comparable to greedy decoding.", "authors": ["Shengyin Sun", "Yiming Li", "Renxi Liu", "Xinqi Li", "Hui-Ling Zhen", "Weizhe Lin", "Chen Chen", "Xianzhi Yu", "Mingxuan Yuan", "Chen Ma"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-15", "url": "https://arxiv.org/abs/2605.15609", "pdf_url": "https://arxiv.org/pdf/2605.15609v1", "arxiv_id": "2605.15609", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "6233bae9530757780e00012ac3b436f2dfb46e51647691339f185f6928c4d269", "sources": ["arxiv", "semantic_scholar"], "title": "An Interpretable Latency Model for Speculative Decoding in LLM Serving", "abstract": "Speculative decoding (SD) accelerates large language model (LLM) inference by using a smaller draft model to propose multiple tokens that are verified by a larger target model in parallel. While prior work demonstrates substantial speedups in isolated or fixed-batch settings, the behavior of SD in production serving systems remains poorly understood: request load varies over time, and effective batch size emerges from the serving system rather than being directly controlled or observed. In this work, we develop a simple and interpretable latency model for SD in LLM serving. We infer effective batch size from request rate using Little's Law and decompose per-request demand into load-independent and load-dependent components for prefill, drafting, and verification. We validate our model using extensive measurements from vLLM across verifier and drafter model sizes, prefill and decode lengths, request rates, draft lengths, and acceptance probabilities. The model accurately describes observed latency, explains why speedups often diminish as server load increases, and characterizes how draft length, acceptance rate, and verifier-drafter size shape latency across serving conditions, with implications for configuring SD in deployed systems. We further show how the framework extends to mixture of experts models, where sparse expert activation changes the effective service costs across load regimes. Together, our results provide a structured framework for understanding SD in real LLM serving systems.", "authors": ["Linghao Kong", "Megan Flynn", "Michael Peng", "Nir Shavit", "Mark Kurtz", "Alexandre Marques"], "categories": ["cs.LG", "cs.PF"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-14", "url": "https://arxiv.org/abs/2605.15051", "pdf_url": "https://arxiv.org/pdf/2605.15051v1", "arxiv_id": "2605.15051", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "e76ad8097ba449fe5930b44fe71d728fffdbd7c97ee1f9462fcc8e0dd934a78a", "sources": ["arxiv", "semantic_scholar"], "title": "Performance-Driven Policy Optimization for Speculative Decoding with Adaptive Windowing", "abstract": "Speculative decoding accelerates LLM inference by having a lightweight draft model propose speculative windows of candidate tokens for parallel verification by a larger target model. In practice, speculative efficiency is often bottlenecked by hard-to-draft positions, where an early mismatch truncates the accepted prefix and invalidates the rest of the speculative window. Most learning-based drafters are still optimized with token-level supervised objectives, even though speculative utility is inherently window-level and prefix-sensitive. We propose PPOW (Performance-Driven Policy Optimization with Adaptive Windowing), a reinforcement learning framework that shifts drafter optimization from token-level imitation to window-level optimization. PPOW combines a Cost-Aware Speedup Reward, a Distribution-Based Proximity Reward, and Adaptive Divergence-Aware Windowing, which prioritizes informative windows with high confidence-weighted draft-target divergence. PPOW achieves average acceptance lengths of 6.29-6.52 and speedups of 3.39-4.36$\\times$ across multiple model families and benchmarks under a unified decoding protocol. These results show that performance-driven window-level optimization is a practical approach to improving speculative decoding efficiency.", "authors": ["Jie Jiang", "Xing Sun", "Ruotian Chen", "Jianan Su", "Kaixin Shen"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-14", "url": "https://arxiv.org/abs/2605.14978", "pdf_url": "https://arxiv.org/pdf/2605.14978v2", "arxiv_id": "2605.14978", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "fd16601b7856f57778cc38080ebd190ab19273c733dd1d25022fcc6626babede", "sources": ["arxiv", "semantic_scholar"], "title": "Factorization-Error-Free Discrete Diffusion Language Model via Speculative Decoding", "abstract": "Discrete diffusion language models improve generation efficiency through parallel token prediction, but standard $X_0$ prediction methods introduce factorization errors by approximating the clean token posterior with independent token-wise distributions. This paper proposes Factorization-Error-Free Discrete Diffusion Language Modeling (FeF-DLLM), which replaces independent clean-token prediction with an exact prefix-conditioned factorization of the clean posterior to better preserve token dependencies. To reduce the sequential cost introduced by prefix conditioning, FeF-DLLM further incorporates speculative decoding within diffusion denoising, accelerating inference while maintaining the parallel prediction and re-masking properties of DLLMs. Theoretically, we prove that FeF-DLLM generates from the true joint distribution and derive its expected acceleration ratio. Experiments on GSM8K, MATH, HumanEval, and MBPP demonstrate that our method improves accuracy by an average of 5.04 percentage points while achieving an average inference speedup of $3.86\\times$.", "authors": ["Xun Fang", "Yunchen Li", "Hang Yuan", "Zhou Yu"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-14", "url": "https://arxiv.org/abs/2605.14305", "pdf_url": "https://arxiv.org/pdf/2605.14305v1", "arxiv_id": "2605.14305", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "9b950f8c90e0109bf8cabf8714f681dce58ea72a67fd29d48a25bb6455820bd5", "sources": ["arxiv", "semantic_scholar"], "title": "Mistletoe: Stealthy Acceleration-Collapse Attacks on Speculative Decoding", "abstract": "Speculative decoding has become a widely adopted technique for accelerating large language model (LLM) inference by drafting multiple candidate tokens and verifying them with a target model in parallel. Its efficiency, however, critically depends on the average accepted length $τ$, i.e., how many draft tokens survive each verification step. In this work, we identify a new mechanism-level vulnerability in model-based speculative decoding: the drafter is trained to approximate the target model distribution, but this approximation is inevitably imperfect. Such a drafter-target mismatch creates a hidden attack surface where small perturbations can preserve the target model's visible behavior while substantially reducing draft-token acceptability. We propose Mistletoe, a stealthy acceleration-collapse attack against speculative decoding. Mistletoe directly targets the acceptance mechanism of speculative decoding. It jointly optimizes a degradation objective that decreases drafter-target agreement and a semantic-preservation objective that constrains the target model's output distribution. To resolve the conflict between these objectives, we introduce a null-space projection mechanism, where degradation gradients are projected away from the local semantic-preserving direction, suppressing draft acceptance while minimizing semantic drift. Experiments on various speculative decoding systems show that Mistletoe substantially reduces average accepted length $τ$, collapses speedup, and lowers averaged token throughput, while preserving output quality and perplexity. Our work highlights that speculative decoding introduces a mechanism-level attack surface beyond existing output robustness, calling for more robust designs of LLM acceleration systems.", "authors": ["Shuoyang Sun", "Chang Dai", "Hao Fang", "Kuofeng Gao", "Xinhao Zhong", "Yi Sun", "Fan Mo", "Shu-Tao Xia", "Bin Chen"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-13", "url": "https://arxiv.org/abs/2605.14005", "pdf_url": "https://arxiv.org/pdf/2605.14005v2", "arxiv_id": "2605.14005", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "567d698b4cee5f5e90c568426a26905573fa25f2e8e073bce18d902f7d66843c", "sources": ["arxiv", "semantic_scholar"], "title": "PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding", "abstract": "Speculative decoding can significantly accelerate LLM inference, especially given that its cloud-edge collaborative deployment offers cloud workload offloading, offline robustness, and privacy enhancement. However, existing collaborative inference frameworks with speculative decoding are constrained by (i) sequential token generation and communication with low resource utilization, and (ii) inflexible cloud non-autoregressive verification (NAV) triggering that induces premature verification or costly rollbacks. In this paper, we propose PipeSD, an efficient cloud-edge collaborative pipeline inference framework with speculative decoding. PipeSD overlaps token generation and communication by a token-batch pipeline scheduling mechanism optimized by dynamic programming, and improves verification flexibility through a dual-threshold NAV triggering mechanism with a lightweight Bayesian optimization autotuner. We implement PipeSD using llama-cpp-python, PyTorch, and FastAPI, and evaluate it on a real-world cloud-edge testbed with two draft-target model pairs across four scenarios. Results show that PipeSD consistently outperforms state-of-the-art baselines, achieving 1.16x-2.16x speedup and reducing energy consumption by 14.3%-25.3%.", "authors": ["Yunhe Han", "Yunqi Gao", "Bing Hu", "Mahdi Boloursaz Mashhadi", "Yitong Duan", "Pei Xiao", "Yanfeng Zhang"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-13", "url": "https://arxiv.org/abs/2605.13319", "pdf_url": "https://arxiv.org/pdf/2605.13319v3", "arxiv_id": "2605.13319", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "22486e20043601fed8f581f23590b4a874519b61d5542a08d07c0367703c5263", "sources": ["arxiv", "semantic_scholar"], "title": "CATS: Cascaded Adaptive Tree Speculation for Memory-Limited LLM Inference Acceleration", "abstract": "Auto-regressive decoding in Large Language Models (LLMs) is inherently memory-bound: every generation step requires loading the model weights and intermediate results from memory (e.g., High-Bandwidth Memory (HBM) for GPU servers), making throughput bottlenecked by memory bandwidth rather than compute. Speculative decoding addresses this by enabling parallel verification of multiple draft tokens, effectively amortizing the cost of each target-model call. However, existing speculative decoding methods are designed under the assumption that HBM is sufficiently large to hold both the target model and an auxiliary draft model simultaneously -- an assumption that breaks down on memory-constrained devices such as edge platforms with limited DRAM. We analyze the inference bottleneck in this memory-limited regime and propose CATS, a self-speculative decoding framework that conducts cascaded verification and correction based on the memory budget and parameter offloading patterns on memory-limited devices. This design maximizes token acceptance rate and end-to-end speedup while keeping the peak memory footprint on the device equal to that of the target model alone. We evaluate CATS on different models across five benchmarks on real edge devices. CATS can achieve a wall-clock speedup of up to 5.08x with no degradation in generation quality, outperforming the SOTA method by up to 1.45x under edge memory constraints.", "authors": ["Yuning Han", "Yangchenchen Jin", "Dylan Zhao", "Jingwei Sun"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-11", "url": "https://arxiv.org/abs/2605.11186", "pdf_url": "https://arxiv.org/pdf/2605.11186v1", "arxiv_id": "2605.11186", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "34f0b12c866caa1d6c01d3a9accad01f131880f70807a9fffc022f33b45913de", "sources": ["arxiv", "semantic_scholar"], "title": "GELATO: Generative Entropy- and Lyapunov-based Adaptive Token Offloading for Device-Edge Speculative LLM Inference", "abstract": "The recent growth of on-device Large Language Model (LLM) inference has driven significant interest in device-edge collaborative LLM inference. As a promising architecture, Speculative Decoding (SD) is increasingly adopted where a lightweight draft model rapidly generates candidate tokens to be verified by a powerful target model. However, a fundamental challenge lies in achieving per-token resource scheduling to effectively adapt SD paradigm to resource-constrained edge environment. This paper proposes a Generative Entropy- and Lyapunov-based Adaptive Token Offloading framework, named GELATO, to maximize decoding throughput under energy constraints in a device-edge collaborative SD system. Specifically, an outer drift-plus-penalty loop makes online decisions to establish a reference drafting budget, managing long-term energy-throughput trade-off. Further, a nested entropy-driven generation mechanism executes early exiting to adapt to per-token dynamic generative uncertainty. Theoretical analysis establishes a rigorous performance bound on long-term throughput for GELATO. Extensive evaluations demonstrate that GELATO achieves a globally optimal tradeoff, outperforming state-of-the-art distributed SD architectures by 64.98% in token throughput and reducing energy consumption by 47.47% under resource-constrained environments, while preserving LLM decoding quality.", "authors": ["Zengzipeng Tang", "Yuxuan Sun", "Wei Chen", "Jianwen Ding", "Bo Ai"], "categories": ["cs.NI", "cs.DC", "cs.IT", "cs.LG"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-05-11", "url": "https://arxiv.org/abs/2605.10124", "pdf_url": "https://arxiv.org/pdf/2605.10124v1", "arxiv_id": "2605.10124", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "3ad664e3bced318acf3f428ad5669080dec5a7484cdcf0cd3b181f66a9276b2e", "sources": ["arxiv", "semantic_scholar"], "title": "Attention Drift: What Autoregressive Speculative Decoding Models Learn", "abstract": "Speculative decoding accelerates LLM inference by drafting future tokens with a small model, but drafter models degrade sharply under template perturbation and long-context inputs. We identify a previously-unreported phenomenon we call \\textbf{attention drift}: as the drafter generates successive tokens within a speculation chain, attention progressively moves from the prompt onto its own recently-generated tokens. We observe this across both \\emph{EAGLE3} drafters and \\emph{MTP heads}, suggesting drift is a property of drafter designs. We trace this to the un-normalized residual path between chain steps: the drafter's hidden state magnitude grows monotonically with chain depth, which exhibits dynamics consistent with additional pre-norm transformer layers stacked on the target rather than as a standalone autoregressive predictor. In order to limit the growth, we propose two architectural changes: Post-norm on the drafter hidden states and per-hidden-state RMSNorm after capturing target hidden states. Our interventions improve acceptance length over the current leading model, pre-norm EAGLE3, by up to $2\\times$ under template perturbation, $1.18\\times$ on long-context tasks, and $1.10\\times$ on seven standard benchmarks spanning multi-turn chat, math, and coding. Our changes also allow shorter train-time-test depths to generalize over longer drafting sequences.", "authors": ["Doğaç Eldenk", "Payal Mohapatra", "Yigitcan Comlek", "Kaan Oktay", "Hongyang Zhang", "Stephen Xia"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-11", "url": "https://arxiv.org/abs/2605.09992", "pdf_url": "https://arxiv.org/pdf/2605.09992v1", "arxiv_id": "2605.09992", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "7d16dde4c1901d12e625e7b87a790624aea83eb9fe3c253ce61dd78f0c4c1b5b", "sources": ["arxiv", "semantic_scholar"], "title": "SlimSpec: Low-Rank Draft LM-Head for Accelerated Speculative Decoding", "abstract": "Speculative decoding speeds up autoregressive generation in Large Language Models (LLMs) through a two-step procedure, where a lightweight draft model proposes tokens which the target model then verifies in a single forward pass. Although the drafter network is small in modern architectures, its LM-head still performs projection to a large vocabulary, becoming one of the major computational bottlenecks. In prior work this issue has been predominantly addressed via static or dynamic vocabulary truncation. Yet mitigating the bottleneck, these methods bring in extra complexity, such as special vocabulary curation, sophisticated inference-time logic or modifications of the training setup. In this paper, we propose SlimSpec, a low-rank parameterization of the drafter's LM-head that compresses the inner representation rather than the output, preserving full vocabulary support. We evaluate our method with EAGLE-3 drafter across three target models and diverse benchmarks in both latency- and throughput-bound inference regimes. SlimSpec achieves $4\\text{-}5\\times$ acceleration over the standard LM-head architecture while maintaining a competitive acceptance length, surpassing existing methods by up to $8\\text{-}9\\%$ of the end-to-end speedup. Our method requires minimal adjustments of training and inference pipelines. Combined with the aforementioned speedup improvements, it makes SlimSpec a strong alternative across wide variety of draft LM-head architectures.", "authors": ["Anton Plaksin", "Sergei Krutikov", "Sergei Skvortsov", "Alexander Samarin"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-11", "url": "https://arxiv.org/abs/2605.10453", "pdf_url": "https://arxiv.org/pdf/2605.10453v1", "arxiv_id": "2605.10453", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "dd6635734fc61cb56b5334026e94ff9e387b38ce2d4f3b026ba09b3b91b33d20", "sources": ["arxiv", "semantic_scholar"], "title": "31.1 A 14.08-to-135.69Token/s ReRAM-on-Logic Stacked Outlier-Free Large-Language-Model Accelerator with Block-Clustered Weight-Compression and Adaptive Parallel-Speculative-Decoding", "abstract": "This work presents a 55nm speculative decoding-based LLM accelerator with bumping-based face-to-face ReRAM-on-logic stacking technology. It features a local rotation unit for outlier-free low-bit quantization, a stacking-aware PNM architecture co-designed with blockwise vector quantization to reduce weight EMA overheads, and an adaptive parallel speculative decoding scheme with an out-of-order scheduler for high resource and bandwidth utilization. Our chip achieves 14.08-to-135.69token/s and 4.46-to-7.17x speedup over vanilla speculative decoding.", "authors": ["Pingcheng Dong", "Yonghao Tan", "Xuejiao Liu", "Peng Luo", "Yu Liu", "Di Pang", "Songchen Ma", "Xijie Huang", "Shih-Yang Liu", "Dong Zhang", "Zhichao Lu", "Luhong Liang", "Chi-Ying Tsui", "Fengbin Tu", "Liang Zhao", "Kwang-Ting Cheng"], "categories": ["cs.AR"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-10", "url": "https://arxiv.org/abs/2605.09375", "pdf_url": "https://arxiv.org/pdf/2605.09375v1", "arxiv_id": "2605.09375", "doi": "10.1109/ISSCC49663.2026.11409211", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE International Solid-State Circuits Conference", "quality_score": 0.55}
{"id": "a29542d6f8c8ab3c35c60194a6cea3711fde8a615c571236a70ceb258ee53963", "sources": ["arxiv", "semantic_scholar"], "title": "Test-Time Speculation", "abstract": "Speculative decoding accelerates LLM inference by using a fast draft model to generate tokens and a more accurate target model to verify them. Its performance depends on the $\\textit{acceptance length}$, or number of draft tokens accepted by the target. Our studies show that the acceptance length of even state-of-the-art speculators, like DFlash, EAGLE-3 and PARD degrade with generation length, reaching values close to 1 (i.e. no speedup) within just a few thousand output tokens, making speculators ineffective for long-response tasks. Acceptance lengths decline because most speculators are trained offline on short sequences, but are forced to match the target model on much longer outputs at inference, well beyond their training distribution. To address this issue, we propose $\\textit{Test-Time Speculation (TTS)}$, an online distillation approach that continuously adapts the speculator at test-time. TTS leverages the key insight that the token verification step already invokes the target model for each draft token, providing the training signal needed to adapt the draft at no additional cost. Treating the draft as the student and the target as a teacher, TTS adjusts the draft over several speculation rounds, with each update improving the draft's accuracy as generation proceeds. Our results across multiple models from the Qwen-3, Qwen-3.5, and Llama3.1 families show that TTS improves acceptance lengths over state-of-the-art speculators by up to $72\\%$ and $41\\%$ on average, with the benefits scaling with increased generation lengths.", "authors": ["Avinash Kumar", "Sujay Sanghavi", "Poulami Das"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-10", "url": "https://arxiv.org/abs/2605.09329", "pdf_url": "https://arxiv.org/pdf/2605.09329v2", "arxiv_id": "2605.09329", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "5c3811110cc4312de6e3cb80ec3b1ef0fd644caa70282544d91025ce5abbf671", "sources": ["arxiv", "semantic_scholar"], "title": "PARD-2: Target-Aligned Parallel Draft Model for Dual-Mode Speculative Decoding", "abstract": "Speculative decoding accelerates Large Language Models (LLMs) inference by using a lightweight draft model to propose candidate tokens that are verified in parallel by the target model. However, existing draft model training objectives are not directly aligned with the inference-time goal of maximizing consecutive token acceptance. To address this issue, we reformulate the draft model optimization objective, shifting the focus from token prediction accuracy to the overall acceptance length. In this paper, we build upon PARD to propose PARD-2, a dual-mode speculative decoding framework with Confidence-Adaptive Token (CAT) optimization. This approach adaptively reweights each token to better align with the verification process. Notably, PARD-2 enables a single draft model to support both target-dependent and target-independent modes. Experiments across diverse models and tasks demonstrate that PARD-2 achieves up to 6.94$\\times$ lossless acceleration, surpassing EAGLE-3 by 1.9$\\times$ and PARD by 1.3$\\times$ on Llama3.1-8B. Our code is available at https://github.com/AMD-AGI/PARD.", "authors": ["Zihao An", "Taichi Liu", "Ziqiong Liu", "Dong Li", "Ruofeng Liu", "Emad Barsoum"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-09", "url": "https://arxiv.org/abs/2605.08632", "pdf_url": "https://arxiv.org/pdf/2605.08632v1", "arxiv_id": "2605.08632", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/AMD-AGI/PARD", "venue": null, "quality_score": 0.65}
{"id": "52c38a8cb1d85c08fb561f42484fd3d3c0e98bc7d34f949d45dfc8b7f51eefb4", "sources": ["arxiv", "semantic_scholar"], "title": "CASCADE: Context-Aware Relaxation for Speculative Image Decoding", "abstract": "Autoregressive generation is a powerful approach for high-fidelity image synthesis, but it remains computationally demanding and slow even on the most advanced accelerators. While speculative decoding has been explored to mitigate this bottleneck, existing approaches fail to achieve efficiency gains comparable to those observed in text generation. A key limitation is the target model's high uncertainty during image generation, which leads to high draft token rejection rates. In this work, we identify previously overlooked patterns in the target model's behavior that emerge naturally in tree-based speculative decoding. Specifically, we formalize two properties, semantic interchangeability and convergence, arising from the redundancies in the target model's hidden state representations. By capturing these redundancies across the depth and breadth of the predicted token tree, our method identifies principled opportunities for acceptance relaxation without requiring additional training. Additionally, we enhance standalone drafter performance by injecting the redundancy signals from the target model into drafter training with minimal modification. We evaluate our approach across multiple text-to-image models and drafter architectures. Results show that CASCADE achieves state-of-the-art speedups for drafter-based speculative decoding, with up to 3.6x acceleration, while maintaining image quality and text-prompt fidelity.", "authors": ["Selin Yildirim", "Subhajit Dutta Chowdhury", "Mohammad Mahdi Kamani", "Vikram Appia", "Deming Chen"], "categories": ["cs.CV", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-08", "url": "https://arxiv.org/abs/2605.07230", "pdf_url": "https://arxiv.org/pdf/2605.07230v1", "arxiv_id": "2605.07230", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "556e0fb86d22b2ce102091cf9d3573bfcee44200d3faa76548bd3d8e30d348c6", "sources": ["arxiv", "semantic_scholar"], "title": "Future Validity is the Missing Statistic: From Impossibility to $Φ$-Estimation for Grammar-Faithful Speculative Decoding", "abstract": "Grammar-constrained generation is often combined with local vocabulary masking and speculative decoding, but the resulting sampling law is not the grammar-conditional distribution users usually intend. We show that any speculative decoder with local mask access, Leviathan rejection, and rollback soundness samples from the locally projected distribution $μ^{\\mathrm{proj}}$ rather than the grammar-conditional distribution $μ^\\star$. This extends the GAD impossibility result to speculative decoding; on Dyck grammars with Qwen3-8B, the total-variation gap can reach 0.996. We identify the future-validity function $Φ_t(y)=\\Pr_p[\\mathrm{valid\\ completion}\\mid y]$ as the missing correction statistic. The target distribution is a Doob transform of the base model with $h=Φ$, while local masking corresponds to setting $h$ to one. With exact $Φ$, our oracle decoder FVO-Spec samples exactly from $μ^\\star$; with approximate $Φ$, we bound the resulting total-variation error. Because exact future validity is hard for general context-free grammars, we evaluate estimator hierarchies on tractable Dyck and finite JSON languages. OneStep reduces Dyck TV by 14% with under 1% throughput overhead, exact dynamic programming reduces it by 97%, and finite-language correction closes JSON gaps to numerical precision. All fidelity claims are scoped to enumerable grammars and token tries.", "authors": ["Wenhua Nie", "Zijie Meng", "Kun Zou", "Zheng Lin", "Ziwei Li", "Haoran Zheng", "Jyh-Shing Roger Jang", "Hao Zhang"], "categories": ["cs.LG", "cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-05-08", "url": "https://arxiv.org/abs/2605.07698", "pdf_url": "https://arxiv.org/pdf/2605.07698v1", "arxiv_id": "2605.07698", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "61030784c8e3e6e22215ccee3ac44b6a840d2c81ff8349b6c7d4186a72acc723", "sources": ["arxiv", "semantic_scholar"], "title": "SpecBlock: Block-Iterative Speculative Decoding with Dynamic Tree Drafting", "abstract": "Speculative decoding accelerates LLM inference by drafting a tree of candidate continuations and verifying it in one target forward. Existing drafters fall into two camps with opposite weaknesses. Autoregressive drafters such as EAGLE-3 preserve dependence along each draft path but call the drafter once per tree depth, making drafting a non-trivial share of per-iteration latency. Parallel drafters cut drafter calls by predicting multiple future positions in one forward, but each position is predicted without seeing the others, producing paths the verifier rejects. In this paper, we propose SpecBlock, a block-iterative drafter that combines path dependence with cheap drafting. Each drafter forward produces K dependent positions and we call this a block. The draft tree grows through repeated block expansions. Two mechanisms explicitly carry path dependence to keep later draft positions accurate. Within each block, a layer-wise shift carries the previous position's hidden state into every decoder layer. Across blocks, each new block can start from any position of the previous block, inheriting its hidden state to extend the path. To spend verifier budget where acceptance is likely, a co-trained rank head replaces the fixed top-k tree by allocating per-position branching during drafting. To avoid training the drafter on prefixes it never produces at inference, a valid-prefix mask drops the loss at later positions once an earlier one is wrong. Beyond static drafting, a cost-aware bandit at deployment uses free verifier feedback to update the drafter selectively, only when the expected throughput gain exceeds the update cost. Experiments show that SpecBlock improves mean speedup by 8-13% over EAGLE-3 at 44-52% of its drafting cost, and cost-aware adaptation extends this lead to 11-19%.", "authors": ["Weijie Shi", "Qiang Xu", "Fan Deng", "Yaguang Wu", "Jiarun Liu", "Yehong Xu", "Hao Chen", "Jia Zhu", "Jiajie Xu", "Xiangjun Huang", "Jian Yang", "Xiaofang Zhou"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-08", "url": "https://arxiv.org/abs/2605.07243", "pdf_url": "https://arxiv.org/pdf/2605.07243v2", "arxiv_id": "2605.07243", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "3a6bdbeafb8b7fb1375eeb609731f435e26de11aca2fc70c7435b2b22499b33d", "sources": ["arxiv", "semantic_scholar"], "title": "Parallel Prefix Verification for Speculative Generation", "abstract": "We introduce PARSE (PArallel pRefix Speculative Engine), a speculative generation framework that accelerates large language model (LLM) inference by parallelizing prefix verification on a semantic level. Existing speculative decoding methods are fundamentally limited by token-level equivalence: the target model must verify each token, leading to short acceptance lengths and modest speedups. Moving to semantic or segment-level verification can substantially increase acceptance granularity, but prior approaches rely on sequential verification, introducing significant overhead and limiting practical gains. PARSE introduces parallel prefix verification, enabling semantic-level verification without sequential checks. Given a full draft from a draft model, the target model evaluates correctness across multiple prefixes in a single forward pass using a custom attention mask, directly identifying the maximal valid prefix. This eliminates sequential segment verification, and makes verification compute-efficient. PARSE is orthogonal to token-level speculative decoding and can be composed with it for additional gains. Across models and benchmarks, PARSE delivers $1.25\\times$ to $4.3\\times$ throughput gain over the target model, and $1.6\\times$ to $4.5\\times$ when composed with EAGLE-3, all with negligible accuracy degradation. This demonstrates parallel prefix verification as an effective, general approach to accelerating LLM inference.", "authors": ["Yuncheng Yao", "Yuxuan Xia", "Shengjie Wang", "Danyang Zhuo"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-05", "url": "https://arxiv.org/abs/2605.04263", "pdf_url": "https://arxiv.org/pdf/2605.04263v1", "arxiv_id": "2605.04263", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "d81d6e1d37e9f015a370d0543a05dcc40c1e2b701bf4eb7930a05f67e8c10082", "sources": ["arxiv", "semantic_scholar"], "title": "SpecKV: Adaptive Speculative Decoding with Compression-Aware Gamma Selection", "abstract": "Speculative decoding accelerates large language model (LLM) inference by using a small draft model to propose candidate tokens that a larger target model verifies. A critical hyperparameter in this process is the speculation length $γ$, which determines how many tokens the draft model proposes per step. Nearly all existing systems use a fixed $γ$ (typically 4), yet empirical evidence suggests that the optimal value varies across task types and, crucially, depends on the compression level applied to the target model. In this paper, we present SpecKV, a lightweight adaptive controller that selects $γ$ per speculation step using signals extracted from the draft model itself. We profile speculative decoding across 4 task categories, 4 speculation lengths, and 3 compression levels (FP16, INT8, NF4), collecting 5,112 step-level records with per-step acceptance rates, draft entropy, and draft confidence. We demonstrate that the optimal $γ$ shifts across compression regimes and that draft model confidence and entropy are strong predictors of acceptance rate (correlation $\\approx$ 0.56). SpecKV uses a small MLP trained on these signals to maximize expected tokens per speculation step, achieving a 56.0% improvement over the fixed-$γ=4$ baseline with only 0.34 ms overhead per decision (<0.5% of step time). The improvement is statistically significant (p < 0.001, paired bootstrap test). We release all profiling data, trained models, and notebooks as open-source artifacts.", "authors": ["Shikhar Shukla"], "categories": ["cs.LG", "cs.AI", "cs.CL", "cs.DC", "eess.SY"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2026-05-04", "url": "https://arxiv.org/abs/2605.02888", "pdf_url": "https://arxiv.org/pdf/2605.02888v2", "arxiv_id": "2605.02888", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Amorfati123/SpecKV", "venue": null, "quality_score": 0.65}
{"id": "7eb7a5f9140ce7d603710d1b02e2f9077cf214e4dcad0197f32c9d296d186c6c", "sources": ["arxiv", "semantic_scholar"], "title": "SPECTRE: Hybrid Ordinary-Parallel Speculative Serving for Resource-Efficient LLM Inference", "abstract": "LLM serving platforms are increasingly deployed as multi-model cloud systems, where user demand is often long-tailed: a few popular large models receive most requests, while many smaller tail models remain underutilized. We propose \\textbf{SPECTRE} (Parallel \\textbf{SPEC}ulative Decoding with a Multi-\\textbf{T}enant \\textbf{RE}mote Drafter), a serving framework that reuses underutilized tail-model services as remote drafters for heavily loaded large-model services through speculative decoding. SPECTRE enables draft generation and target-side verification to run in parallel, and makes such parallelism effective through three techniques: a hybrid ordinary-parallel speculative decoding strategy guided by a threshold derived from throughput analysis, speculative priority scheduling to preserve draft--target overlap under multi-tenant traffic, and draft-side prompt compression to reduce draft latency. We implement SPECTRE in \\texttt{SGLang} and evaluate it across multiple draft--target model pairs, reasoning benchmarks, real-world long-context workloads, and a wide range of batch sizes. Results show that SPECTRE consistently improves large-model serving throughput while causing only minor interference to the native workloads of tail-model services. In large-model deployments, including Qwen3-235B-A22B with TP=8, SPECTRE achieves up to \\textbf{2.28$\\times$ speedup} over autoregressive decoding and up to an additional \\textbf{66\\% relative improvement} over the strongest speculative decoding baselines. Talk is cheap, we show you the code: https://github.com/sgl-project/sglang/pull/22272.", "authors": ["Jincheng Xie", "Yawen Ling", "Qi Xiao", "Feiyu Zhang", "Zhongyi Huang", "Wen Hu", "Yu Zheng"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-04", "url": "https://arxiv.org/abs/2605.08151", "pdf_url": "https://arxiv.org/pdf/2605.08151v2", "arxiv_id": "2605.08151", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/sgl-project/sglang/pull/22272", "venue": null, "quality_score": 0.65}
{"id": "f7c2ece51dedc23b887e2321514ae8b6fac663c98779cb2677d30a8ecdda52ed", "sources": ["arxiv", "semantic_scholar"], "title": "CoVSpec: Efficient Device-Edge Co-Inference for Vision-Language Models via Speculative Decoding", "abstract": "Vision-language models (VLMs) have demonstrated strong capabilities in multimodal perception and reasoning. However, deploying large VLMs on mobile devices remains challenging due to their substantial computational and memory demands. A practical alternative is device-edge co-inference, where a lightweight draft VLM on the mobile device collaborates with a larger target VLM on the edge server via speculative decoding. Nevertheless, directly extending speculative decoding to VLMs suffers from severe inefficiency due to excessive visual-token computation and high communication overhead. To address these challenges, we propose CoVSpec, an efficient collaborative speculative decoding framework for VLM inference. Specifically, we first develop a training-free visual token reduction framework that prunes redundant visual tokens on the mobile device by jointly considering query relevance, token activity, and low-rank dependency. Moreover, we design an adaptive drafting strategy that dynamically adjusts both the verification frequency and the draft length. In addition, we introduce a parallel branching mechanism with decoupled verification-correction to improve draft-side utilization during target-side verification and reduce correction-related transmission overhead. Experiments on multiple benchmarks show that CoVSpec achieves up to 2.21x higher throughput than target-only inference and reduces communication overhead by more than 96% compared with baselines, without compromising task accuracy.", "authors": ["Yuanyuan Jia", "Shunpu Tang", "Qianqian Yang"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-04", "url": "https://arxiv.org/abs/2605.02218", "pdf_url": "https://arxiv.org/pdf/2605.02218v1", "arxiv_id": "2605.02218", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "acc73fea3b2b4266f59431d60739e647587f4f5c55623edc5a55aa7930becbff", "sources": ["arxiv", "semantic_scholar"], "title": "Component-Aware Self-Speculative Decoding in Hybrid Language Models", "abstract": "Speculative decoding accelerates autoregressive inference by drafting candidate tokens with a fast model and verifying them in parallel with the target. Self-speculative methods avoid the need for an external drafter but have been studied exclusively in homogeneous Transformer architectures. We introduce component-aware self-speculative decoding, the first method to exploit the internal architectural heterogeneity of hybrid language models, isolating the SSM/linear-attention subgraph as a zero-cost internal draft. We evaluate this on two architecturally distinct hybrid families: Falcon-H1 (parallel: Mamba-2 + attention per layer) and Qwen3.5 (sequential: interleaved linear and attention layers), with a pure Transformer control (Qwen2.5). Parallel hybrids achieve acceptance rates of alpha = 0.68 at draft length k=2 under greedy decoding, while sequential hybrids yield only alpha = 0.038 -- an 18x gap attributable to how each architecture integrates its components. The property is scale-invariant: Falcon-H1 at 3B reproduces the rates observed at 0.5B. We further show that perplexity degradation from a companion ablation study predicts speculative viability without running speculative decoding: a 3.15x ratio (Falcon) maps to alpha = 0.37 at k=4, while 81.96x (Qwen) maps to alpha = 0.019. For sequential hybrids, generic LayerSkip achieves 12x higher acceptance rates than the component-aware strategy. The composition pattern of hybrid models -- not merely the presence of alternative components -- determines whether component-level self-speculation is viable.", "authors": ["Hector Borobia", "Elies Seguí-Mas", "Guillermina Tormo-Carbó"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-01", "url": "https://arxiv.org/abs/2605.01106", "pdf_url": "https://arxiv.org/pdf/2605.01106v1", "arxiv_id": "2605.01106", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/hecboar/hybrid-speculative-decoding", "venue": null, "quality_score": 0.65}
{"id": "0b98f82ee509f41751bed8e17e73020891c3538fff332e5bb6b4f462d1f04c0b", "sources": ["arxiv", "semantic_scholar"], "title": "Making Every Verified Token Count: Adaptive Verification for MoE Speculative Decoding", "abstract": "Tree-based speculative decoding accelerates autoregressive generation by verifying multiple draft candidates in parallel, but this advantage weakens for sparse Mixture-of-Experts (MoE) models. As the draft tree grows, different branches activate different experts, expanding the union of activated experts and substantially increasing target-side verification cost. We propose EVICT, a training-free, hyperparameter-free, and lossless adaptive verification method for MoE speculative decoding. EVICT makes every verified token count by truncating the draft tree before target verification and retaining only the cost-effective prefix. It leverages fine-grained drafter signals to estimate candidate benefit, combines them with offline-profiled verification cost, and remains highly compatible with the high-performance graph-based serving framework SGLang. Extensive experiments on diverse MoE backbones and benchmarks show that EVICT achieves up to 2.35x speedup over autoregressive decoding and an average 1.21x speedup over the state-of-the-art baseline EAGLE-3, while significantly reducing unnecessary expert activations during verification.", "authors": ["Lehan Pan", "Ziyang Tao", "Ruoyu Pang", "Xiao Wang", "Jianjun Zhao", "Yanyong Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-05-01", "url": "https://arxiv.org/abs/2605.00342", "pdf_url": "https://arxiv.org/pdf/2605.00342v1", "arxiv_id": "2605.00342", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.35}
{"id": "ded1c9f9be141eff511a71ddb73fe8bd9c1aa84402e568afd7a02240e2d62440", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating RL Post-Training Rollouts via System-Integrated Speculative Decoding", "abstract": "RL post-training of frontier language models is increasingly bottlenecked by autoregressive rollout generation, making rollout acceleration a central systems challenge. Many existing efficiency methods improve throughput by changing the rollout or optimization regime, for example, through off-policy execution, replay, or lower-precision generation. We study speculative decoding as a lossless acceleration primitive for RL rollouts that preserves the target model's output distribution. We implement speculative decoding in NeMo-RL with a vLLM backend, supporting both synchronous and asynchronous pipelines and enabling speculation during RL rollouts. This benefit is realizable across speculation mechanisms, such as pretrained MTP heads, small external draft models or even techniques such as Eagle3, which are traditionally applied after RL phase. This yields a deployment path for state-of-the-art speculative decoding inside RL training. In a reasoning post-training workload at 8B scale under synchronous RL, speculative decoding improves rollout throughput by 1.8x. Using a high-fidelity performance simulator, we project that combining speculative decoding with asynchronous RL yields up to 2.5x end-to-end training speedup at 235B scale.", "authors": ["Hayate Iso", "Tiyasa Mitra", "Sudipta Mondal", "Rasoul Shafipour", "Venmugil Elango", "Terry Kong", "Yuki Huang", "Seonjin Na", "Izzy Putterman", "Benjamin Chislett", "Maor Ashkenazi", "Joseph Guman", "Gerald Shen", "Tugrul Konuk", "Ashwath Aithal", "Ritika Borkar", "Ran Zilberstein", "Bita Rouhani"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-29", "url": "https://arxiv.org/abs/2604.26779", "pdf_url": "https://arxiv.org/pdf/2604.26779v1", "arxiv_id": "2604.26779", "doi": "10.48550/arXiv.2604.26779", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.55}
{"id": "460a2479e928c9b61782fb98c048b4f071996aab29820ffca4d2d6313b8e4995", "sources": ["arxiv", "semantic_scholar"], "title": "When Hidden States Drift: Can KV Caches Rescue Long-Range Speculative Decoding?", "abstract": "Speculative decoding accelerates LLM inference, but SOTA hidden-state-based drafters suffer from long-range decay: draft accuracy degrades as the speculative step increases. Existing work attributes this decay to train-inference mismatch and proposes test-time training (TTT) as a remedy, yet we observe that long-range decay persists even in TTT-trained drafters. We revisit long-range decay from the perspective of context information preservation. In hidden-state reuse, we argue the target hidden state acts as a biased context compression: it aggregates historical token information according to the attention query at the current position, yielding a compact representation optimized for immediate next-token prediction. This compression can suppress information less relevant to the current query but important for later speculative steps. In contrast, the target model's KV cache serves as an explicit context, retaining the complete set of token-wise KV representations. We therefore posit the KV-Reuse Hypothesis: allowing the draft model to reuse the target KV cache can provide richer signals for long-horizon drafting. To test this hypothesis, we introduce KVShot, a diagnostic framework that compares three reuse paradigms: hidden-only, KV-only, and hybrid. Extensive evaluations on Qwen3-8B show that KV-Reuse improves long-range acceptance, although end-to-end speedups remain marginal under current training pipelines. Our analysis identifies two key structural bottlenecks: shallow drafters struggle to estimate target queries accurately, and draft-side KV projections receive sparse gradient signals. These findings suggest that realizing the full potential of KV-aware decoding requires moving beyond TTT toward block-wise training paradigms. By exposing these bottlenecks, KVShot provides a foundational diagnostic testbed and a clear roadmap for designing next-generation inference architectures.", "authors": ["Tianyu Liu", "Yuhao Shen", "Xinyi Hu", "Baolin Zhang", "Hengxin Zhang", "Jun Dai", "Jun Zhang", "Shuang Ge", "Lei Chen", "Yue Li", "MingCheng Wan"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-29", "url": "https://arxiv.org/abs/2604.26412", "pdf_url": "https://arxiv.org/pdf/2604.26412v2", "arxiv_id": "2604.26412", "doi": "10.48550/arXiv.2604.26412", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.55}
{"id": "3444375581815c62ca166972a5aeb5e3b90d11b34a941f4804cf9e1815bfe138", "sources": ["arxiv", "semantic_scholar"], "title": "An Empirical Study of Speculative Decoding on Software Engineering Tasks", "abstract": "Large Language Models (LLMs) have become widely used for Software Engineering (SE) tasks, spanning from function-level code generation to complex repository-level workflows. However, the high latency of autoregressive inference remains a significant bottleneck, hindering their deployment in interactive environments. While Speculative Decoding (SD) offers a promising technique for lossless acceleration, prior research on long-context repository-level tasks and complex agentic interactions remains limited. To bridge this gap, we present the first systematic empirical study to evaluate the effectiveness of SD in SE tasks. We systematically benchmark a comprehensive spectrum of strategies, encompassing both model-based and model-free methods, across representative generation, editing, and repair scenarios. Our empirical results indicate that SD demonstrates clear potential for accelerating inference, particularly for smaller models that achieve higher speedups than those of their larger counterparts. We find that the effectiveness of SD methods varies across different task scenarios. Model-based approaches are well-suited for code generation, whereas model-free methods are better adapted to repository-level repair and editing scenarios. Furthermore, we observe that the repetitiveness of SE tasks improves the performance of model-free methods. In contrast to natural language tasks, the higher predictability of SE tasks allows for more aggressive hyperparameters. Our findings are summarized as guidelines to help increase inference efficiency for SE scenarios.", "authors": ["Yijia Li", "Junkai Chen", "Xing Hu", "Xin Xia"], "categories": ["cs.SE"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-29", "url": "https://arxiv.org/abs/2604.26469", "pdf_url": "https://arxiv.org/pdf/2604.26469v3", "arxiv_id": "2604.26469", "doi": "10.48550/arXiv.2604.26469", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.55}
{"id": "b762005c1bb065e44f1f3b490d7b31badc0882f1c0e7a1ccf1733c659b6b3b45", "sources": ["arxiv", "semantic_scholar"], "title": "SpecFed: Accelerating Federated LLM Inference with Speculative Decoding and Compressed Transmission", "abstract": "Federated inference enhances LLM performance in edge computing through weighted averaging of distributed model predictions. However, autoregressive LLM inference requires frequent full-model forward passes across workers, severely limiting decoding throughput. Distributed deployment further aggravates this due to a communication bottleneck: each worker must transmit full token probability distributions per draft token, dominating end-to-end latency. To address these challenges, we introduce speculative decoding to enable parallel LLM processing and propose a top-K compressed transmission scheme with two server-side reconstruction strategies. We theoretically analyze the robustness of our method in terms of local reconstruction error, aggregation bias, and acceptance-rate bias, and derive corresponding bounds. Experiments demonstrate that our scheme achieves high generation fidelity while significantly reducing communication overhead.", "authors": ["Ce Zheng", "Xinghan Wang", "Jiahong Ning", "Yuxuan Shi", "Ning Huang", "Tingting Yang"], "categories": ["eess.SP", "cs.DC"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2026-04-28", "url": "https://arxiv.org/abs/2604.25777", "pdf_url": "https://arxiv.org/pdf/2604.25777v1", "arxiv_id": "2604.25777", "doi": "10.48550/arXiv.2604.25777", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.55}
{"id": "e43cf4d320579220f4880a28fbaa065ddf4dd98b3d1a323dcdcc719ba68cd9e4", "sources": ["arxiv", "semantic_scholar"], "title": "AHASD: Asynchronous Heterogeneous Architecture for LLM Adaptive Drafting Speculative Decoding on Mobile Devices", "abstract": "Speculative decoding enhances the inference efficiency of large language models (LLMs) by generating drafts using a small draft language model (DLM) and verifying them in batches with a large target language model (TLM). However, adaptive drafting inference on a mobile single-NPU-PIM system faces idle overhead in traditional operator-level synchronous execution and wasted computation in asynchronous execution due to fluctuations in draft length. This paper introduces AHASD, a task-level asynchronous mobile NPU-PIM heterogeneous architecture for speculative decoding. Notably, AHASD achieves parallel drafting on the PIM and verification on a single NPU through task-level DLM-TLM decoupling and specifically, it incorporates Entropy-History-Aware Drafting Control and Time-Aware Pre-Verification Control to dynamically manage adaptive drafting algorithm execution and pre-verification timing, suppressing invalid drafting based on low-confidence drafts. Additionally, AHASD integrates Attention Algorithm Units and Gated Task Scheduling Units within LPDDR5-PIM to enable attention link localization and sub-microsecond task switching on the PIM side. Experimental results for different LLMs and adaptive drafting algorithms show that AHASD achieves up to 4.2$\\times$ in throughput and 5.6$\\times$ in energy efficiency improvements over a GPU-only baseline, and 1.5$\\times$ in throughput and 1.24$\\times$ in energy efficiency gains over the state-of-the-art GPU+PIM baseline, with hardware overhead below 3% of the DRAM area.", "authors": ["Ma Zirui", "Fan Zhihua", "Li Wenxing", "Wu Haibin", "Zhang Fulin", "Ye Xiaochun", "Li Wenming"], "categories": ["cs.AR", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-28", "url": "https://arxiv.org/abs/2604.25326", "pdf_url": "https://arxiv.org/pdf/2604.25326v3", "arxiv_id": "2604.25326", "doi": "10.1145/3770743.3803965", "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/MAdrid1011/AHASD", "venue": "arXiv.org", "quality_score": 0.85}
{"id": "a809065401a8ed4e20d45eb999e222ac3577319fa1245a226ef27cd9450c2b88", "sources": ["arxiv", "semantic_scholar"], "title": "FASER: Fine-Grained Phase Management for Speculative Decoding in Dynamic LLM Serving", "abstract": "Speculative decoding (SD) is a widely used approach for accelerating decode-heavy LLM inference workloads. While online inference workloads are highly dynamic, existing SD systems are rigid and take a coarse-grained approach to SD management. They typically set the speculative token length for an entire batch and serialize the execution of the draft and verification phases. Consequently, these systems fall short at adapting to volatile online inference traffic. Under low load, they exhibit prolonged latency because the draft phase blocks the verification phase for the entire batch, leaving GPU computing resources underutilized. Conversely, under high load, they waste computation on rejected tokens during the verification phase, overloading GPU resources. We introduce FASER, a novel system that features fine-grained SD phase management. First, FASER minimizes computational waste by dynamically adjusting the speculative length for each request within a continuous batch and by performing early pruning of rejected tokens inside the verification phase. Second, FASER breaks the verification phase into frontiers, or chunks, to overlap them with the draft phase. This overlap is achieved via fine-grained spatial multiplexing with minimal resource interference. Our FASER prototype in vLLM improves throughput by up to 53% and reduces latency by up to 1.92$\\times$ compared to state-of-the-art systems.", "authors": ["Wenyan Chen", "Chengzhi Lu", "Yanying Lin", "Dmitrii Ustiugov"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-22", "url": "https://arxiv.org/abs/2604.20503", "pdf_url": "https://arxiv.org/pdf/2604.20503v1", "arxiv_id": "2604.20503", "doi": "10.48550/arXiv.2604.20503", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.55}
{"id": "dd1acd20e7b27782a72bf65ecaa7f0f3446db53948d30fe37eba90bfa0f505e7", "sources": ["arxiv", "semantic_scholar"], "title": "DiP-SD: Distributed Pipelined Speculative Decoding for Efficient LLM Inference at the Edge", "abstract": "Speculative decoding has emerged as a promising technique for large language model (LLM) inference by accelerating autoregressive decoding via draft-then-verify. This paper studies a new edge scenario with multi-user inference, where draft tokens are generated locally on devices and subsequently offloaded to a centralized edge server for batch verification. The key challenge is to sustain high throughput under coupled decisions of (i) batching and pipeline scheduling and (ii) per user draft token length. We propose DiP-SD, which exploits two complementary parallelism dimensions: device-level distributed drafting and phase-level draft-verify pipelining. We formulate a throughput-maximization objective, defined as the expected number of accepted tokens per unit time, and jointly optimize the number of batches, user-to-batch assignment, and integer draft lengths. To solve the resulting fractional mixed-integer program, DiP-SD scans the batch number and iteratively alternates between an association subproblem and a draft-length subproblem. Numerical results under a Qwen3-1.7B/Qwen3-32B device-edge deployment show that DiP-SD achieves up to 17.89x throughput over autoregressive decoding (AD) and 1.93x over AD with greedy batching.", "authors": ["Yaodan Xu", "Sheng Zhou", "Zhisheng Niu"], "categories": ["cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-04-22", "url": "https://arxiv.org/abs/2604.20919", "pdf_url": "https://arxiv.org/pdf/2604.20919v1", "arxiv_id": "2604.20919", "doi": "10.48550/arXiv.2604.20919", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.55}
{"id": "7449dd94dbbb12be98fd6837c186930f2a831cd1a1aa91815547cff5fea264ee", "sources": ["arxiv", "semantic_scholar"], "title": "LLM-Viterbi: Semantic-Aware Decoding for Convolutional Codes", "abstract": "Traditional wireless communications rely solely on bit-level channel coding for error correction, without exploiting the inherent linguistic structure of the data source. This paper proposes a large language model (LLM) Viterbi decoder that integrates LLM priors into the Viterbi decoding for text transmission over AWGN channels. The proposed decoder maintains multiple candidate paths during the Viterbi decoding and periodically evaluates path reliabilities using a fine-tuned Byte-level T5 (ByT5) language model. By combining channel reliability metrics with semantic probability from the LLM, it outputs the path that maximizes the joint likelihood of channel observations and linguistic coherence. Simulations show that our decoder achieves significant performance gains over conventional Viterbi decoding in terms of both block error rate (BLER) and semantic similarity. For convolutional codes with constraint length 3, it achieves approximately 1.5 dB more coding gain in BLER, with over 50% improvements in semantic similarity. The framework can extend to other structured data sources beyond text.", "authors": ["Zhengtong Li", "Chentao Yue", "Jiafu Hao", "Branka Vucetic", "Yonghui Li"], "categories": ["cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-04-21", "url": "https://arxiv.org/abs/2604.19035", "pdf_url": "https://arxiv.org/pdf/2604.19035v1", "arxiv_id": "2604.19035", "doi": "10.48550/arXiv.2604.19035", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Todd-6/LLM-Viterbi", "venue": "arXiv.org", "quality_score": 0.85}
{"id": "eff37c71bb5300f9d772a02406e08c03f6ddf26ef24cb0c7def66604a9277e81", "sources": ["arxiv", "semantic_scholar"], "title": "WISV: Wireless-Informed Semantic Verification for Distributed Speculative Decoding in Device-Edge LLM Inference", "abstract": "While distributed device-edge speculative decoding enhances resource utilization across heterogeneous nodes, its performance is often bottlenecked by conventional token-level verification strategies. Such rigid alignment leads to excessive rejections, significantly diminishing the accepted sequence length and increasing interaction rounds under fluctuating wireless conditions. In this paper, we propose WISV (Wireless-Informed Semantic Verification), a novel distributed speculative decoding framework that goes beyond strict token-level matching via a channel-aware semantic acceptance policy. WISV integrates a lightweight decision head into the edge-side target LLM to dynamically evaluate speculative tokens by synthesizing high-dimensional hidden representations with instantaneous channel state information (CSI). To optimize the trade-off between verification fidelity and communication overhead, we further design two tailored communication protocols: full-hidden upload and mismatch-first selective-hidden upload. Extensive simulations using a 1B drafter and an 8B target model demonstrate that WISV achieves up to a 60.8% increase in accepted length, a 37.3% reduction in interaction rounds, and a 31.4% improvement in end-to-end latency compared to vanilla speculative decoding across tested settings, while maintaining a negligible task accuracy drop (<1%). Finally, we validate WISV on a hardware testbed comprising an NVIDIA Jetson AGX Orin and an A40-equipped server, confirming its real-world efficacy in accelerating edge-deployed LLM inference.", "authors": ["Zixuan Liu", "Zhiyong Chen", "Nan Xue", "Shengkang Chen", "Jiangchao Yao", "Meixia Tao", "Wenjun Zhang"], "categories": ["cs.IT", "cs.AI"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-04-20", "url": "https://arxiv.org/abs/2604.17701", "pdf_url": "https://arxiv.org/pdf/2604.17701v1", "arxiv_id": "2604.17701", "doi": "10.48550/arXiv.2604.17701", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.55}
{"id": "50311ae68c1790c8d319f811a79a0560cbf138c1c1215fd7696c6e1039c13080", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding for Autoregressive Video Generation", "abstract": "Autoregressive video diffusion is emerging as a promising paradigm for streaming video synthesis, with step distillation serving as the primary means of accelerating inference. Whether speculative decoding, the dominant acceleration strategy for large language models, can be effectively adapted to autoregressive video generation remains an open question, because video blocks are continuous spatiotemporal tensors with no token-level distribution for exact rejection sampling. We introduce SDVG, which brings speculative decoding to block-based autoregressive video diffusion by replacing token verification with an image-quality router. A 1.3B drafter proposes candidate blocks via four denoising steps; each block is VAE-decoded and scored by ImageReward using worst-frame aggregation--taking the minimum per-frame reward to catch single-frame artifacts that averaging would mask. Blocks scoring above a fixed threshold tau are accepted into the 14B target's KV cache; the rest are regenerated by the target. Two additional design choices prove critical: the first block is always force-rejected to anchor scene composition, and tau serves as a single knob that traces a smooth quality-speed Pareto frontier. On 1003 MovieGenVideoBench prompts (832x480), SDVG retains 98.1% of target-only VisionReward quality (0.0773 vs. 0.0788) at a 1.59x speedup with tau=-0.7, and reaches 2.09x at 95.7% quality retention--while consistently outperforming draft-only generation by over +17%. The framework is training-free, requires no architectural changes, and can be seamlessly integrated into existing autoregressive video generation pipelines.", "authors": ["Yuezhou Hu", "Jintao Zhang"], "categories": ["cs.CV", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-19", "url": "https://arxiv.org/abs/2604.17397", "pdf_url": "https://arxiv.org/pdf/2604.17397v1", "arxiv_id": "2604.17397", "doi": "10.48550/arXiv.2604.17397", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5489}
{"id": "f72a41800ee2e901b6e9ea04de7ec99806dd37b891352b6e55e02772b9261cc8", "sources": ["arxiv", "semantic_scholar"], "title": "Faster LLM Inference via Sequential Monte Carlo", "abstract": "Speculative decoding (SD) accelerates language model inference by drafting tokens from a cheap proposal model and verifying them against an expensive target model via rejection sampling. Because rejection truncates the draft block at the first error, throughput degrades when draft and target diverge. Rather than rejecting draft tokens outright, we propose to reweight them. To this end, we introduce sequential Monte Carlo speculative decoding (SMC-SD), which replaces token-level rejection with importance-weighted resampling over a population of draft particles. SMC-SD is a principled approximate inference scheme that trades exactness for additional speed, while preserving theoretical bounds on its per-step approximation error. Because LLM inference is memory bandwidth-bound, the arithmetic needed to draft particles and to score them in parallel comes nearly for free -- SMC-SD uses idle compute to turn verification into a vectorized, fixed-size operation with no rollback. Empirically, SMC-SD achieves 2.36x speed-up over speculative decoding and a 5.2x speed-up over autoregressive decoding, while remaining within 3% of the target model's accuracy on reasoning, instruction-following, and coding benchmarks.", "authors": ["Yahya Emara", "Mauricio Barba da Costa", "Chi-Chih Chang", "Cameron Freer", "Tim Vieira", "Ryan Cotterell", "Mohamed S. Abdelfattah"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-17", "url": "https://arxiv.org/abs/2604.15672", "pdf_url": "https://arxiv.org/pdf/2604.15672v1", "arxiv_id": "2604.15672", "doi": "10.48550/arXiv.2604.15672", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5466}
{"id": "11c4239976d3da1d0ae54c133e70c465a42e83efccb1c40bef43d4595c3a07d9", "sources": ["arxiv", "semantic_scholar"], "title": "EvoSpec: Evolving Speculative Decoding via Real-Time Vocabulary and Parameter Adaptation", "abstract": "Speculative decoding accelerates Large Language Model inference via a draft-then-verify paradigm, yet the output projection layer becomes a bottleneck as vocabulary sizes scale. While existing static pruning methods effectively reduce this overhead, they suffer from precipitous drops in acceptance rate in specialized domains or topic-switching scenarios due to their inability to capture dynamic distribution shifts. To address this, we introduce EvoSpec, a framework that enables real-time evolution of the draft model through dynamic vocabulary and parameter adaptation. Unlike static or purely retrieval-based approaches, EvoSpec employs a context-aware mechanism that retrieves critical long-tail tokens via efficient semantic and statistical indexing. Furthermore, we propose a lightweight online alignment strategy utilizing curriculum learning to continually minimize the distributional gap between the draft and target models. Extensive evaluations across specialized domains (coding, law, and medicine) confirm that EvoSpec overcomes the limitations of static baselines. On EAGLE-3, it achieves a 1.13x speedup in these settings over the state-of-the-art static baseline FR-Spec, with 27\\% lower memory overhead than standard online adaptation.", "authors": ["Shuyu Zhang", "Lingfeng Pan", "Qicheng Wang", "Yaqi Shi", "Yueyang Tan", "Ruyu Yan", "Jiaqi Chen", "Lixing Du", "Lu Wang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-17", "url": "https://arxiv.org/abs/2605.27390", "pdf_url": "https://arxiv.org/pdf/2605.27390v2", "arxiv_id": "2605.27390", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3478}
{"id": "96a8ba45ae2a2e9e3b1e291b0ddb98ee2018cafd2dc910bbb6c3d851d8251b71", "sources": ["arxiv", "semantic_scholar"], "title": "RACER: Retrieval-Augmented Contextual Rapid Speculative Decoding", "abstract": "Autoregressive decoding in Large Language Models (LLMs) generates one token per step, causing high inference latency. Speculative decoding (SD) mitigates this through a guess-and-verify strategy, but existing training-free variants face trade-offs: retrieval-based drafts break when no exact match exists, while logits-based drafts lack structural guidance. We propose $\\textbf{RACER}$ ($\\textbf{R}$etrieval-$\\textbf{A}$ugmented $\\textbf{C}$ont$\\textbf{e}$xtual $\\textbf{R}$apid Speculative Decoding), a lightweight and training-free method that integrates retrieved exact patterns with logit-driven future cues. This unification supplies both reliable anchors and flexible extrapolation, yielding richer speculative drafts. Experiments on Spec-Bench, HumanEval, and MGSM-ZH demonstrate that RACER consistently accelerates inference, achieving more than $2\\times$ speedup over autoregressive decoding, and outperforms prior training-free methods, offering a scalable, plug-and-play solution for efficient LLM decoding. Our source code is available at $\\href{https://github.com/hkr04/RACER}{https://github.com/hkr04/RACER}$.", "authors": ["Zihong Zhang", "Zuchao Li", "Lefei Zhang", "Ping Wang", "Hai Zhao"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-16", "url": "https://arxiv.org/abs/2604.14885", "pdf_url": "https://arxiv.org/pdf/2604.14885v1", "arxiv_id": "2604.14885", "doi": "10.48550/arXiv.2604.14885", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/hkr04/RACER}{https://github.com/hkr04/RACER}$", "venue": "arXiv.org", "quality_score": 0.8429}
{"id": "ff11338d64b751e1e89defdd12f992c998af0a5a52ca22a5f8f535b99839070d", "sources": ["arxiv", "semantic_scholar"], "title": "Acceptance Dynamics Across Cognitive Domains in Speculative Decoding", "abstract": "Speculative decoding accelerates large language model (LLM) inference. It uses a small draft model to propose a tree of future tokens. A larger target model then verifies these tokens in a single batched forward pass. Despite the growing body of work on speculative methods, the degree to which the cognitive characteristics of a task affect acceptance probability remains largely unexplored. We present an empirical study of tree-based speculative decoding acceptance dynamics. Our study spans four well-established NLP benchmark domains: code generation, mathematical reasoning, logical reasoning, and open-ended chat. For this, we use TinyLlama-1.1B as the draft model against Llama-2-7B-Chat-GPTQ as the target. Over 99,768 speculative nodes collected from 200 prompts, we derive per-domain acceptance rates, expected accepted lengths, depth-acceptance profiles, and entropy-acceptance correlations. We find that task type is a stronger predictor of acceptance than tree depth. Furthermore, only the chat domain consistently yields an expected accepted length exceeding 1.0 token per step. We also show that the entropy-acceptance correlation is consistently negative but weak across all domains (rho in [-0.20, -0.15]). Counterintuitively, chat produces the highest entropy yet the highest acceptance rate. We attribute this divergence to the lexical predictability of RLHF-aligned register. These findings have direct implications for domain-aware speculation budgets and draft-model selection strategies. Index Terms--speculative decoding, large language model inference, tree attention, draft model, acceptance probability, LLM efficiency", "authors": ["Saif Mahmoud"], "categories": ["cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-16", "url": "https://arxiv.org/abs/2604.14682", "pdf_url": "https://arxiv.org/pdf/2604.14682v1", "arxiv_id": "2604.14682", "doi": "10.48550/arXiv.2604.14682", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5454}
{"id": "ea1fd3b3a361eec707cee3fbad1b5fe03b0a7edff2022ad3e8586671e11dd7b9", "sources": ["arxiv", "semantic_scholar"], "title": "From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning", "abstract": "Speculative decoding (SD) accelerates large language model inference by allowing a lightweight draft model to propose outputs that a stronger target model verifies. However, its token-centric nature allows erroneous steps to propagate. Prior approaches mitigate this using external reward models, but incur additional latency, computational overhead, and limit generalizability. We propose SpecGuard, a verification-aware speculative decoding framework that performs step-level verification using only model-internal signals. At each step, SpecGuard samples multiple draft candidates and selects the most consistent step, which is then validated using an ensemble of two lightweight model-internal signals: (i) an attention-based grounding score that measures attribution to the input and previously accepted steps, and (ii) a log-probability-based score that captures token-level confidence. These signals jointly determine whether a step is accepted or recomputed using the target, allocating compute selectively. Experiments across a range of reasoning benchmarks show that SpecGuard improves accuracy by 3.6% while reducing latency by ~11%, outperforming both SD and reward-guided SD.", "authors": ["Kiran Purohit", "Ramasuri Narayanam", "Soumyabrata Pal"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-16", "url": "https://arxiv.org/abs/2604.15244", "pdf_url": "https://arxiv.org/pdf/2604.15244v1", "arxiv_id": "2604.15244", "doi": "10.48550/arXiv.2604.15244", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5454}
{"id": "71f962a2fbfbf4fae4f1964e1f136c1654e986a8b5437e30bd606d975f68e17d", "sources": ["arxiv", "semantic_scholar"], "title": "ConfLayers: Adaptive Confidence-based Layer Skipping for Self-Speculative Decoding", "abstract": "Self-speculative decoding is an inference technique for large language models designed to speed up generation without sacrificing output quality. It combines fast, approximate decoding using a compact version of the model as a draft model with selective re-evaluation by the full target model. Some existing methods form the draft model by dynamically learning which layers to skip during inference, effectively creating a smaller subnetwork to speed up computation. However, using heuristic-based approaches to select layers to skip can often be simpler and more effective. In this paper, we propose ConfLayers, a dynamic plug-and-play approach to forming the draft model in self-speculative decoding via confidence-based intermediate layer skipping. The process iteratively computes confidence scores for all layers, selects layers to skip based on an adaptive threshold, evaluates the performance of the resulting set, and updates the best selection until no further improvement is achieved or a maximum number of iterations is reached. This framework avoids the overhead and complexity of training a layer skipping policy and can provide more consistent speed-quality trade-offs while preserving the adaptivity of the draft model to diverse tasks and datasets. The performance evaluation of ConfLayers across different models and datasets shows that our novel approach offers up to 1.4x speedup over vanilla LLM generation.", "authors": ["Walaa Amer", "Uday das", "Fadi Kurdahi"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-16", "url": "https://arxiv.org/abs/2604.14612", "pdf_url": "https://arxiv.org/pdf/2604.14612v1", "arxiv_id": "2604.14612", "doi": "10.48550/arXiv.2604.14612", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5454}
{"id": "e24d78808fe21c52724f192798875a78c89a5d2a9e29f75d5c7214ad3235ec60", "sources": ["arxiv", "semantic_scholar"], "title": "ELMoE-3D: Leveraging Intrinsic Elasticity of MoE for Hybrid-Bonding-Enabled Self-Speculative Decoding in On-Premises Serving", "abstract": "Mixture-of-Experts (MoE) models have become the dominant architecture for large-scale language models, yet on-premises serving remains fundamentally memory-bound as batching turns sparse per-token compute into dense memory activation. Memory-centric architectures (PIM, NMP) improve bandwidth but leave compute underutilized under MoE's low arithmetic intensity at high batch sizes. Speculative decoding (SD) trades idle compute for fewer target invocations, yet verification must load experts even for rejected tokens, severely limiting its benefit in MoE especially at low batch sizes. We propose ELMoE-3D, a hybrid-bonding (HB)-based HW-SW co-designed framework that unifies cache-based acceleration and speculative decoding to offer overall speedup across batch sizes. We identify two intrinsic elasticity axes of MoE-expert and bit-and jointly scale them to construct Elastic Self-Speculative Decoding (Elastic-SD), which serves as both an expert cache and a strongly aligned self-draft model accelerated by high HB bandwidth. Our LSB-augmented bit-sliced architecture exploits inherent redundancy in bit-slice representations to natively support bit-nested execution. On our 3D-stacked hardware, ELMoE-3D achieves an average $6.6\\times$ speedup and $4.4\\times$ energy efficiency gain over naive MoE serving on xPU across batch sizes 1-16, and delivers $2.2\\times$ speedup and $1.4\\times$ energy efficiency gain over the best-performing prior accelerator baseline.", "authors": ["Yuseon Choi", "Jingu Lee", "Jungjun Oh", "Sunjoo Whang", "Byeongcheol Kim", "Minsung Kim", "Hoi-Jun Yoo", "Sangjin Kim"], "categories": ["cs.LG", "cs.AI", "cs.AR", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-16", "url": "https://arxiv.org/abs/2604.14626", "pdf_url": "https://arxiv.org/pdf/2604.14626v2", "arxiv_id": "2604.14626", "doi": "10.48550/arXiv.2604.14626", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5454}
{"id": "5cca20b3d9d46eca9c699f6df570933331a5dd66f22de603d0a045fc28497734", "sources": ["arxiv", "semantic_scholar"], "title": "Calibrated Speculative Decoding: Frequency-Guided Candidate Selection for Efficient Inference", "abstract": "Speculative decoding accelerates autoregressive generation by letting draft tokens bypass full verification, but conventional frameworks suffer from frequent false rejections, particularly when draft models produce semantically correct but lexically divergent outputs. In this paper, we present Calibrated Speculative Decoding (CSD), a training-free framework that recovers valid tokens discarded by standard verification. Guided by the principle of \"Frequency-Guided Candidate Selection and Probability-Guarded Acceptance,\" CSD incorporates two lightweight modules: Online Correction Memory, which aggregates historical rejections to propose recurring divergence patterns as rescue candidates, and Semantic Consistency Gating, which verifies candidate admissibility using probability ratios instead of exact token matching. Our evaluation across diverse large language models demonstrates that CSD outperforms existing methods, achieving a peak throughput speedup of 2.33x. CSD preserves model accuracy across all tasks while further boosting performance on complex reasoning datasets. These results establish CSD as a highly effective, lightweight solution for practical LLM deployments.", "authors": ["Xuwen Zhou", "Fangxin Liu", "Chao Wang", "Xiao Zheng", "Hao Zheng", "Min He", "Li Jiang", "Haibing Guan"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-15", "url": "https://arxiv.org/abs/2604.13634", "pdf_url": "https://arxiv.org/pdf/2604.13634v1", "arxiv_id": "2604.13634", "doi": "10.48550/arXiv.2604.13634", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5443}
{"id": "2f923df658b8fb3996804d613961ca293ca8d83b52e4944244dd8beed1c66cab", "sources": ["arxiv", "semantic_scholar"], "title": "ToolSpec: Accelerating Tool Calling via Schema-Aware and Retrieval-Augmented Speculative Decoding", "abstract": "Tool calling has greatly expanded the practical utility of large language models (LLMs) by enabling them to interact with external applications. As LLM capabilities advance, effective tool use increasingly involves multi-step, multi-turn interactions to solve complex tasks. However, the resulting growth in tool interactions incurs substantial latency, posing a key challenge for real-time LLM serving. Through empirical analysis, we find that tool-calling traces are highly structured, conform to constrained schemas, and often exhibit recurring invocation patterns. Motivated by this, we propose ToolSpec, a schema-aware, retrieval-augmented speculative decoding method for accelerating tool calling. ToolSpec exploits predefined tool schemas to generate accurate drafts, using a finite-state machine to alternate between deterministic schema token filling and speculative generation for variable fields. In addition, ToolSpec retrieves similar historical tool invocations and reuses them as drafts to further improve efficiency. ToolSpec presents a plug-and-play solution that can be seamlessly integrated into existing LLM workflows. Experiments across multiple benchmarks demonstrate that ToolSpec achieves up to a 4.2x speedup, substantially outperforming existing training-free speculative decoding methods.", "authors": ["Heming Xia", "Yongqi Li", "Cunxiao Du", "Mingbo Song", "Wenjie Li"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-15", "url": "https://arxiv.org/abs/2604.13519", "pdf_url": "https://arxiv.org/pdf/2604.13519v2", "arxiv_id": "2604.13519", "doi": "10.48550/arXiv.2604.13519", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5443}
{"id": "8c95dc8000e2c620d3bdf43001fdcbdcacae537b921dc28c99367e816acf72d9", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Speculative Decoding with Block Diffusion Draft Trees", "abstract": "Speculative decoding accelerates autoregressive language models by using a lightweight drafter to propose multiple future tokens, which the target model then verifies in parallel. DFlash shows that a block diffusion drafter can generate an entire draft block in a single forward pass and achieve state-of-the-art speculative decoding performance, outperforming strong autoregressive drafters such as EAGLE-3. Vanilla DFlash, however, still verifies only a single drafted trajectory per round, potentially limiting its acceptance length. We introduce DDTree (Diffusion Draft Tree), a method that constructs a draft tree directly from the per-position distributions of a block diffusion drafter. Under a fixed node budget, DDTree uses a simple best-first heap algorithm to select the continuations that are most likely to match the target model according to a surrogate defined by the draft model's output. The resulting tree is verified efficiently in a single target model forward pass using an ancestor-only attention mask. Because DDTree builds on DFlash, a leading draft model for speculative decoding, these gains place DDTree among the leading approaches to speculative decoding.", "authors": ["Liran Ringel", "Yaniv Romano"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-14", "url": "https://arxiv.org/abs/2604.12989", "pdf_url": "https://arxiv.org/pdf/2604.12989v1", "arxiv_id": "2604.12989", "doi": "10.48550/arXiv.2604.12989", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5431}
{"id": "c029f2bf77b0863e78c5a12a5c361abf01587f2cc84cb92e9ef4b3bb3336dfcb", "sources": ["arxiv", "semantic_scholar"], "title": "SENSE: Semantic Embedding Navigation with Soft-gated Evaluation for Retrieval-based Speculative Decoding", "abstract": "Speculative Decoding (SD) accelerates Large Language Model (LLM) inference by employing a lightweight draft model to propose candidate tokens, which are verified in parallel by the target model, without compromising generation quality. While Retrieval-based Speculative Decoding (RSD) is favored for its plug-and-play versatility, its potential is impeded by rigid lexical dependencies, rendering both retrieval and verification brittle to surface-level variations. To address this, we propose SENSE (Semantic Embedding Navigation with Soft-gated Evaluation). By anchoring retrieval on the hidden states of the target model, SENSE establishes robust semantic alignment, which empowers the Soft-gated Evaluation module to validate semantic equivalence rather than surface forms. To ensure rigorous benchmarking, we deconstruct existing methods into atomic primitives within a unified framework, facilitating granular, component-level comparison. Extensive experiments across diverse domains demonstrate that SENSE outperforms multiple baselines on the LLaMA and Qwen families, attaining up to 4.09 mean acceptance length and 3.26x speedup, while preserving generation quality. Our code will be released upon publication.", "authors": ["Shaowen Chen", "Zhicheng Liao", "Hongwei Wang"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-14", "url": "https://arxiv.org/abs/2606.00021", "pdf_url": "https://arxiv.org/pdf/2606.00021v1", "arxiv_id": "2606.00021", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3456}
{"id": "e4b7bad1d0e5c499d78182cd4e54e27ee1d87b9c954936d69b2bb8bc3fb76c74", "sources": ["arxiv", "semantic_scholar"], "title": "SOLARIS: Speculative Offloading of Latent-bAsed Representation for Inference Scaling", "abstract": "Recent advances in recommendation scaling laws have led to foundation models of unprecedented complexity. While these models offer superior performance, their computational demands make real-time serving impractical, often forcing practitioners to rely on knowledge distillation-compromising serving quality for efficiency. To address this challenge, we present SOLARIS (Speculative Offloading of Latent-bAsed Representation for Inference Scaling), a novel framework inspired by speculative decoding. SOLARIS proactively precomputes user-item interaction embeddings by predicting which user-item pairs are likely to appear in future requests, and asynchronously generating their foundation model representations ahead of time. This approach decouples the costly foundation model inference from the latency-critical serving path, enabling real-time knowledge transfer from models previously considered too expensive for online use. Deployed across Meta's advertising system serving billions of daily requests, SOLARIS achieves 0.67% revenue-driving top-line metrics gain, demonstrating its effectiveness at scale.", "authors": ["Zikun Liu", "Liang Luo", "Qianru Li", "Zhengyu Zhang", "Wei Ling", "Jingyi Shen", "Zeliang Chen", "Yaning Huang", "Jingxian Huang", "Abdallah Aboelela", "Chonglin Sun", "Feifan Gu", "Fenggang Wu", "Hang Qu", "Huayu Li", "Jill Pan", "Kaidi Pei", "Laming Chen", "Longhao Jin", "Qin Huang", "Tongyi Tang", "Varna Puvvada", "Wenlin Chen", "Xiaohan Wei", "Xu Cao", "Yantao Yao", "Yuan Jin", "Yunchen Pu", "Yuxin Chen", "Zijian Shen", "Zhengkai Zhang", "Jing Zhu", "Dong Liang", "Ellie Wen"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-13", "url": "https://arxiv.org/abs/2604.12110", "pdf_url": "https://arxiv.org/pdf/2604.12110v2", "arxiv_id": "2604.12110", "doi": "10.48550/arXiv.2604.12110", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.542}
{"id": "43a41d9db3e337a905ba1e7c42a98b0ac7f64cb56ff4b3c93b96769b15eaae26", "sources": ["arxiv", "semantic_scholar"], "title": "SpecMoE: A Fast and Efficient Mixture-of-Experts Inference via Self-Assisted Speculative Decoding", "abstract": "The Mixture-of-Experts (MoE) architecture has emerged as a promising approach to mitigate the rising computational costs of large language models (LLMs) by selectively activating parameters. However, its high memory requirements and sub-optimal parameter efficiency pose significant challenges for efficient deployment. Although CPU-offloaded MoE inference systems have been proposed in the literature, they offer limited efficiency, particularly for large batch sizes. In this work, we propose SpecMoE, a memory-efficient MoE inference system based on our self-assisted speculative decoding algorithm. SpecMoE demonstrates the effectiveness of applying speculative decoding to MoE inference without requiring additional model training or fine-tuning. Our system improves inference throughput by up to $4.30\\times$, while significantly reducing bandwidth requirements of both memory and interconnect on memory-constrained systems.", "authors": ["Jehyeon Bang", "Eunyeong Cho", "Ranggi Hwang", "Jinha Chung", "Minsoo Rhu"], "categories": ["cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-11", "url": "https://arxiv.org/abs/2604.10152", "pdf_url": "https://arxiv.org/pdf/2604.10152v1", "arxiv_id": "2604.10152", "doi": "10.48550/arXiv.2604.10152", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5397}
{"id": "4d16de4c79e9bca2245057aa3a10cb125e3eae867fe128fc727acdb23325f531", "sources": ["arxiv", "semantic_scholar"], "title": "DIVERSED: Relaxed Speculative Decoding via Dynamic Ensemble Verification", "abstract": "Speculative decoding is an effective technique for accelerating large language model inference by drafting multiple tokens in parallel. In practice, its speedup is often bottlenecked by a rigid verification step that strictly enforces the accepted token distribution to exactly match the target model. This constraint leads to the rejection of many plausible tokens, lowering the acceptance rate and limiting overall time speedup. To overcome this limitation, we propose Dynamic Verification Relaxed Speculative Decoding (DIVERSED), a relaxed verification framework that improves time efficiency while preserving generation quality. DIVERSED learns an ensemble-based verifier that blends the draft and target model distributions with a task-dependent and context-dependent weight. We provide theoretical justification for our approach and demonstrate empirically that DIVERSED achieves substantially higher inference efficiency compared to standard speculative decoding methods. Code is available at: https://github.com/comeusr/diversed.", "authors": ["Ziyi Wang", "Siva Rajesh Kasa", "Ankith M S", "Santhosh Kumar Kasa", "Jiaru Zou", "Sumit Negi", "Ruqi Zhang", "Nan Jiang", "Qifan Song"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-08", "url": "https://arxiv.org/abs/2604.07622", "pdf_url": "https://arxiv.org/pdf/2604.07622v1", "arxiv_id": "2604.07622", "doi": "10.48550/arXiv.2604.07622", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/comeusr/diversed", "venue": "arXiv.org", "quality_score": 0.8287}
{"id": "b317d3ab008f1b442542c4d81d4c7b801ecd768774a3840a61cf1f57bebe987f", "sources": ["arxiv", "semantic_scholar"], "title": "ConfigSpec: Profiling-Based Configuration Selection for Distributed Edge--Cloud Speculative LLM Serving", "abstract": "Speculative decoding enables collaborative Large Language Model (LLM) inference across cloud and edge by separating lightweight token drafting from heavyweight verification. While prior systems show performance and cost benefits, practical deployment requires navigating a large configuration space spanning draft model variants, quantisation levels, speculative lengths, and heterogeneous edge devices. This paper presents ConfigSpec, a configurationselection framework for distributed speculative LLM serving. ConfigSpec profiles edge devices and draft-target alignment, and models drafting throughput, acceptance rate, and power to evaluate goodput, verification cost efficiency, and energy efficiency across the joint configuration space. Our analysis across three edge platforms and two LLM families reveals structurally conflicting optima. Firstly, goodput is maximised by the smallest, fastest draft model at device-dependent speculative lengths (K*=2-10). Secondly, both cost and energy efficiency converge to K=2 due to a dominant bonus-token effect-with cost favouring the largest drafter for its high acceptance rate and energy favouring the smallest for its low power draw. These conflicts confirm that no single fixed configuration can simultaneously optimise all objectives, underscoring the need for profiling-based configuration selection in disaggregated edge-cloud LLM inference.", "authors": ["Xiangchen Li", "Saeid Ghafouri", "Jiakun Fan", "Babar Ali", "Hans Vandierendonck", "Dimitrios S. Nikolopoulos"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-08", "url": "https://arxiv.org/abs/2604.09722", "pdf_url": "https://arxiv.org/pdf/2604.09722v1", "arxiv_id": "2604.09722", "doi": "10.1145/3802513.3803483", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3412}
{"id": "91d22f655e9b6cffc3cd2efe44d13a8207676a491cb521b5467a0644748e4e28", "sources": ["arxiv", "semantic_scholar"], "title": "NanoSpec: Accelerating Speculative Decoding using Minimalist In-Context Vocabularies", "abstract": "The massive vocabulary sizes of large language models, often exceeding 100k tokens, impose a computational bottleneck on the final linear projection layer during speculative decoding. Existing vocabulary pruning solutions rely on static or coarsely-grained sub-vocabularies that necessitate large active sizes ($\\sim$30k) to maintain draft quality. We propose NanoSpec, a novel training-free approach that breaks this trade-off by dynamically constructing a minimalist, context-aware active vocabulary for each generation step. Leveraging the inherent temporal locality of language generation, NanoSpec achieves high coverage while slashing the average vocabulary size by over $40\\times$ (to $<$3k tokens) without requiring any auxiliary trained parameters. To realize the theoretical benefits of such high sparsity on modern hardware, we introduce a system-algorithm co-design that overcomes the inefficiencies of sparse memory access through asynchronous gathering and GPU-resident state management. As a complementary plug-and-play module, NanoSpec cuts draft time by an average of 51.6\\%, delivering a $1.17$-$1.29\\times$ end-to-end speedup over the state-of-the-art speculative decoding methods EAGLE-2 and EAGLE-3 across 7 tasks and outperforming complex training-based pruning baselines.", "authors": ["Zhiyang Chen", "Daliang Xu", "Yinyuan Zhang", "Chenghua Wang", "Mengwei Xu", "Yun Ma"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-08", "url": "https://arxiv.org/abs/2605.26444", "pdf_url": "https://arxiv.org/pdf/2605.26444v2", "arxiv_id": "2605.26444", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3412}
{"id": "c873da0c66f4c88ca9b2de481640ff59ea34cdd6ef0cee77400ba147da820047", "sources": ["arxiv", "semantic_scholar"], "title": "See the Forest for the Trees: Loosely Speculative Decoding via Visual-Semantic Guidance for Efficient Inference of Video LLMs", "abstract": "Video Large Language Models (Video-LLMs) excel in video understanding but suffer from high inference latency during autoregressive generation. Speculative Decoding (SD) mitigates this by applying a draft-and-verify paradigm, yet existing methods are constrained by rigid exact-match rules, severely limiting the acceleration potential. To bridge this gap, we propose LVSpec, the first training-free loosely SD framework tailored for Video-LLMs. Grounded in the insight that generation is governed by sparse visual-relevant anchors (mandating strictness) amidst abundant visual-irrelevant fillers (permitting loose verification), LVSpec employs a lightweight visual-relevant token identification scheme to accurately pinpoint the former. To further maximize acceptance, we augment this with a position-shift tolerant mechanism that effectively salvages positionally mismatched but semantically equivalent tokens. Experiments demonstrate that LVSpec achieves high fidelity and speed: it preserves >99.8 of target performance while accelerating Qwen2.5-VL-32B by 2.70x and LLaVA-OneVision-72B by 2.94x. Notably, it boosts the mean accepted length and speedup ratio by 136% and 35% compared to SOTA training-free SD methods for Video-LLMs.", "authors": ["Yicheng Ji", "Jun Zhang", "Jinpeng Chen", "Cong Wang", "Lidan Shou", "Gang Chen", "Huan Li"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-07", "url": "https://arxiv.org/abs/2604.05650", "pdf_url": "https://arxiv.org/pdf/2604.05650v2", "arxiv_id": "2604.05650", "doi": "10.48550/arXiv.2604.05650", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5351}
{"id": "f7654b7effe7ed608afc581723dc36692465f2d861631e2f440f5a81f87e5652", "sources": ["arxiv", "semantic_scholar"], "title": "Multi-Drafter Speculative Decoding with Alignment Feedback", "abstract": "Speculative decoding (SD) accelerates large language model (LLM) inference by using a smaller model to draft future tokens, which are then verified by the target LLM. This preserves generation quality by accepting only aligned tokens. However, individual drafters, often trained for specific tasks or domains, exhibit limited effectiveness across diverse applications. To address this, we introduce \\textsc{MetaSD}, a unified framework that integrates multiple drafters into the SD process. MetaSD dynamically allocates computational resources to heterogeneous drafters by leveraging alignment feedback and framing drafter selection as a multi-armed bandit problem. Extensive experiments show MetaSD consistently outperforms single-drafter approaches.", "authors": ["Taehyeon Kim", "Hojung Jung", "Se-Young Yun"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-07", "url": "https://arxiv.org/abs/2604.05417", "pdf_url": "https://arxiv.org/pdf/2604.05417v1", "arxiv_id": "2604.05417", "doi": "10.48550/arXiv.2604.05417", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5351}
{"id": "853edba5b9debb95b784b406e5131475cede78b0682ac41c8344994f6e9e8441", "sources": ["arxiv", "semantic_scholar"], "title": "Epistemic Blinding: An Inference-Time Protocol for Auditing Prior Contamination in LLM-Assisted Analysis", "abstract": "This paper presents epistemic blinding in the context of an agentic system that uses large language models to reason across multiple biological datasets for drug target prioritization. During development, it became apparent that LLM outputs silently blend data-driven inference with memorized priors about named entities - and the blend is invisible: there is no way to determine, from a single output, how much came from the data on the page and how much came from the model's training memory. Epistemic blinding is a simple inference-time protocol that replaces entity identifiers with anonymous codes before prompting, then compares outputs against an unblinded control. The protocol does not make LLM reasoning deterministic, but it restores one critical axis of auditability: measuring how much of an output came from the supplied data versus the model's parametric knowledge. The complete target identification system is described - including LLM-guided evolutionary optimization of scoring functions and blinded agentic reasoning for target rationalization - with demonstration that both stages operate without access to entity identity. In oncology drug target prioritization across four cancer types, blinding changes 16% of top-20 predictions while preserving identical recovery of validated targets. The contamination problem is shown to generalize beyond biology: in S&P 500 equity screening, brand-recognition bias reshapes 30-40% of top-20 rankings across five random seeds. To lower the barrier to adoption, the protocol is released as an open-source tool and as a Claude Code skill that enables one-command epistemic blinding within agentic workflows. The claim is not that blinded analysis produces better results, but that without blinding, there is no way to know to what degree the agent is adhering to the analytical process the researcher designed.", "authors": ["Michael Cuccarese"], "categories": ["cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-07", "url": "https://arxiv.org/abs/2604.06013", "pdf_url": "https://arxiv.org/pdf/2604.06013v1", "arxiv_id": "2604.06013", "doi": "10.48550/arXiv.2604.06013", "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/mcuccarese/epistemic-blinding", "venue": "arXiv.org", "quality_score": 0.827}
{"id": "67c2514fad7827b15b9bccca9d88b97e3b9525d755d92f93b35fc2b0ee19ca7e", "sources": ["arxiv", "semantic_scholar"], "title": "DualDiffusion: A Speculative Decoding Strategy for Masked Diffusion Models", "abstract": "Masked Diffusion Models (MDMs) offer a promising alternative to autoregressive language models by enabling parallel token generation and bidirectional context modeling. However, their inference speed is significantly limited by the inability to cache key-value pairs due to bidirectional attention, requiring $O(N^2)$ computations at each generation step. While recent methods like FastDLLM and DkvCache improve inference speed through attention approximations and caching strategies, they achieve speedups at the cost of generation quality. We propose DualDiffusion, a speculative decoding framework for MDMs that combines fast drafter models (using efficient approximations) with slower, more accurate verifier models. By running multiple steps of a lightweight drafter followed by a single verification step, DualDiffusion achieves a superior Pareto frontier between generation steps and accuracy compared to existing approaches. We evaluate our method on MMLU and GSM8K, demonstrating that DualDiffusion maintains high accuracy while reducing the number of generation steps required, effectively pushing the quality-efficiency trade-off curve for masked diffusion language models.", "authors": ["Satyam Goyal", "Kushal Patel", "Tanush Mittal", "Arjun Laxman"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-06", "url": "https://arxiv.org/abs/2604.05250", "pdf_url": "https://arxiv.org/pdf/2604.05250v1", "arxiv_id": "2604.05250", "doi": "10.48550/arXiv.2604.05250", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.534}
{"id": "30ce6022ff2dfe54ed50bcd76bf29f72f46be2f9ed6649bc9c49fd1eca86369c", "sources": ["arxiv", "semantic_scholar"], "title": "Cactus: Accelerating Auto-Regressive Decoding with Constrained Acceptance Speculative Sampling", "abstract": "Speculative sampling (SpS) has been successful in accelerating the decoding throughput of auto-regressive large language models by leveraging smaller draft models. SpS strictly enforces the generated distribution to match that of the verifier LLM. This is unnecessarily restrictive as slight variations of the verifier's distribution, such as sampling with top-$k$ or temperature, would also be acceptable. Typical acceptance sampling (TAS) alleviates this issue by accepting more tokens using entropy-based heuristics. However, this approach distorts the verifier distribution, potentially degrading output quality when the verifier encodes critical information. In this work, we formalize the speculative sampling algorithm through the lens of constrained optimization. Based on this formulation, we propose Cactus (constrained acceptance speculative sampling), a method that guarantees controlled divergence from the verifier distribution and increasing acceptance rates. Empirical results across a wide range of benchmarks confirm the effectiveness of our approach.", "authors": ["Yongchang Hao", "Lili Mou"], "categories": ["cs.LG", "cs.AI", "math.OC", "stat.ML"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-04-05", "url": "https://arxiv.org/abs/2604.04987", "pdf_url": "https://arxiv.org/pdf/2604.04987v1", "arxiv_id": "2604.04987", "doi": "10.48550/arXiv.2604.04987", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5328}
{"id": "c027a2df4835c9936edcc72c0cab4ae6defe2d2d19e6e70104b581c1777b6b27", "sources": ["arxiv", "semantic_scholar"], "title": "Goose: Anisotropic Speculation Trees for Training-Free Speculative Decoding", "abstract": "Speculative decoding accelerates large language model inference by drafting multiple candidate tokens and verifying them in a single forward pass. Candidates are organized as a tree: deeper trees accept more tokens per step, but adding depth requires sacrificing breadth (fallback options) under a fixed verification budget. Existing training-free methods draft from a single token source and shape their trees without distinguishing candidate quality across origins. We observe that two common training-free token sources - n-gram matches copied from the input context, and statistical predictions from prior forward passes - differ dramatically in acceptance rate (~6x median gap, range 2-18x across five models and five benchmarks). We prove that when such a quality gap exists, the optimal tree is anisotropic (asymmetric): reliable tokens should form a deep chain while unreliable tokens spread as wide branches, breaking through the depth limit of balanced trees. We realize this structure in GOOSE, a training-free framework that builds an adaptive spine tree - a deep chain of high-acceptance context-matched tokens with wide branches of low-acceptance alternatives at each node. We prove that the number of tokens accepted per step is at least as large as that of either source used alone. On five LLMs (7B-33B) and five benchmarks, GOOSE achieves 1.9-4.3x lossless speedup, outperforming balanced-tree baselines by 12-33% under the same budget.", "authors": ["Tao Jin", "Phuong Minh Nguyen", "Naoya Inoue"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-02", "url": "https://arxiv.org/abs/2604.02047", "pdf_url": "https://arxiv.org/pdf/2604.02047v1", "arxiv_id": "2604.02047", "doi": "10.48550/arXiv.2604.02047", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5294}
{"id": "3a497fbfb0502bfd45305ceb09c1bb9ade58a190561a068d502c828d4f7769c6", "sources": ["arxiv", "semantic_scholar"], "title": "SpecTr-GBV: Multi-Draft Block Verification Accelerating Speculative Decoding", "abstract": "Autoregressive language models suffer from high inference latency due to their sequential decoding nature. Speculative decoding (SD) mitigates this by employing a lightweight draft model to propose candidate tokens, which are selectively verified by a larger target model. While existing methods either adopt multi-draft strategies to increase acceptance rates or block verification techniques to jointly verify multiple tokens, they remain limited by treating these improvements in isolation. In this work, we propose SpecTr-GBV, a novel SD method that unifies multi-draft and greedy block verification (GBV) into a single framework. By formulating the verification step as an optimal transport problem over draft and target token blocks, SpecTr-GBV improves both theoretical efficiency and empirical performance. We theoretically prove that SpecTr-GBV achieves the optimal expected acceptance length physically attainable within the framework of i.i.d. draft generation, and this bound improves as the number of drafts increases. Empirically, we evaluate SpecTr-GBV across five datasets and four baselines. Our method achieves superior speedup and significantly higher block efficiency while preserving output quality. In addition, we perform comprehensive ablation studies to evaluate the impact of various hyperparameters in the model.", "authors": ["Yijun Lin", "Jinhao Sheng", "Qingyue Cai", "Feng Zhou"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-04-01", "url": "https://arxiv.org/abs/2604.25925", "pdf_url": "https://arxiv.org/pdf/2604.25925v1", "arxiv_id": "2604.25925", "doi": "10.48550/arXiv.2604.25925", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5282}
{"id": "3d752b08d43b132c7377162cb1892a015776414478211ccb8623a86c45dd53c0", "sources": ["arxiv", "semantic_scholar"], "title": "SJD-VP: Speculative Jacobi Decoding with Verification Prediction for Autoregressive Image Generation", "abstract": "Speculative Jacobi Decoding (SJD) has emerged as a promising method for accelerating autoregressive image generation. Despite its potential, existing SJD approaches often suffer from the low acceptance rate issue of speculative tokens due to token selection ambiguity. Recent works attempt to mitigate this issue primarily from the relaxed token verification perspective but fail to fully exploit the iterative dynamics of decoding. In this paper, we conduct an in-depth analysis and make a novel observation that tokens whose probabilities increase are more likely to match the verification-accepted and correct token. Based on this, we propose a novel Speculative Jacobi Decoding with Verification Prediction (SJD-VP). The key idea is to leverage the change in token probabilities across iterations to guide sampling, favoring tokens whose probabilities increase. This effectively predicts which tokens are likely to pass subsequent verification, boosting the acceptance rate. In particular, our SJD-VP is plug-and-play and can be seamlessly integrated into existing SJD methods. Extensive experiments on standard benchmarks demonstrate that our SJD-VP method consistently accelerates autoregressive decoding while improving image generation quality.", "authors": ["Bingqi Shan", "Baoquan Zhang", "Xiaochen Qi", "Xutao Li", "Yunming Ye", "Liqiang Nie"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-28", "url": "https://arxiv.org/abs/2603.27115", "pdf_url": "https://arxiv.org/pdf/2603.27115v1", "arxiv_id": "2603.27115", "doi": "10.48550/arXiv.2603.27115", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5236}
{"id": "c250d01b0d9a40fa17b9da9abc452c172dc733f062274fabe65ab63a70952883", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating PayPal's Commerce Agent with Speculative Decoding: An Empirical Study on EAGLE3 with Fine-Tuned Nemotron Models", "abstract": "We evaluate speculative decoding with EAGLE3 as an inference-time optimization for PayPal's Commerce Agent, powered by a fine-tuned llama3.1-nemotron-nano-8B-v1 model. Building on prior work (NEMO-4-PAYPAL) that reduced latency and cost through domain-specific fine-tuning, we benchmark EAGLE3 via vLLM against NVIDIA NIM on identical 2xH100 hardware across 40 configurations spanning speculative token counts (gamma=3, gamma=5), concurrency levels (1-32), and sampling temperatures (0, 0.5). Key findings: (1) gamma=3 achieves 22-49% throughput improvement and 18-33% latency reduction at zero additional hardware cost; (2) acceptance rates remain stable at approximately 35.5% for gamma=3 across all conditions; (3) gamma=5 yields diminishing returns (approximately 25% acceptance rate); (4) LLM-as-Judge evaluation confirms fully preserved output quality; and (5) speculative decoding on a single H100 matches or exceeds NIM on two H100s, enabling 50% GPU cost reduction.", "authors": ["Ally Qin", "Jian Wan", "Sarat Mudunuri", "Srinivasan Manoharan"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-27", "url": "https://arxiv.org/abs/2604.19767", "pdf_url": "https://arxiv.org/pdf/2604.19767v1", "arxiv_id": "2604.19767", "doi": "10.48550/arXiv.2604.19767", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5225}
{"id": "bdae8340baa21b72b415b7a3551242849ab776db4818aab69c35557611755ea4", "sources": ["arxiv", "semantic_scholar"], "title": "S2D2: Fast Decoding for Diffusion LLMs via Training-Free Self-Speculation", "abstract": "Block-diffusion language models offer a promising path toward faster-than-autoregressive generation by combining block-wise autoregressive decoding with within-block parallel denoising. However, in the few-step regime needed for practical acceleration, standard confidence-thresholded decoding is often brittle: aggressive thresholds hurt quality, while conservative thresholds require unnecessary denoising steps. Existing approaches that address this issue either require additional training or incur extra test-time compute. We present S2D2, a training-free self-speculative decoding framework for block-diffusion language models. Our key observation is that a block-diffusion model becomes autoregressive when the block size is reduced to one, allowing the same pretrained model to act as both drafter and verifier. S2D2 inserts a speculative verification step into standard block-diffusion decoding and uses lightweight routing policies to decide when verification is worth its cost. This yields a hybrid decoding trajectory in which diffusion proposes tokens in parallel, while the autoregressive mode acts as a local sequence-level critic. Across three mainstream block-diffusion families, S2D2 consistently improves the accuracy-speed tradeoff over strong confidence-thresholding baselines. On SDAR, we observe up to $4.7\\times$ speedup over autoregressive decoding, and up to $1.57\\times$ over a tuned dynamic decoding baseline while improving accuracy by up to $4.5$ points. On LLaDA2.1-Mini, S2D2 remains complementary to built-in self-correction, including a conservative setting where it is $4.4\\times$ faster than the static baseline with slightly higher accuracy.", "authors": ["Ligong Han", "Hao Wang", "Han Gao", "Kai Xu", "Akash Srivastava"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-26", "url": "https://arxiv.org/abs/2603.25702", "pdf_url": "https://arxiv.org/pdf/2603.25702v2", "arxiv_id": "2603.25702", "doi": "10.48550/arXiv.2603.25702", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/phymhan/S2D2", "venue": "arXiv.org", "quality_score": 0.8057}
{"id": "4a38df3d05a310265387354c0143f000c1a40a3d03e88171e55ac7efca89fa73", "sources": ["arxiv", "semantic_scholar"], "title": "Cross-Family Speculative Decoding for Polish Language Models on Apple~Silicon: An Empirical Evaluation of Bielik~11B with UAG-Extended MLX-LM", "abstract": "Speculative decoding accelerates LLM inference by using a small draft model to propose k candidate tokens for a target model to verify. While effective for same-tokenizer pairs on high-bandwidth GPUs, its applicability to cross-family pairs with mismatched tokenizers and consumer-grade unified memory remains underexplored. We extend the MLX-LM framework with Universal Assisted Generation (UAG) to enable cross-tokenizer speculative decoding on Apple Silicon. We evaluate Bielik 11B-Instruct (Mistral-based) as the target model, paired with three draft models: Bielik 1.5B (Qwen-based with custom tokenizer), Qwen2.5-1.5B, and Llama 3.2-1B. Experiments on three Polish-language datasets (Wikipedia, pl_alpaca, synthetic) use draft lengths k in {2, 4, 6} to compare naive and context-aware token translation. Results show: (1) context-aware translation consistently improves acceptance rates across all configurations; (2) the Polish-specialized Bielik 1.5B achieves lower acceptance than general-purpose Qwen2.5 and Llama 3.2 drafters; (3) throughput on Apple Silicon is content-dependent, reaching 1.7x speedup for structured text but failing for varied instructions; and (4) verification cost on unified memory does not amortize as theory predicts because both models are memory-bandwidth bound, making sequential drafting expensive relative to batched verification. We propose a hardware-aware speedup formula and characterize conditions for cross-family speculative decoding on Apple Silicon. This is the first systematic evaluation of cross-family speculative decoding for Polish LLMs and the first empirical study of UAG-based decoding on unified memory architectures.", "authors": ["Krzysztof Fonal"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-22", "url": "https://arxiv.org/abs/2604.16368", "pdf_url": "https://arxiv.org/pdf/2604.16368v2", "arxiv_id": "2604.16368", "doi": "10.48550/arXiv.2604.16368", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5168}
{"id": "cfe21e34399c12f1675a9430a3aa83aa511cf579cd960b145f080e20e98065ae", "sources": ["arxiv", "semantic_scholar"], "title": "ParallelVLM: Lossless Video-LLM Acceleration with Visual Alignment Aware Parallel Speculative Decoding", "abstract": "Although current Video-LLMs achieve impressive performance in video understanding tasks, their autoregressive decoding efficiency remains constrained by the massive number of video tokens. Visual token pruning can partially ease this bottleneck, yet existing approaches still suffer from information loss and yield only modest acceleration in decoding. In this paper, we propose ParallelVLM, a training-free draft-then-verify speculative decoding framework that overcomes both mutual waiting and limited speedup-ratio problems between draft and target models in long-video settings. ParallelVLM features two parallelized stages that maximize hardware utilization and incorporate an Unbiased Verifier-Guided Pruning strategy to better align the draft and target models by eliminating the positional bias in attention-guided pruning. Extensive experiments demonstrate that ParallelVLM effectively expands the draft window by $1.6\\sim1.8\\times$ with high accepted lengths, and accelerates various video understanding benchmarks by 3.36$\\times$ on LLaVA-Onevision-72B and 2.42$\\times$ on Qwen2.5-VL-32B compared with vanilla autoregressive decoding.", "authors": ["Quan Kong", "Yuhao Shen", "Yicheng Ji", "Huan Li", "Cong Wang"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-20", "url": "https://arxiv.org/abs/2603.19610", "pdf_url": "https://arxiv.org/pdf/2603.19610v2", "arxiv_id": "2603.19610", "doi": "10.48550/arXiv.2603.19610", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5145}
{"id": "9ec11f88936ceb5debb2fc82a2b85a9a258463900f58df8d8380f032b8400920", "sources": ["arxiv", "semantic_scholar"], "title": "A Pipelined Collaborative Speculative Decoding Framework for Efficient Edge-Cloud LLM Inference", "abstract": "Recent advancements and widespread adoption of Large Language Models (LLMs) in both industry and academia have catalyzed significant demand for LLM serving. However, traditional cloud services incur high costs, while on-device inference alone faces challenges due to limited resources. Edge-cloud collaboration emerges as a key research direction to combine the strengths of both paradigms, yet efficiently utilizing limited network bandwidth while fully leveraging and balancing the computational capabilities of edge devices and the cloud remains an open problem. To address these challenges, we propose Pipelined Collaborative Speculative Decoding Framework (PicoSpec), a novel, general-purpose, and training-free speculative decoding framework for LLM edge-cloud collaborative inference. We design an asynchronous pipeline that resolves the mutual waiting problem inherent in vanilla speculative decoding within edge collaboration scenarios, which concurrently executes a Small Language Model (SLM) on the edge device and a LLM in the cloud. Meanwhile, to mitigate the significant communication latency caused by transmitting vocabulary distributions, we introduce separate rejection sampling with sparse compression, which completes the rejection sampling with only a one-time cost of transmitting the compressed vocabulary. Experimental results demonstrate that our solution outperforms baseline and existing methods, achieving up to 2.9 speedup.", "authors": ["Yida Zhang", "Zhiyong Gao", "Shuaibing Yue", "Jie Li", "Rui Wang"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-19", "url": "https://arxiv.org/abs/2603.19133", "pdf_url": "https://arxiv.org/pdf/2603.19133v2", "arxiv_id": "2603.19133", "doi": "10.48550/arXiv.2603.19133", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5133}
{"id": "f41fea8a25a5f7a5bade1a111f0c3b7c5a231af4297bbd0884723459ec875e04", "sources": ["arxiv", "semantic_scholar"], "title": "SpecForge: A Flexible and Efficient Open-Source Training Framework for Speculative Decoding", "abstract": "Large language models incur high inference latency due to sequential autoregressive decoding. Speculative decoding alleviates this bottleneck by using a lightweight draft model to propose multiple tokens for batched verification. However, its adoption has been limited by the lack of high-quality draft models and scalable training infrastructure. We introduce SpecForge, an open-source, production-oriented framework for training speculative decoding models with full support for EAGLE-3. SpecForge incorporates target-draft decoupling, hybrid parallelism, optimized training kernels, and integration with production-grade inference engines, enabling up to 9.9x faster EAGLE-3 training for Qwen3-235B-A22B. In addition, we release SpecBundle, a suite of production-grade EAGLE-3 draft models trained with SpecForge for mainstream open-source LLMs. Through a systematic study of speculative decoding training recipes, SpecBundle addresses the scarcity of high-quality drafts in the community, and our draft models achieve up to 4.48x end-to-end inference speedup on SGLang, establishing SpecForge as a practical foundation for real-world speculative decoding deployment.", "authors": ["Shenggui Li", "Chao Wang", "Yikai Zhu", "Yubo Wang", "Fan Yin", "Shuai Shi", "Yefei Chen", "Xiaomin Dong", "Qiaoling Chen", "Jin Pan", "Ji Li", "Laixin Xie", "Yineng Zhang", "Lei Yu", "Yonggang Wen", "Ivor Tsang", "Tianwei Zhang"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-19", "url": "https://arxiv.org/abs/2603.18567", "pdf_url": "https://arxiv.org/pdf/2603.18567v1", "arxiv_id": "2603.18567", "doi": "10.48550/arXiv.2603.18567", "citation_count": 4, "influential_citation_count": 0, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.7933}
{"id": "88d6a550b2363922bfcd189efc53766d327fd7a9668ef999cd090974fd5a9516", "sources": ["arxiv", "semantic_scholar"], "title": "MMSpec: Benchmarking Speculative Decoding for Vision-Language Models", "abstract": "Vision-language models (VLMs) achieve strong performance on multimodal tasks but suffer from high inference latency due to large model sizes and long multimodal contexts. Speculative decoding has recently emerged as an effective acceleration technique, yet its behavior in VLMs remains insufficiently understood. We introduce MMSpec, the first benchmark for evaluating speculative decoding in vision-language models. MMSpec contains 600 multimodal samples across six task categories and integrates ten representative speculative decoding algorithms under a unified evaluation framework. Our study reveals three key findings: (1) methods designed for text-only LLMs degrade in multimodal scenarios, (2) vision awareness becomes increasingly important at larger batch sizes, and (3) throughput speedup alone does not reliably reflect latency performance. Motivated by these findings, we propose ViSkip, a plug-and-play speculative decoding method that dynamically adapts speculation to vision tokens and achieves state-of-the-art performance.", "authors": ["Hui Shen", "Xin Wang", "Ping Zhang", "Yunta Hsieh", "Qi Han", "Zhongwei Wan", "Ziheng Zhang", "Jingxuan Zhang", "Jing Xiong", "Ziyuan Liu", "Yifan Zhang", "Hangrui Cao", "Chenyang Zhao", "Mi Zhang"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-16", "url": "https://arxiv.org/abs/2603.14989", "pdf_url": "https://arxiv.org/pdf/2603.14989v1", "arxiv_id": "2603.14989", "doi": "10.48550/arXiv.2603.14989", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5099}
{"id": "2b2222250aa6a1ce1dbd18608d969416eeb11ae94626a4c78c643c812ad71ac6", "sources": ["arxiv", "semantic_scholar"], "title": "When Drafts Evolve: Speculative Decoding Meets Online Learning", "abstract": "Speculative decoding has emerged as a widely adopted paradigm for accelerating large language model inference, where a lightweight draft model rapidly generates candidate tokens that are then verified in parallel by a larger target model. However, due to limited model capacity, drafts often struggle to approximate the target distribution, resulting in shorter acceptance lengths and diminished speedup. A key yet under-explored observation is that speculative decoding inherently provides verification feedback that quantifies the deviation between the draft and target models at no additional cost. This process naturally forms an iterative \"draft commits-feedback provides-draft adapts\" evolving loop, which precisely matches the online learning paradigm. Motivated by this connection, we propose OnlineSpec, a unified framework that systematically leverages interactive feedback to continuously evolve draft models. Grounded in dynamic regret minimization, we establish a formal link between online learning performance and speculative system's acceleration rate, and develop novel algorithms via modern online learning techniques, including optimistic online learning that adaptively reuses historical gradients as predictive update hints, and online ensemble learning that dynamically maintains multiple draft models. Our algorithms are equipped with theoretical justifications and improved acceleration rates, achieving up to 24% speedup over seven benchmarks and three foundation models.", "authors": ["Yu-Yang Qian", "Hao-Cong Wu", "Yichao Fu", "Hao Zhang", "Peng Zhao"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-13", "url": "https://arxiv.org/abs/2603.12617", "pdf_url": "https://arxiv.org/pdf/2603.12617v1", "arxiv_id": "2603.12617", "doi": "10.48550/arXiv.2603.12617", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5065}
{"id": "ad0b7ebfcb444929dbf27ba4be4f0c965b48be60371d65288f1cba1c7d528740", "sources": ["arxiv", "semantic_scholar"], "title": "Self-Speculative Decoding for LLM-based ASR with CTC Encoder Drafts", "abstract": "We propose self-speculative decoding for speech-aware LLMs by using the CTC encoder as a draft model to accelerate auto-regressive (AR) inference and improve ASR accuracy. Our three-step procedure works as follows: (1) if the frame entropies of the CTC output distributions are below a threshold, the greedy CTC hypothesis is accepted as final; (2) otherwise, the CTC hypothesis is verified in a single LLM forward pass using a relaxed acceptance criterion based on token likelihoods; (3) if verification fails, AR decoding resumes from the accepted CTC prefix. Experiments on nine corpora and five languages show that this approach can simultaneously accelerate decoding and reduce WER. On the HuggingFace Open ASR benchmark with a 1B parameter LLM and 440M parameter CTC encoder, we achieve a record 5.58% WER and improve the inverse real time factor by a factor of 4.4 with only a 12% relative WER increase over AR search. Code and model weights are publicly available under a permissive license.", "authors": ["George Saon", "Samuel Thomas", "Takashi Fukuda", "Tohru Nagano", "Avihu Dekel", "Luis Lastras"], "categories": ["eess.AS"], "fields_of_study": ["Engineering"], "published_date": "2026-03-11", "url": "https://arxiv.org/abs/2603.11243", "pdf_url": "https://arxiv.org/pdf/2603.11243v1", "arxiv_id": "2603.11243", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3208}
{"id": "620fa0ce4f1a630349e109eebd49b958c67bc89e28f7c52f41751f5fa04fee86", "sources": ["arxiv", "semantic_scholar"], "title": "How Contrastive Decoding Enhances Large Audio Language Models?", "abstract": "While Contrastive Decoding (CD) has proven effective at enhancing Large Audio Language Models (LALMs), the underlying mechanisms driving its success and the comparative efficacy of different strategies remain unclear. This study systematically evaluates four distinct CD strategies across diverse LALM architectures. We identify Audio-Aware Decoding and Audio Contrastive Decoding as the most effective methods. However, their impact varies significantly by model. To explain this variability, we introduce a Transition Matrix framework to map error pattern shifts during inference. Our analysis demonstrates that CD reliably rectifies errors in which models falsely claim an absence of audio or resort to uncertainty-driven guessing. Conversely, it fails to correct flawed reasoning or confident misassertions. Ultimately, these findings provide a clear guideline for determining which LALM architectures are most suitable for CD enhancement based on their baseline error profiles.", "authors": ["Tzu-Quan Lin", "Wei-Ping Huang", "Yi-Cheng Lin", "Hung-yi Lee"], "categories": ["cs.SD", "cs.CL", "eess.AS"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2026-03-10", "url": "https://arxiv.org/abs/2603.09232", "pdf_url": "https://arxiv.org/pdf/2603.09232v1", "arxiv_id": "2603.09232", "doi": "10.48550/arXiv.2603.09232", "citation_count": 3, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/nervjack2/LALM-Contrastive-Decoding-Error-Profiles", "venue": "arXiv.org", "quality_score": 0.7774}
{"id": "d3e4dfe242d37c453fbdb6e9c38dec65e502284122b36fe7132d40b090065e1c", "sources": ["arxiv", "semantic_scholar"], "title": "ECHO: Elastic Speculative Decoding with Sparse Gating for High-Concurrency Scenarios", "abstract": "Speculative Decoding promises to accelerate the inference of Large Language Models, yet its efficacy often degrades in production-grade serving. Existing evaluations typically overlook the compute-bound nature of high-concurrency regimes, where verification compute becomes the dominant bottleneck. Consequently, prior methods face a dilemma: static trees incur massive verification waste, while dynamic trees suffer from cumulative misjudgments and kernel incompatibility. To bridge this gap, we introduce ECHO, a high concurrency-oriented framework integrated into SGLang that reformulates speculative execution as a budgeted scheduling problem. Crucially, ECHO employs sparse confidence gating to manage the batch as a unified super-tree, elastically pivoting budget between depth and width to co-optimize the trade-off between reducing global verification steps and maximizing per-step efficiency. Extensive evaluations across diverse model scales-particularly the industrial-grade Qwen3-235B-demonstrate that ECHO consistently outperforms SOTA methods in both low-load and high-load scenarios, achieving up to 5.35x walltime speedup and delivering over 20% relative speedup gain.", "authors": ["Xinyi Hu", "Yuhao Shen", "Baolin Zhang", "Hengxin Zhang", "Jun Dai", "Shuang Ge", "Lei Chen", "Yue Li", "Mingcheng Wan"], "categories": ["cs.DC", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-10", "url": "https://arxiv.org/abs/2604.09603", "pdf_url": "https://arxiv.org/pdf/2604.09603v2", "arxiv_id": "2604.09603", "doi": "10.48550/arXiv.2604.09603", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.503}
{"id": "919011a72c3e4352bb81d9265a2d7affb7eab9efb814664b756cfab01ae6ea42", "sources": ["arxiv", "semantic_scholar"], "title": "Learning Adaptive LLM Decoding", "abstract": "Decoding from large language models (LLMs) typically relies on fixed sampling hyperparameters (e.g., temperature, top-p), despite substantial variation in task difficulty and uncertainty across prompts and individual decoding steps. We propose to learn adaptive decoding policies that dynamically select sampling strategies at inference time, conditioned on available compute resources. Rather than fine-tuning the language model itself, we introduce lightweight decoding adapters trained with reinforcement learning and verifiable terminal rewards (e.g. correctness on math and coding tasks). At the sequence level, we frame decoding as a contextual bandit problem: a policy selects a decoding strategy (e.g. greedy, top-k, min-p) for each prompt, conditioned on the prompt embedding and a parallel sampling budget. At the token level, we model decoding as a partially observable Markov decision process (POMDP), where a policy selects sampling actions at each token step based on internal model features and the remaining token budget. Experiments on the MATH and CodeContests benchmarks show that the learned adapters improve the accuracy-budget tradeoff: on MATH, the token-level adapter improves Pass@1 accuracy by up to 10.2% over the best static baseline under a fixed token budget, while the sequence-level adapter yields 2-3% gains under fixed parallel sampling. Ablation analyses support the contribution of both sequence- and token-level adaptation.", "authors": ["Chloe H. Su", "Zhe Ye", "Samuel Tenka", "Aidan Yang", "Soonho Kong", "Udaya Ghai"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-10", "url": "https://arxiv.org/abs/2603.09065", "pdf_url": "https://arxiv.org/pdf/2603.09065v2", "arxiv_id": "2603.09065", "doi": "10.48550/arXiv.2603.09065", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.503}
{"id": "2637d50b286df8bbee11e4a10a0a06661adf804b34461dba537faed835ea5169", "sources": ["arxiv", "semantic_scholar"], "title": "EAGLE-Pangu: Accelerator-Safe Tree Speculative Decoding on Ascend NPUs", "abstract": "Autoregressive decoding remains a primary bottleneck in large language model (LLM) serving, motivating speculative decoding methods that reduce expensive teacher-model invocations by verifying multiple candidate tokens per step. Tree-structured speculation further increases parallelism, but is often brittle when ported across heterogeneous backends and accelerator stacks, where attention masking, KV-cache layouts, and indexing semantics are not interchangeable. We present EAGLE-Pangu, a reproducible system that ports EAGLE-3-style tree speculative decoding to a Pangu teacher backend on Ascend NPUs. EAGLE-Pangu contributes (i) an explicit branch/commit cache manager built on the Cache API, (ii) accelerator-safe tree tensorization that removes undefined negative indices by construction and validates structural invariants, and (iii) a fused-kernel-compatible teacher verification path with a debuggable eager fallback. On 240 turns from MT-Bench and HumanEval-style prompts, EAGLE-Pangu improves end-to-end decoding throughput by 1.27x on average, up to 2.46x at p99, over teacher-only greedy decoding in the fused-kernel performance path. We also provide a fused-kernel-free reference path with structured traces and invariant checks to support reproducible debugging and ablation across execution modes and tree budgets.", "authors": ["Chang Han", "Yijie Hu", "Jingling Liu"], "categories": ["cs.LG", "cs.PL"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-09", "url": "https://arxiv.org/abs/2603.08088", "pdf_url": "https://arxiv.org/pdf/2603.08088v1", "arxiv_id": "2603.08088", "doi": "10.48550/arXiv.2603.08088", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5019}
{"id": "abaf72d9dcb8237f170e98af3b8e9308c2e965d7c8234d464348ebbfec8e115b", "sources": ["arxiv", "semantic_scholar"], "title": "Speculating Experts Accelerates Inference for Mixture-of-Experts", "abstract": "Mixture-of-Experts (MoE) models have gained popularity as a means of scaling the capacity of large language models (LLMs) while maintaining sparse activations and reduced per-token compute. However, in memory-constrained inference settings, expert weights must be offloaded to CPU, creating a performance bottleneck from CPU-GPU transfers during decoding. We propose an expert prefetching scheme that leverages currently computed internal model representations to speculate future experts, enabling memory transfers to overlap with computation. Across multiple MoE architectures, we demonstrate that future experts can be reliably predicted by these internal representations. We also demonstrate that executing speculated experts generally maintains downstream task accuracy, thus preserving more effective compute-memory overlap by eliminating the need to re-fetch true router-selected experts. Integrated into an optimized inference engine, our approach achieves up to 14\\% reduction in time per output token (TPOT) over on-demand loading of experts from CPU memory. For MoEs where speculative execution alone yields suboptimal accuracy, we further examine lightweight estimators that improve expert prediction hit rates, thereby reducing performance degradation. Our code is released in open-source at https://github.com/axonn-ai/yalis/tree/offload_prefetch.", "authors": ["Vivan Madan", "Prajwal Singhania", "Abhinav Bhatele", "Tom Goldstein", "Ashwinee Panda"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-09", "url": "https://arxiv.org/abs/2603.19289", "pdf_url": "https://arxiv.org/pdf/2603.19289v1", "arxiv_id": "2603.19289", "doi": "10.48550/arXiv.2603.19289", "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/axonn-ai/yalis/tree/offload_prefetch", "venue": "arXiv.org", "quality_score": 0.7756}
{"id": "940d2378569b310049174fb9a710aa7b4bf69280e958af18c8a3bc4f4319269d", "sources": ["arxiv", "semantic_scholar"], "title": "ConFu: Contemplate the Future for Better Speculative Sampling", "abstract": "Speculative decoding has emerged as a powerful approach to accelerate large language model (LLM) inference by employing lightweight draft models to propose candidate tokens that are subsequently verified by the target model. The effectiveness of this paradigm critically depends on the quality of the draft model. While recent advances such as the EAGLE series achieve state-of-the-art speedup, existing draft models remain limited by error accumulation: they condition only on the current prefix, causing their predictions to drift from the target model over steps. In this work, we propose \\textbf{ConFu} (Contemplate the Future), a novel speculative decoding framework that enables draft models to anticipate the future direction of generation. ConFu introduces (i) contemplate tokens and soft prompts that allow the draft model to leverage future-oriented signals from the target model at negligible cost, (ii) a dynamic contemplate token mechanism with MoE to enable context-aware future prediction, and (iii) a training framework with anchor token sampling and future prediction replication that learns robust future prediction. ConFu improves token acceptance rates and generation speed over EAGLE-3 by 8--11\\% on Llama-3 3B/8B and by approximately 20\\% on Qwen-3 4B across downstream tasks. We believe our work is the first to bridge speculative decoding with continuous reasoning tokens, offering a new direction for accelerating LLM inference.", "authors": ["Zongyue Qin", "Raghavv Goel", "Mukul Gagrani", "Risheek Garrepalli", "Mingu Lee", "Yizhou Sun"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-09", "url": "https://arxiv.org/abs/2603.08899", "pdf_url": "https://arxiv.org/pdf/2603.08899v3", "arxiv_id": "2603.08899", "doi": "10.48550/arXiv.2603.08899", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5019}
{"id": "85ab3f800c4a10936be523e43141f048f35767e726ca2642a30d4b8cd256376b", "sources": ["arxiv", "semantic_scholar"], "title": "Balancing Coverage and Draft Latency in Vocabulary Trimming for Faster Speculative Decoding", "abstract": "Speculative decoding accelerates inference for Large Language Models by using a lightweight draft model to propose candidate tokens that are verified in parallel by a larger target model. Prior work shows that the draft model often dominates speculative decoding latency, since it generates tokens sequentially and incurs high cost from its language modeling head as vocabulary size grows. This exposes a fundamental trade-off in draft model design: larger vocabularies improve token coverage and agreement with the target model, but incur higher draft latency, while smaller vocabularies reduce latency at the risk of missing tokens required for accurate draft generation. We address this trade-off through vocabulary trimming for draft models, motivated by the observation that domain-specific workloads use only a small fraction of the full vocabulary. We cast draft vocabulary selection as a constrained optimization problem that balances token coverage and draft latency. Coverage is computed over assistant responses in the training data, while latency is estimated using architecture-aware FLOPs that capture the cost of the language modeling head as a function of vocabulary size. We optimize a utility function with a Tree-structured Parzen Estimator to efficiently explore the coverage-latency Pareto frontier under a minimum coverage constraint. Experiments show improved speculative decoding throughput while reducing draft vocabularies by up to 97% with high coverage. On domain-specific tasks, we achieve up to 16% latency reduction and 20% throughput improvement, and up to 6.7% throughput gains on diverse out-of-distribution tasks.", "authors": ["Ofir Ben Shoham"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-05", "url": "https://arxiv.org/abs/2603.05210", "pdf_url": "https://arxiv.org/pdf/2603.05210v1", "arxiv_id": "2603.05210", "doi": "10.48550/arXiv.2603.05210", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4973}
{"id": "9965bae6110495f0a86782639cc6e909e3b5d6df537f588e4dc5832c4fc4fc5d", "sources": ["arxiv", "semantic_scholar"], "title": "SLO-Aware Compute Resource Allocation for Prefill-Decode Disaggregated LLM Inference", "abstract": "Prefill-Decode (P/D) disaggregation has emerged as a widely adopted optimization strategy for Large Language Model (LLM) inference. However, there currently exists no well-established methodology for determining the optimal number of P/D hardware resources, subject to constraints on total throughput, service level objectives (SLOs), and request characteristics - specifically input and output lengths. To address this gap, we propose a hybrid approach that combines theoretical modeling with empirical benchmarking. First, we present a theoretical model for calculating P/D resource counts, which is based on total throughput requirements, request input and output lengths, as well as prefill and decode throughput. Then, to obtain the actual prefill and decode throughput under SLO constraints, we model the prefill process using M/M/1 queuing theory, deriving the achieved prefill throughput from the benchmarked maximum prefill throughput and Time-To-First-Token (TTFT). For the decode phase, we determine the decode batch sizes that meet Time-Per-Output-Token (TPOT) requirements and obtain the corresponding decode throughput through empirical measurements. Our experimental results demonstrate that the proposed method can accurately predict optimal P/D resource allocation in real-world LLM inference scenarios.", "authors": ["Luchang Li", "Dongfang Li", "Bozhao Gong", "Yu Zhang"], "categories": ["cs.DC", "cs.IT", "cs.LG"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-03-05", "url": "https://arxiv.org/abs/2603.04716", "pdf_url": "https://arxiv.org/pdf/2603.04716v1", "arxiv_id": "2603.04716", "doi": "10.48550/arXiv.2603.04716", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4973}
{"id": "9f9e27c8313a7425ad13a0442ab2169c45f549374fe6e4b1aa88ce20b0ed42e7", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Speculative Decoding", "abstract": "Autoregressive decoding is bottlenecked by its sequential nature. Speculative decoding has become a standard way to accelerate inference by using a fast draft model to predict upcoming tokens from a slower target model, and then verifying them in parallel with a single target model forward pass. However, speculative decoding itself relies on a sequential dependence between speculation and verification. We introduce speculative speculative decoding (SSD) to parallelize these operations. While a verification is ongoing, the draft model predicts likely verification outcomes and prepares speculations pre-emptively for them. If the actual verification outcome is then in the predicted set, a speculation can be returned immediately, eliminating drafting overhead entirely. We identify three key challenges presented by speculative speculative decoding, and suggest principled methods to solve each. The result is Saguaro, an optimized SSD algorithm. Our implementation is on average 30% faster than optimized speculative decoding baselines and up to 5x faster than autoregressive decoding with open source inference engines.", "authors": ["Tanishq Kumar", "Tri Dao", "Avner May"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-03", "url": "https://arxiv.org/abs/2603.03251", "pdf_url": "https://arxiv.org/pdf/2603.03251v3", "arxiv_id": "2603.03251", "doi": "10.48550/arXiv.2603.03251", "citation_count": 8, "influential_citation_count": 0, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.765}
{"id": "c88adb913b99bb02f7b5e4f49fd9a2d7173101b964065942f0a52a2bf37500dd", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating OpenPangu Inference on NPU via Speculative Decoding", "abstract": "To mitigate the Memory Wall bottleneck encountered by Large Language Models (LLMs) during inference on \\textbf{NPU} hardware, and addressing the scarcity of native support for mainstream speculative decoding algorithms on domestic infrastructure, this study presents an end-to-end speculative inference acceleration scheme for OpenPangu-7B.", "authors": ["Yuntao Dai", "Jing Wu", "Hang Gu", "Teng Wang"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-03", "url": "https://arxiv.org/abs/2603.03383", "pdf_url": "https://arxiv.org/pdf/2603.03383v1", "arxiv_id": "2603.03383", "doi": "10.48550/arXiv.2603.03383", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.495}
{"id": "d8e14c6dab3bff9aa086aa6ad24ffafbf19d0230d1470b6a7f4d7c8a10e82bfc", "sources": ["arxiv", "semantic_scholar"], "title": "Learning to Draft: Adaptive Speculative Decoding with Reinforcement Learning", "abstract": "Speculative decoding accelerates large language model (LLM) inference by using a small draft model to generate candidate tokens for a larger target model to verify. The efficacy of this technique hinges on the trade-off between the time spent on drafting candidates and verifying them. However, current state-of-the-art methods rely on a static time allocation, while recent dynamic approaches optimize for proxy metrics like acceptance length, often neglecting the true time cost and treating the drafting and verification phases in isolation. To address these limitations, we introduce Learning to Draft (LTD), a novel method that directly optimizes for throughput of each draft-and-verify cycle. We formulate the problem as a reinforcement learning environment and train two co-adaptive policies to dynamically coordinate the draft and verification phases. This encourages the policies to adapt to each other and explicitly maximize decoding efficiency. We conducted extensive evaluations on five diverse LLMs and four distinct tasks. Our results show that LTD achieves speedup ratios ranging from 2.24x to 4.32x, outperforming the state-of-the-art method Eagle3 up to 36.4%.", "authors": ["Jiebin Zhang", "Zhenghan Yu", "Liang Wang", "Nan Yang", "Eugene J. Yu", "Zheng Li", "Yifan Song", "Dawei Zhu", "Xingxing Zhang", "Furu Wei", "Sujian Li"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-02", "url": "https://arxiv.org/abs/2603.01639", "pdf_url": "https://arxiv.org/pdf/2603.01639v1", "arxiv_id": "2603.01639", "doi": "10.48550/arXiv.2603.01639", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4939}
{"id": "69cfaa5a10370e697fe383985d3b7d207f7a18f1057f0789847a2a3d3cfd7cd4", "sources": ["arxiv", "semantic_scholar"], "title": "Quasar: Quantized Self-Speculative Acceleration for Rapid Inference via Memory-Efficient Verification", "abstract": "Speculative Decoding (SD) has emerged as a premier technique for accelerating Large Language Model (LLM) inference by decoupling token generation into rapid drafting and parallel verification. While recent advancements in self-speculation and lookahead decoding have successfully minimized drafting overhead, they have shifted the primary performance bottleneck to the verification phase. Since verification requires a full forward pass of the target model, it remains strictly memory-bandwidth bound, fundamentally limiting the maximum achievable speedup.In this paper, we introduce \\textbf{Quasar} (\\textbf{Qua}ntized \\textbf{S}elf-speculative \\textbf{A}cceleration for \\textbf{R}apid Inference), a novel, training-free framework designed to overcome this \"memory wall\" by employing low-bit quantization specifically for the verification stage. Our empirical analysis reveals that while aggressive structural pruning significantly degrades verification accuracy, quantization-based verification preserves the logit distribution with high fidelity while effectively halving memory traffic. Extensive experiments on state-of-the-art models (e.g., OpenPangu and Qwen3) demonstrate that Quasar maintains a speculative acceptance length comparable to full-precision methods while achieving a $1.28\\times$ improvement in end-to-end throughput. Being orthogonal to existing drafting strategies, Quasar offers a generic and efficient pathway to accelerate the verification leg of speculative execution. Code is available at https://github.com/Tom-HG/Quasar.", "authors": ["Guang Huang", "Zeyi Wen"], "categories": ["cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-02", "url": "https://arxiv.org/abs/2603.01399", "pdf_url": "https://arxiv.org/pdf/2603.01399v1", "arxiv_id": "2603.01399", "doi": "10.48550/arXiv.2603.01399", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Tom-HG/Quasar", "venue": "arXiv.org", "quality_score": 0.7632}
{"id": "3fc5790597f01c3d465898874dbabfe2e0af582688e4cb6ae667c8ccf46d118a", "sources": ["arxiv", "semantic_scholar"], "title": "KERV: Kinematic-Rectified Speculative Decoding for Embodied VLA Models", "abstract": "Vision-Language-Action (VLA) models build a token-domain robot control paradigm, yet suffer from low speed. Speculative Decoding (SD) is an optimization strategy that can boost inference speed. Two key issues emerge when integrating VLA and SD: first, SD relies on re-inference to address token errors, which is computationally expensive; second, to mitigate token errors, the acceptance threshold in SD requires careful adjustment. Existing works fail to address the above two issues effectively. Meanwhile, as the bridge between AI and the physical world, existing embodied intelligence has overlooked the application of robotic kinematics. To address these issues, we innovatively combine token-domain VLA models with kinematic-domain prediction for SD, proposing a kinematic-rectified SD framework named KERV. We employ a kinematics-based Kalman Filter to predict actions and compensate for SD errors, avoiding costly re-inference. Moreover, we design a kinematics-based adjustment strategy to dynamically rectify the acceptance threshold, addressing the difficulty of threshold determination. Experimental results across diverse tasks and environments demonstrate that KERV achieves 27%~37% acceleration with nearly no Success Rate loss.", "authors": ["Zihao Zheng", "Zhihao Mao", "Maoliang Li", "Jiayu Chen", "Xinhao Sun", "Zhaobo Zhang", "Donggang Cao", "Hong Mei", "Xiang Chen"], "categories": ["cs.RO", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-02", "url": "https://arxiv.org/abs/2603.01581", "pdf_url": "https://arxiv.org/pdf/2603.01581v2", "arxiv_id": "2603.01581", "doi": "10.48550/arXiv.2603.01581", "citation_count": 6, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4939}
{"id": "3c52ce4295805bbf03880da0a8385ba7b764d0c84ceaa0df7970243c6137230c", "sources": ["arxiv", "semantic_scholar"], "title": "SJD-PV: Speculative Jacobi Decoding with Phrase Verification for Autoregressive Image Generation", "abstract": "Autoregressive (AR) image models have recently demonstrated remarkable generative capability, but their sequential nature results in significant inference latency. Existing training-free acceleration methods typically verify tokens independently, overlooking the strong co-occurrence patterns between adjacent visual tokens. This independence assumption often leads to contextual inconsistency and limits decoding efficiency. In this work, we introduce a novel training-free acceleration framework that performs phrase-level speculative verification, enabling the model to jointly validate multiple correlated tokens within each decoding window. To construct such phrase units, we analyze token co-occurrence statistics from the training corpus and group frequently co-occurring tokens into semantically coherent visual phrases. During inference, the proposed phrase-level verification evaluates aggregated likelihood ratios over each phrase, allowing simultaneous acceptance of multiple tokens while preserving generation quality. Extensive experiments on autoregressive text-to-image generation show that our method significantly reduces the number of function evaluations (NFE) and achieves up to 30% faster decoding without compromising visual fidelity. Our findings reveal that modeling short-range token co-occurrence provides an effective and general principle for accelerating autoregressive inference.", "authors": ["Zhehao Yu", "Baoquan Zhang", "Bingqi Shan", "Xinhao Liu", "Dongliang Zhou", "Guotao Liang", "Guangming Ye", "Yunming Ye"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2026-03-02", "url": "https://arxiv.org/abs/2603.06666", "pdf_url": "https://arxiv.org/pdf/2603.06666v1", "arxiv_id": "2603.06666", "doi": "10.48550/arXiv.2603.06666", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4939}
{"id": "0f9321eafacdebba715e3ab1b3c36c9ef8568ec76d554dec4501d2d6ac5d1092", "sources": ["arxiv", "semantic_scholar"], "title": "LK Losses: Direct Acceptance Rate Optimization for Speculative Decoding", "abstract": "Speculative decoding accelerates autoregressive large language model (LLM) inference by using a lightweight draft model to propose candidate tokens that are then verified in parallel by the target model. The speedup is significantly determined by the acceptance rate, yet standard training minimizes Kullback-Leibler (KL) divergence as a proxy objective. While KL divergence and acceptance rate share the same global optimum, small draft models, having limited capacity, typically converge to suboptimal solutions where minimizing KL does not guarantee maximizing acceptance rate. To address this issue, we propose LK losses, special training objectives that directly target acceptance rate. Comprehensive experiments across four draft architectures and six target models, ranging from 8B to 685B parameters, demonstrate consistent improvements in acceptance metrics across all configurations compared to the standard KL-based training. We evaluate our approach on general, coding and math domains and report gains of up to 8-10% in average acceptance length. LK losses are easy to implement, introduce no computational overhead and can be directly integrated into any existing speculator training framework, making them a compelling alternative to the existing draft training objectives.", "authors": ["Alexander Samarin", "Sergei Krutikov", "Anton Shevtsov", "Sergei Skvortsov", "Filipp Fisin", "Alexander Golubev"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-27", "url": "https://arxiv.org/abs/2602.23881", "pdf_url": "https://arxiv.org/pdf/2602.23881v2", "arxiv_id": "2602.23881", "doi": "10.48550/arXiv.2602.23881", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4904}
{"id": "a1ec30d62dc75806d0aa475f0ef3030088d533404888c920379eaa673c8aa51d", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding Scaling Laws (SDSL): Throughput Optimization Made Simple", "abstract": "Speculative decoding is a technique that uses multiple language models to accelerate infer- ence. Previous works have used an experi- mental approach to optimize the throughput of the inference pipeline, which involves LLM training and can be costly. This study of spec- ulative decoding proposes a theory that ana- lytically connects the key hyperparameters of pre-trained LLMs to the throughput efficiency of a downstream SD-based inference system. The theory allows the prediction of throughput- optimal hyperparameters for the components of an inference system before their pre-training.", "authors": ["Amirhossein Bozorgkhoo", "Igor Molybog"], "categories": ["cs.CL", "cs.IT", "cs.LG"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-02-25", "url": "https://arxiv.org/abs/2603.11053", "pdf_url": "https://arxiv.org/pdf/2603.11053v1", "arxiv_id": "2603.11053", "doi": "10.48550/arXiv.2603.11053", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4881}
{"id": "614457d2f5b19b2841174f379475807c919b6cd0de49039533020870a4263f20", "sources": ["arxiv", "semantic_scholar"], "title": "MineDraft: A Framework for Batch Parallel Speculative Decoding", "abstract": "Speculative decoding (SD) accelerates large language model inference by using a smaller draft model to propose draft tokens that are subsequently verified by a larger target model. However, the performance of standard SD is often limited by the strictly sequential execution of these drafting and verification stages. To address this, this paper proposes MineDraft, a batch parallel speculative decoding (PSD) framework designed to effectively hide drafting latency by overlapping it with verification. Our theoretical analysis shows that PSD is substantially more efficient than standard SD. MineDraft realizes the PSD through a novel batch-parallel design that maintains two batches of requests, overlapping drafting for one batch with verification for the other. Our experimental results show significant improvements of MineDraft in both throughput (up to 75%) and end-to-end latency (up to 39%) over standard SD. Furthermore, we have implemented MineDraft as a plugin for vLLM, demonstrating its practicality for production-ready inference systems.", "authors": ["Zhenwei Tang", "Arun Verma", "Zijian Zhou", "Zhaoxuan Wu", "Alok Prakash", "Daniela Rus", "Bryan Kian Hsiang Low"], "categories": ["cs.CL", "cs.AI", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-24", "url": "https://arxiv.org/abs/2603.18016", "pdf_url": "https://arxiv.org/pdf/2603.18016v2", "arxiv_id": "2603.18016", "doi": "10.48550/arXiv.2603.18016", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.487}
{"id": "3a07305ce7f891698355f95b58fc4c94e26624805574d9a871b4fa5f098db6cd", "sources": ["arxiv", "semantic_scholar"], "title": "Greedy Multi-Path Block Verification for Faster Decoding in Speculative Sampling", "abstract": "The goal of $L$-step speculative decoding is to accelerate autoregressive decoding of a target model by using a cheaper draft model to generate a candidate path of $L$ tokens. Based on a verification algorithm involving target and draft model probabilities, a prefix of the candidate sequence is accepted, and an additional correction token is sampled from a residual distribution to ensure that the final output adheres to the target distribution. While standard speculative decoding uses a verification algorithm which is independent at each token on the path, a recent extension called block verification uses a joint condition involving all sampled on-path probabilities. Block verification (BV) was shown to be optimal over all verification algorithms which use only on-path probabilities, improving on standard speculative decoding. In this work, we first show that block verification is optimal even over verification algorithms that use off-path probabilities, by constructing an information-agnostic linear program (LP). Further, we can extend our LP to the setting where the draft model samples multiple candidate paths, and use it to construct a natural class of multi-path block verification generalizations. While computing the optimal algorithm in this class is not tractable, by considering a stricter class of greedy algorithms, we can formulate an efficient method called greedy multi-path block verification (GBV). Empirically, GBV can improve block efficiency by over 30% and reduce decoding walltimes by over 15% relative to BV. On Llama-3 70B, GBV can improve the end-to-end decoding throughput over SOTA multi-path verification methods by more than 15%.", "authors": ["Rahul Thomas", "Arka Pal"], "categories": ["cs.IT", "cs.LG"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-02-18", "url": "https://arxiv.org/abs/2602.16961", "pdf_url": "https://arxiv.org/pdf/2602.16961v1", "arxiv_id": "2602.16961", "doi": "10.48550/arXiv.2602.16961", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4801}
{"id": "7ff93673f3d0de80cf9a5d268e83d2cfd323d63015bd65be696aa45af8421926", "sources": ["arxiv", "semantic_scholar"], "title": "Privacy-Aware Split Inference with Speculative Decoding for Large Language Models over Wide-Area Networks", "abstract": "We present a practical system for privacy-aware large language model (LLM) inference that splits a transformer between a trusted local GPU and an untrusted cloud GPU, communicating only intermediate activations over the network. Our system addresses the unique challenges of autoregressive LLM decoding over high-latency wide-area networks (WANs), contributing: (1) an asymmetric layer split where embedding and unembedding layers remain local, ensuring raw tokens never leave the trusted device; (2) the first application of lookahead decoding to split inference over WANs, amortizing network round-trip latency across multiple tokens per iteration; (3) an empirical inversion attack evaluation showing that split depth provides a tunable privacy-performance tradeoff -- an attacker can recover ~59%% of tokens at a 2-layer split but only ~35%% at an 8-layer split, with minimal throughput impact; (4) ablation experiments showing that n-gram speculation accepts 1.2-1.3 tokens per decoding step on average (peak of 7 observed on code), with acceptance rates consistent across model scales; (5) formal verification that lookahead decoding produces token-identical output to sequential decoding under greedy argmax, with zero quality degradation; and (6) scaling validation on Mistral NeMo 12B (40 layers), demonstrating that the system generalizes to larger models with only 4.9 GB local VRAM and matching 7B throughput. Evaluated on Mistral 7B and NeMo 12B over a ~80ms WAN link, our system achieves 8.7-9.3 tok/s (7B) and 7.8-8.7 tok/s (12B) with lookahead decoding, with an RTT decomposition model (validated at <6.2%% cross-validation error) projecting 15-19 tok/s at 20ms RTT.", "authors": ["Michael Cunningham"], "categories": ["cs.CR", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-18", "url": "https://arxiv.org/abs/2602.16760", "pdf_url": "https://arxiv.org/pdf/2602.16760v1", "arxiv_id": "2602.16760", "doi": "10.48550/arXiv.2602.16760", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4801}
{"id": "f8028e71a8ea62bb19e7f8fc6c259b427149351e6c3097da0bd5378774eca8b0", "sources": ["arxiv", "semantic_scholar"], "title": "MoE-Spec: Expert Budgeting for Efficient Speculative Decoding", "abstract": "Speculative decoding accelerates Large Language Model (LLM) inference by verifying multiple drafted tokens in parallel. However, for Mixture-of-Experts (MoE) models, this parallelism introduces a severe bottleneck: large draft trees activate many unique experts, significantly increasing memory pressure and diminishing speedups from speculative decoding relative to autoregressive decoding. Prior methods reduce speculation depth when MoE verification becomes expensive. We propose MoE-Spec, a training-free verification-time expert budgeting method that decouples speculation depth from memory cost by enforcing a fixed expert capacity limit at each layer, loading only the experts that contribute most to verification and dropping the long tail of rarely used experts that drive bandwidth overhead. Experiments across multiple model scales and datasets show that this method yields 10--30\\% higher throughput than state-of-the-art speculative decoding baselines (EAGLE-3) at comparable quality, with flexibility to trade accuracy for further latency reductions through tighter budgets.", "authors": ["Bradley McDanel", "Steven Li", "Sruthikesh Surineni", "Harshit Khaitan"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-17", "url": "https://arxiv.org/abs/2602.16052", "pdf_url": "https://arxiv.org/pdf/2602.16052v1", "arxiv_id": "2602.16052", "doi": "10.48550/arXiv.2602.16052", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.479}
{"id": "dae9fa5aba7ff05350cfa8cd1cf97257baf7db1b7dfd4ea5f2a7e23946ab3084", "sources": ["arxiv", "semantic_scholar"], "title": "Sparrow: Text-Anchored Window Attention with Visual-Semantic Glimpsing for Speculative Decoding in Video LLMs", "abstract": "Although speculative decoding is widely used to accelerate Vision-Language Models (VLMs) inference, it faces severe performance collapse when applied to Video Large Language Models (Vid-LLMs). The draft model typically falls into the trap of attention dilution and negative visual gain due to key-value cache explosion and context window mismatches. We observe a visual semantic internalization phenomenon in Vid-LLMs, indicating that critical visual semantics are implicitly encoded into text hidden states during deep-layer interactions, which renders raw visual inputs structurally redundant during deep inference. To address this, we propose the Sparrow framework, which first utilizes visually-aware text-anchored window attention via hidden state reuse to fully offload visual computation to the target model, and leverages intermediate-layer visual state bridging to train the draft model with semantic-rich intermediate states, thereby filtering out low-level visual noise. Additionally, a multi-token prediction strategy is introduced to bridge the training-inference distribution shift. Experiments show that Sparrow achieves an average speedup of 2.82x even with 25k visual tokens, effectively resolving the performance degradation in long sequences and offering a practical solution for real-time long video tasks.", "authors": ["Libo Zhang", "Zhaoning Zhang", "Wangyang Hong", "Peng Qiao", "Dongsheng Li"], "categories": ["cs.CV", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-17", "url": "https://arxiv.org/abs/2602.15318", "pdf_url": "https://arxiv.org/pdf/2602.15318v1", "arxiv_id": "2602.15318", "doi": "10.48550/arXiv.2602.15318", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.479}
{"id": "d0581e9dc5f171405f2f0f7344445b8c654db77136e29c8ed6317ec5451b93e4", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding with a Speculative Vocabulary", "abstract": "Speculative decoding has rapidly emerged as a leading approach for accelerating language model (LM) inference, as it offers substantial speedups while yielding identical outputs. This relies upon a small draft model, tasked with predicting the outputs of the target model. State-of-the-art speculative decoding methods use a draft model consisting of a single decoder layer and output embedding matrix, with the latter dominating drafting time for the latest LMs. Recent work has sought to address this output distribution bottleneck by reducing the vocabulary of the draft model. Although this can improve throughput, it compromises speculation effectiveness when the target token is out-of-vocabulary. In this paper, we argue for vocabulary speculation as an alternative to a reduced vocabulary. We propose SpecVocab, an efficient and effective method that selects a vocabulary subset per decoding step. Across a variety of tasks, we demonstrate that SpecVocab can achieve a higher acceptance length than state-of-the-art speculative decoding approach, EAGLE-3. Notably, this yields up to an 8.1% increase in average throughput over EAGLE-3.", "authors": ["Miles Williams", "Young D. Kwon", "Rui Li", "Alexandros Kouris", "Stylianos I. Venieris"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-14", "url": "https://arxiv.org/abs/2602.13836", "pdf_url": "https://arxiv.org/pdf/2602.13836v1", "arxiv_id": "2602.13836", "doi": "10.48550/arXiv.2602.13836", "citation_count": 2, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4755}
{"id": "065d91c968177c81423660a305f9818dde4293c39bac8fe49ed564c09c76f506", "sources": ["arxiv", "semantic_scholar"], "title": "MoE-SpAc: Efficient MoE Inference Based on Speculative Activation Utility in Heterogeneous Edge Scenarios", "abstract": "Mixture-of-Experts (MoE) models enable scalable performance but face severe memory constraints on edge devices. Existing offloading strategies struggle with I/O bottlenecks due to the dynamic, low-information nature of autoregressive expert activation. In this paper, we propose to repurpose Speculative Decoding (SD) not merely as a compute accelerator, but as an informative lookahead sensor for memory management, supported by our theoretical and empirical analyses. Hence, we introduce MoE-SpAc, an MoE inference framework that integrates a Speculative Utility Estimator to track expert demand, a Heterogeneous Workload Balancer to dynamically partition computation via online integer optimization, and an Asynchronous Execution Engine to unify the prefetching and eviction in the same utility space. Extensive experiments on seven benchmarks demonstrate that MoE-SpAc achieves a 42% improvement in TPS over the SOTA SD-based baseline, and an average 4.04x speedup over all standard baselines. Code is available at https://github.com/lshAlgorithm/MoE-SpAc .", "authors": ["Shuhuai Li", "Jianghao Lin", "Dongdong Ge", "Yinyu Ye"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-12", "url": "https://arxiv.org/abs/2603.09983", "pdf_url": "https://arxiv.org/pdf/2603.09983v1", "arxiv_id": "2603.09983", "doi": "10.48550/arXiv.2603.09983", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/lshAlgorithm/MoE-SpAc", "venue": "arXiv.org", "quality_score": 0.7314}
{"id": "07d03f57bc1ae6cdd9f65c6895083a47037543009da6a609916b37ce9f736c39", "sources": ["arxiv", "semantic_scholar"], "title": "Training-free Dropout Sampling for Semantic Token Acceptance in Speculative Decoding", "abstract": "Speculative decoding accelerates large language model inference by proposing tokens with a lightweight draft model and selectively accepting them using a target model. This work introduces DropMatch, a novel approach that matches draft tokens to the predictive distribution of the target model via Monte Carlo dropout applied exclusively to the LM head, enabling sampling-based acceptance decisions. By generating multiple decoding paths, our method forms an empirical token distribution against which draft tokens are evaluated for consistency. This acceptance mechanism enables the model to adaptively control the size of decoding paths under an appropriate dropout probability, preventing substantial distortion of the target model predictive distribution. The proposed method operates in a training-free, data-free, and calibration-free manner, requires no architectural modification to pretrained models, and can be orthogonally integrated with a wide range of existing speculative decoding and inference acceleration techniques. Experiments across multiple benchmarks demonstrate that our approach increases acceptance length while maintaining competitive task performance, yielding inference speedups ranging from 1.09x to 1.33x over the standard baseline, and up to an additional 1.09x speedup when applied on top of EAGLE3.", "authors": ["Jeongtae Lee", "Minjung Jo", "Hyunjoon Jeong", "Gunho Park", "Sunghyeon Woo", "Joonghoon Kim", "Se Jung Kwon", "Dongsoo Lee"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-11", "url": "https://arxiv.org/abs/2603.03333", "pdf_url": "https://arxiv.org/pdf/2603.03333v1", "arxiv_id": "2603.03333", "doi": "10.48550/arXiv.2603.03333", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4721}
{"id": "44af42eb5fe0a5d159f7cdf5e9750863cf446e41b32b72954094fcb96ff61429", "sources": ["arxiv", "semantic_scholar"], "title": "StreamServe: Adaptive Speculative Flows for Low-Latency Disaggregated LLM Serving", "abstract": "Efficient LLM serving must balance throughput and latency across diverse, bursty workloads. We introduce StreamServe, a disaggregated prefill decode serving architecture that combines metric aware routing across compute lanes with adaptive speculative decoding that tunes speculation depth online from runtime signals. StreamServe comprises four components: StreamScheduler for request orchestration, FlowGuard for multi signal routing, PipeServe Engine for disaggregated prefill decode execution on multi GPU, and SpecuStream for runtime adaptive speculation. We evaluate StreamServe on four benchmarks ALPACA, GSM8K, HUMANEVAL, and SUM with 80 queries each and 320 total using 4 A800 40GB GPUs configured as two stream pairs. Across these workloads, StreamServe reduces latency by 11 to 18 times relative to tensor parallel vLLM baselines and reaches throughput up to 2235 tokens per second on summarization tasks. Time per output token remains stable across configurations, indicating that the gains arise from architectural efficiency rather than token quality degradation. Although evaluated on a single node 4 GPU setup, these results suggest that jointly adapting routing and speculation within a disaggregated framework creates a distinct operating regime for LLM inference.", "authors": ["Satyam Kumar", "Arpit Singh Gautam", "Kailash Talreja", "Saurabh Jha"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-11", "url": "https://arxiv.org/abs/2604.09562", "pdf_url": "https://arxiv.org/pdf/2604.09562v1", "arxiv_id": "2604.09562", "doi": "10.48550/arXiv.2604.09562", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4721}
{"id": "f41985dc055efb96a1291d4e6a3cced3615908215c8532e7e515d27ae5eaad48", "sources": ["arxiv", "semantic_scholar"], "title": "SPEED-Bench: A Unified and Diverse Benchmark for Speculative Decoding", "abstract": "Speculative Decoding (SD) has emerged as a critical technique for accelerating Large Language Model (LLM) inference. Unlike deterministic system optimizations, SD performance is inherently data-dependent, meaning that diverse and representative workloads are essential for accurately measuring its effectiveness. Existing benchmarks suffer from limited task diversity, inadequate support for throughput-oriented evaluation, and a reliance on high-level implementations that fail to reflect production environments. To address this, we introduce SPEED-Bench, a comprehensive suite designed to standardize SD evaluation across diverse semantic domains and realistic serving regimes. SPEED-Bench offers a carefully curated Qualitative data split, selected by prioritizing semantic diversity across the data samples. Additionally, it includes a Throughput data split, allowing speedup evaluation across a range of concurrencies, from latency-sensitive low-batch settings to throughput-oriented high-load scenarios. By integrating with production engines like vLLM and TensorRT-LLM, SPEED-Bench allows practitioners to analyze system behaviors often masked by other benchmarks. We highlight this by quantifying how synthetic inputs overestimate real-world throughput, identifying batch-size dependent optimal draft lengths and biases in low-diversity data, and analyzing the caveats of vocabulary pruning in state-of-the-art drafters. We release SPEED-Bench to establish a unified evaluation standard for practical comparisons of SD algorithms.", "authors": ["Talor Abramovich", "Maor Ashkenazi", "Izzy Putterman", "Benjamin Chislett", "Tiyasa Mitra", "Bita Darvish Rouhani", "Ran Zilberstein", "Yonatan Geifman"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-10", "url": "https://arxiv.org/abs/2604.09557", "pdf_url": "https://arxiv.org/pdf/2604.09557v2", "arxiv_id": "2604.09557", "doi": "10.48550/arXiv.2604.09557", "citation_count": 2, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4709}
{"id": "91d76cd55b98f485a8d9d828dc0284cb4bc1d2f4892450aff3fc42d234b84fbe", "sources": ["arxiv", "semantic_scholar"], "title": "Benchmarking the Energy Savings with Speculative Decoding Strategies", "abstract": "Speculative decoding has emerged as an effective method to reduce latency and inference cost of LLM inferences. However, there has been inadequate attention towards the energy requirements of these models. To address this gap, this paper presents a comprehensive survey of energy requirements of speculative decoding strategies, with detailed analysis on how various factors -- model size and family, speculative decoding strategies, and dataset characteristics -- influence the energy optimizations.", "authors": ["Rohit Dutta", "Paramita Koley", "Soham Poddar", "Janardan Misra", "Sanjay Podder", "Naveen Balani", "Saptarshi Ghosh", "Niloy Ganguly"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-09", "url": "https://arxiv.org/abs/2602.09113", "pdf_url": "https://arxiv.org/pdf/2602.09113v1", "arxiv_id": "2602.09113", "doi": "10.48550/arXiv.2602.09113", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Conference of the European Chapter of the Association for Computational Linguistics", "quality_score": 0.4698}
{"id": "754971f7a2e1001cff3b1343c3a7c0299b9876bd6e6f05f2f92eea981d717fb1", "sources": ["arxiv", "semantic_scholar"], "title": "Compiler-Assisted Speculative Sampling for Accelerated LLM Inference on Heterogeneous Edge Devices", "abstract": "LLM deployment on resource-constrained edge devices faces severe latency constraints, particularly in real-time applications where delayed responses can compromise safety or usability. Among many approaches to mitigate the inefficiencies of sequential token-by-token generation, Speculative Decoding (SD) has emerged as a promising technique. However, SD at the edge is hindered by two major challenges: (1) integrating SD into a compiler-based workflow without sacrificing performance or programmability, and (2) exploiting the heterogeneous compute resources of modern SoCs through carefully designed partitioning strategies. This work addresses these challenges by using an analytical cost model that explores heterogeneous hardware configurations and guides coarse-grained partitioning of LLM subgraphs, particularly with edge-typical short input sequence lengths. The cost model predicts when speculative sampling and heterogeneous execution are jointly beneficial and is validated on an edge device featuring a hexacore Cortex-A CPU and a Mali GPU, revealing up to 1.68$\\times$ speedup for translation tasks, closely matching analytic expectations.", "authors": ["Alejandro Ruiz y Mesa", "Guilherme Korol", "Moritz Riesterer", "João Paulo Cardoso de Lima", "Jeronimo Castrillon"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-08", "url": "https://arxiv.org/abs/2602.08060", "pdf_url": "https://arxiv.org/pdf/2602.08060v2", "arxiv_id": "2602.08060", "doi": "10.48550/arXiv.2602.08060", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4686}
{"id": "5e54c53680c7efd6fbb43b4542d5b82996063a9b9f92b6f1b4aba13739fa942f", "sources": ["arxiv", "semantic_scholar"], "title": "Vegas: Self-Speculative Decoding with Verification-Guided Sparse Attention", "abstract": "Long-context large language model (LLM) inference has become the norm for today's AI applications. However, it is severely bottlenecked by the increasing memory demands of its KV cache. Previous works have shown that self-speculative decoding with sparse attention, where tokens are drafted using a subset of the KV cache and verified in parallel against the full KV cache, speeds up inference in a lossless manner. However, they rely on a standalone KV selection algorithm to select the KV entries used for drafting and overlook the fact that the criticality of each KV entry is inherently computed during verification. In this paper, we propose Vegas, a self-speculative decoding method with verification-guided sparse attention. Vegas identifies critical KV cache entries as a byproduct of verification and computes attention only over these entries when drafting subsequent tokens. This not only improves the draft token acceptance rate but also incurs low KV selection overhead, thereby improving decoding throughput. Vegas achieves a 1.25$\\times$-2.81$\\times$ speedup in decoding throughput over default vLLM and a 1.15$\\times$-1.29$\\times$ speedup over state-of-the-art sparse attention-based self-speculative decoding methods. Our code is available at https://github.com/platformxlab/vegas.", "authors": ["Yikang Yue", "Yuqi Xue", "Jian Huang"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-06", "url": "https://arxiv.org/abs/2602.07223", "pdf_url": "https://arxiv.org/pdf/2602.07223v2", "arxiv_id": "2602.07223", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/platformxlab/vegas", "venue": null, "quality_score": 0.5511}
{"id": "092017e2097c57adce599a86e9bb7123673ae9cc3667d737dfd01c581e146d1b", "sources": ["arxiv", "semantic_scholar"], "title": "Anchored Decoding: Provably Reducing Copyright Risk for Any Language Model", "abstract": "Language models (LMs) tend to memorize portions of their training data and emit verbatim spans. When the underlying sources are sensitive or copyright-protected, such reproduction raises issues of consent and compensation for creators and compliance risks for developers. We propose Anchored Decoding, a plug-and-play inference-time method for suppressing verbatim copying: it enables decoding from any risky LM trained on mixed-license data by keeping generation in bounded proximity to a permissively trained safe LM. Anchored Decoding adaptively allocates a user-chosen information budget over the generation trajectory and enforces per-step constraints that yield a sequence-level guarantee, enabling a tunable risk-utility trade-off. To make Anchored Decoding practically useful, we introduce a new permissively trained safe model (TinyComma 1.8B), as well as Anchored$_{\\mathrm{Byte}}$ Decoding, a byte-level variant of our method that enables cross-vocabulary fusion via the ByteSampler framework (Hayase et al., 2025). Across six model pairs on long-form metrics for copying risk and utility, Anchored and Anchored$_{\\mathrm{Byte}}$ Decoding define a new Pareto frontier, preserving near-original fluency and factuality while closing up to 75% of the measurable copying gap between the risky baseline and a safe reference, at a modest inference overhead.", "authors": ["Jacqueline He", "Jonathan Hayase", "Wen-tau Yih", "Sewoong Oh", "Luke Zettlemoyer", "Pang Wei Koh"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-06", "url": "https://arxiv.org/abs/2602.07120", "pdf_url": "https://arxiv.org/pdf/2602.07120v2", "arxiv_id": "2602.07120", "doi": "10.48550/arXiv.2602.07120", "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/jacqueline-he/anchored-decoding", "venue": "arXiv.org", "quality_score": 0.7207}
{"id": "fa131ba8f9cc9b5f3c200aadceabdbf6675f23d692788be68da85b430825700d", "sources": ["arxiv", "semantic_scholar"], "title": "DFlash: Block Diffusion for Flash Speculative Decoding", "abstract": "Autoregressive large language models (LLMs) deliver strong performance but require inherently sequential decoding, leading to high inference latency and poor GPU utilization. Speculative decoding mitigates this bottleneck by using a fast draft model whose outputs are verified in parallel by the target LLM; however, existing methods still rely on autoregressive drafting, which remains sequential and limits practical speedups. Diffusion LLMs offer a promising alternative by enabling parallel generation, but current diffusion models typically underperform compared with autoregressive models. In this paper, we introduce DFlash, a speculative decoding framework that employs a lightweight block diffusion model for parallel drafting. By generating draft tokens in a single forward pass and conditioning the draft model on context features extracted from the target model, DFlash enables efficient drafting with high-quality outputs and higher acceptance rates. Experiments show that DFlash achieves over 6x lossless acceleration across a range of models and tasks, delivering up to 2.5x higher speedup than the state-of-the-art speculative decoding method EAGLE-3.", "authors": ["Jian Chen", "Yesheng Liang", "Zhijian Liu"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-05", "url": "https://arxiv.org/abs/2602.06036", "pdf_url": "https://arxiv.org/pdf/2602.06036v2", "arxiv_id": "2602.06036", "doi": "10.48550/arXiv.2602.06036", "citation_count": 33, "influential_citation_count": 14, "has_code": true, "code_url": "https://github.com/z-lab/dflash", "venue": "arXiv.org", "quality_score": 0.719}
{"id": "efd442573bfe236afe3518764162863d47ee0a03543753f6fe9a95538d2c2342", "sources": ["arxiv", "semantic_scholar"], "title": "SDFP: Speculative Decoding with FIT-Pruned Models for Training-Free and Plug-and-Play LLM Acceleration", "abstract": "Large language models (LLMs) underpin interactive multimedia applications such as captioning, retrieval, recommendation, and creative content generation, yet their autoregressive decoding incurs substantial latency. Speculative decoding reduces latency using a lightweight draft model, but deployment is often limited by the cost and complexity of acquiring, tuning, and maintaining an effective draft model. Recent approaches usually require auxiliary training or specialization, and even training-free methods incur costly search or optimization. We propose SDFP, a fully training-free and plug-and-play framework that builds the draft model via Fisher Information Trace (FIT)-based layer pruning of a given LLM. Using layer sensitivity as a proxy for output perturbation, SDFP removes low-impact layers to obtain a compact draft while preserving compatibility with the original model for standard speculative verification. SDFP needs no additional training, hyperparameter tuning, or separately maintained drafts, enabling rapid, deployment-friendly draft construction. Across benchmarks, SDFP delivers 1.32x-1.5x decoding speedup without altering the target model's output distribution, supporting low-latency multimedia applications.", "authors": ["Hanyu Wei", "Zunhai Su", "Peng Lu", "Chao Li", "Spandan Tiwari", "Ashish Sirasao", "Yuhan Dong"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-05", "url": "https://arxiv.org/abs/2602.05499", "pdf_url": "https://arxiv.org/pdf/2602.05499v1", "arxiv_id": "2602.05499", "doi": "10.48550/arXiv.2602.05499", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4652}
{"id": "b4513f6f0d82b39b39430b8c68903a284d0eddfbd3884947b569a90266aeeb4a", "sources": ["arxiv", "semantic_scholar"], "title": "Variational Speculative Decoding: Rethinking Draft Training from Token Likelihood to Sequence Acceptance", "abstract": "Speculative decoding accelerates inference for (M)LLMs, yet a training-decoding discrepancy persists: while existing methods optimize single greedy trajectories, decoding involves verifying and ranking multiple sampled draft paths. We propose Variational Speculative Decoding (VSD), formulating draft training as variational inference over latent proposals (draft paths). VSD maximizes the marginal probability of target-model acceptance, yielding an ELBO that promotes high-quality latent proposals while minimizing divergence from the target distribution. To enhance quality and reduce variance, we incorporate a path-level utility and optimize via an Expectation-Maximization procedure. The E-step draws Monte Carlo samples from an oracle-filtered posterior, while the M-step maximizes weighted likelihood using Adaptive Rejection Weighting (ARW) and Confidence-Aware Regularization (CAR). Theoretical analysis confirms that VSD increases expected acceptance length and speedup. Extensive experiments across LLMs and MLLMs show that VSD achieves up to a 9.6% speedup over EAGLE-3 and 7.9% over ViSpec, significantly improving decoding efficiency.", "authors": ["Xiandong Zou", "Jianshu Li", "Jing Huang", "Pan Zhou"], "categories": ["cs.LG", "cs.AI", "math.PR"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-02-05", "url": "https://arxiv.org/abs/2602.05774", "pdf_url": "https://arxiv.org/pdf/2602.05774v4", "arxiv_id": "2602.05774", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.296}
{"id": "1f5da4d1db6c0e6343c3dd91b420c3358309f3bca24067be9d30d3d1bbaad33f", "sources": ["arxiv", "semantic_scholar"], "title": "TIDE: Temporal Incremental Draft Engine for Self-Improving LLM Inference", "abstract": "Speculative decoding can substantially accelerate LLM inference, but realizing its benefits in practice is challenging due to evolving workloads and system-level constraints. We present TIDE (Temporal Incremental Draft Engine), a serving-engine-native framework that integrates online draft adaptation directly into high-performance LLM inference systems. TIDE reuses target model hidden states generated during inference as training signals, enabling zero-overhead draft adaptation without reloading the target model, and employs adaptive runtime control to activate speculation and training only when beneficial. TIDE exploits heterogeneous clusters by mapping decoupled inference and training to appropriate GPU classes. Across diverse real-world workloads, TIDE achieves up to 1.15x throughput improvement over static speculative decoding while reducing draft training time by 1.67x compared to approaches that recompute training signals.", "authors": ["Jiyoung Park", "Hankyu Jang", "Changseok Song", "Wookeun Jung"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-05", "url": "https://arxiv.org/abs/2602.05145", "pdf_url": "https://arxiv.org/pdf/2602.05145v1", "arxiv_id": "2602.05145", "doi": "10.48550/arXiv.2602.05145", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4652}
{"id": "b0f2c8b875e126495561622b98c904e33b28b54775b4984e682f473f252452aa", "sources": ["arxiv", "semantic_scholar"], "title": "LycheeDecode: Accelerating Long-Context LLM Inference via Hybrid-Head Sparse Decoding", "abstract": "The proliferation of long-context large language models (LLMs) exposes a key bottleneck: the rapidly expanding key-value cache during decoding, which imposes heavy memory and latency costs. While recent approaches attempt to alleviate this by sharing a single set of crucial tokens across layers, such coarse-grained sharing undermines model performance by neglecting the functional diversity of attention heads. To address this, we propose LycheeDecode, an efficient decoding method centered on a fine-grained hybrid-head attention mechanism that employs a hardware-efficient top-k selection strategy. Specifically, the novel HardKuma-based mechanism partitions attention heads into a small subset of retrieval heads that dynamically identify crucial tokens and a majority of sparse heads that reuse them for efficient computation. Through extensive experiments on leading models like Llama3 and Qwen3 across diverse benchmarks for long-context understanding (e.g., LongBench, RULER) and complex reasoning (e.g., AIME24, OlympiadBench), we demonstrate that LycheeDecode achieves generative quality comparable to, and at times surpassing even the full-attention baseline. Crucially, this is accomplished with up to a 2.7x speedup at a 128K context length. By preserving the functional diversity of attention heads, our fine-grained strategy overcomes the performance bottlenecks of existing methods, providing a powerful and validated pathway to both efficient and high-quality long-context LLM inference.", "authors": ["Gang Lin", "Dongfang Li", "Zhuoen Chen", "Yukun Shi", "Xuhui Chen", "Baotian Hu", "Min Zhang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-04", "url": "https://arxiv.org/abs/2602.04541", "pdf_url": "https://arxiv.org/pdf/2602.04541v1", "arxiv_id": "2602.04541", "doi": "10.48550/arXiv.2602.04541", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4641}
{"id": "56ab42c904e0b27e7cd5a547cc0b176f6ea976631f9580d89704d71bd3bfd8dc", "sources": ["arxiv", "semantic_scholar"], "title": "Beyond Tokens: Semantic-Aware Speculative Decoding for Efficient Inference by Probing Internal States", "abstract": "Large Language Models (LLMs) achieve strong performance across many tasks but suffer from high inference latency due to autoregressive decoding. The issue is exacerbated in Large Reasoning Models (LRMs), which generate lengthy chains of thought. While speculative decoding accelerates inference by drafting and verifying multiple tokens in parallel, existing methods operate at the token level and ignore semantic equivalence (i.e., different token sequences expressing the same meaning), leading to inefficient rejections. We propose SemanticSpec, a semantic-aware speculative decoding framework that verifies entire semantic sequences instead of tokens. SemanticSpec introduces a semantic probability estimation mechanism that probes the model's internal hidden states to assess the likelihood of generating sequences with specific meanings. Experiments on four benchmarks show that SemanticSpec achieves up to 2.7x speedup on DeepSeekR1-32B and 2.1x on QwQ-32B, consistently outperforming token-level and sequence-level baselines in both efficiency and effectiveness.", "authors": ["Ximing Dong", "Shaowei Wang", "Dayi Lin", "Boyuan Chen", "Ahmed E. Hassan"], "categories": ["cs.CL", "cs.PF"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-03", "url": "https://arxiv.org/abs/2602.03708", "pdf_url": "https://arxiv.org/pdf/2602.03708v2", "arxiv_id": "2602.03708", "doi": "10.48550/arXiv.2602.03708", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4629}
{"id": "61cc828d9569c50d6d74243c810300b60afc6d63d8278674a18a2478358f09ac", "sources": ["arxiv", "semantic_scholar"], "title": "Make Every Draft Count: Hidden State based Speculative Decoding", "abstract": "Speculative decoding has emerged as a pivotal technique to accelerate LLM inference by employing a lightweight draft model to generate candidate tokens that are subsequently verified by the target model in parallel. However, while this paradigm successfully increases the arithmetic intensity of memory-bound inference, it causes significant compute inefficiency: the majority of draft tokens fail verification and are discarded, resulting in waste of computation. Motivated by the goal of recollecting this wasted computation, we propose a novel system that transforms discarded drafts into reusable tokens. Our key insight is to perform auto-regressive prediction at the hidden states level and postpone the integrating token information after the hidden states generation, so the draft hidden states are not contaminated by incorrect tokens, enabling hidden state reuse. To implement such a system, first we introduce a draft model architecture based on auto-regressive hidden states, which preserves richer semantics than token-based drafters to facilitate draft repurposing. Second, we design an efficient token information injection mechanism that leverages our specialized draft model to construct high-quality draft token trees and enables resampling tokens from verification failures. Third, we eliminate the overhead hidden in our design to further maximize hardware utilization. We conducted extensive evaluations against various baselines, demonstrating up to a 3.3x speedup against standard speculative decoding.", "authors": ["Yuetao Chen", "Xuliang Wang", "Xinzhou Zheng", "Ming Li", "Peng Wang", "Hong Xu"], "categories": ["cs.CL", "cs.AI", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-02", "url": "https://arxiv.org/abs/2602.21224", "pdf_url": "https://arxiv.org/pdf/2602.21224v1", "arxiv_id": "2602.21224", "doi": "10.48550/arXiv.2602.21224", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4618}
{"id": "b4f398f7a2d4c0ae59e38cc0ea5a6ce47816c77d0a7e11e49ef64e1e9ad1fe0c", "sources": ["arxiv", "semantic_scholar"], "title": "PRISM: Parametrically Refactoring Inference for Speculative Sampling Draft Models", "abstract": "Large Language Models (LLMs), constrained by their auto-regressive nature, suffer from slow decoding. Speculative decoding methods have emerged as a promising solution to accelerate LLM decoding, attracting attention from both systems and AI research communities. Recently, the pursuit of better draft quality has driven a trend toward parametrically larger draft models, which inevitably introduces substantial computational overhead. While existing work attempts to balance the trade-off between prediction accuracy and compute latency, we address this fundamental dilemma through architectural innovation. We propose PRISM, which disaggregates the computation of each predictive step across different parameter sets, refactoring the computational pathways of draft models to successfully decouple model capacity from inference cost. Through extensive experiments, we demonstrate that PRISM outperforms all existing draft architectures, achieving exceptional acceptance lengths while maintaining minimal draft latency for superior end-to-end speedup. We also re-examine scaling laws with PRISM, revealing that PRISM scales more effectively with expanding data volumes than other draft architectures. Through rigorous and fair comparison, we show that PRISM boosts the decoding throughput of an already highly optimized inference engine by more than 2.6x.", "authors": ["Xuliang Wang", "Yuetao Chen", "Maochan Zhen", "Fang Liu", "Xinzhou Zheng", "Xingwu Liu", "Hong Xu", "Ming Li"], "categories": ["cs.AI", "cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-02", "url": "https://arxiv.org/abs/2602.01762", "pdf_url": "https://arxiv.org/pdf/2602.01762v1", "arxiv_id": "2602.01762", "doi": "10.48550/arXiv.2602.01762", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4618}
{"id": "769e41cd1d22874d58fb8564839543c8a6179b06ca1fbc86a685124cb1e6449f", "sources": ["arxiv", "semantic_scholar"], "title": "PACER: Blockwise Pre-verification for Speculative Decoding with Adaptive Length", "abstract": "Speculative decoding (SD) is a powerful technique for accelerating the inference process of large language models (LLMs) without sacrificing accuracy. Typically, SD employs a small draft model to generate a fixed number of draft tokens, which are then verified in parallel by the target model. However, our experiments reveal that the optimal draft length varies significantly across different decoding steps. This variation suggests that using a fixed draft length limits the potential for further improvements in decoding speed. To address this challenge, we propose Pacer, a novel approach that dynamically controls draft length using a lightweight, trainable pre-verification layer. This layer pre-verifies draft tokens blockwise before they are sent to the target model, allowing the draft model to stop token generation if the blockwise pre-verification fails. We implement Pacer on multiple SD model pairs and evaluate its performance across various benchmarks. Our results demonstrate that Pacer achieves up to 2.66x Speedup over autoregressive decoding and consistently outperforms standard speculative decoding. Furthermore, when integrated with Ouroboros, Pacer attains up to 3.09x Speedup.", "authors": ["Situo Zhang", "Yifan Zhang", "Zichen Zhu", "Hankun Wang", "Da Ma", "Danyang Zhang", "Lu Chen", "Kai Yu"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-02-01", "url": "https://arxiv.org/abs/2602.01274", "pdf_url": "https://arxiv.org/pdf/2602.01274v1", "arxiv_id": "2602.01274", "doi": "10.48550/arXiv.2602.01274", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4606}
{"id": "89d1e1fc975cced6084cc06ce3ffba8e4ba20a8b94bfd5fd49714bbf63298b44", "sources": ["arxiv", "semantic_scholar"], "title": "SAGE: Accelerating Vision-Language Models via Entropy-Guided Adaptive Speculative Decoding", "abstract": "Speculative decoding has emerged as a promising approach to accelerate inference in vision-language models (VLMs) by enabling parallel verification of multiple draft tokens. However, existing methods rely on static tree structures that remain fixed throughout the decoding process, failing to adapt to the varying prediction difficulty across generation steps. This leads to suboptimal acceptance lengths and limited speedup. In this paper, we propose SAGE, a novel framework that dynamically adjusts the speculation tree structure based on real-time prediction uncertainty. Our key insight is that output entropy serves as a natural confidence indicator with strong temporal correlation across decoding steps. SAGE constructs deeper-narrower trees for high-confidence predictions to maximize speculation depth, and shallower-wider trees for uncertain predictions to diversify exploration. SAGE improves acceptance lengths and achieves faster acceleration compared to static tree baselines. Experiments on multiple benchmarks demonstrate the effectiveness of SAGE: without any loss in output quality, it delivers up to $3.36\\times$ decoding speedup for LLaVA-OneVision-72B and $3.18\\times$ for Qwen2.5-VL-72B.", "authors": ["Yujia Tong", "Tian Zhang", "Yunyang Wan", "Kaiwei Lin", "Jingling Yuan", "Chuang Hu"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-31", "url": "https://arxiv.org/abs/2602.00523", "pdf_url": "https://arxiv.org/pdf/2602.00523v1", "arxiv_id": "2602.00523", "doi": "10.48550/arXiv.2602.00523", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4595}
{"id": "9a5688d2f937cce7542260fd24d7eadb2ff8471d3cb861d072d4477441b2eb39", "sources": ["arxiv", "semantic_scholar"], "title": "LLMs as High-Dimensional Nonlinear Autoregressive Models with Attention: Training, Alignment and Inference", "abstract": "Large language models (LLMs) based on transformer architectures are typically described through collections of architectural components and training procedures, obscuring their underlying computational structure. This review article provides a concise mathematical reference for researchers seeking an explicit, equation-level description of LLM training, alignment, and generation. We formulate LLMs as high-dimensional nonlinear autoregressive models with attention-based dependencies. The framework encompasses pretraining via next-token prediction, alignment methods such as reinforcement learning from human feedback (RLHF), direct preference optimization (DPO), rejection sampling fine-tuning (RSFT), and reinforcement learning from verifiable rewards (RLVR), as well as autoregressive generation during inference. Self-attention emerges naturally as a repeated bilinear--softmax--linear composition, yielding highly expressive sequence models. This formulation enables principled analysis of alignment-induced behaviors (including sycophancy), inference-time phenomena (such as hallucination, in-context learning, chain-of-thought prompting, and retrieval-augmented generation), and extensions like continual learning, while serving as a concise reference for interpretation and further theoretical development.", "authors": ["Vikram Krishnamurthy"], "categories": ["cs.LG", "cs.AI", "cs.CL", "eess.SP"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2026-01-31", "url": "https://arxiv.org/abs/2602.00426", "pdf_url": "https://arxiv.org/pdf/2602.00426v1", "arxiv_id": "2602.00426", "doi": "10.48550/arXiv.2602.00426", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4595}
{"id": "f7a595a710135729e69c3c16bd32c089b67df85f11132127cf0243363a1df9a4", "sources": ["arxiv", "semantic_scholar"], "title": "TriSpec: Ternary Speculative Decoding via Lightweight Proxy Verification", "abstract": "Inference efficiency in Large Language Models (LLMs) is fundamentally limited by their serial, autoregressive generation, especially as reasoning becomes a key capability and response sequences grow longer. Speculative decoding (SD) offers a powerful solution, providing significant speed-ups through its lightweight drafting and parallel verification mechanism. While existing work has nearly saturated improvements in draft effectiveness and efficiency, this paper advances SD from a new yet critical perspective: the verification cost. We propose TriSpec, a novel ternary SD framework that, at its core, introduces a lightweight proxy to significantly reduce computational cost by approving easily verifiable draft sequences and engaging the full target model only when encountering uncertain tokens. TriSpec can be integrated with state-of-the-art SD methods like EAGLE-3 to further reduce verification costs, achieving greater acceleration. Extensive experiments on the Qwen3 and DeepSeek-R1-Distill-Qwen/LLaMA families show that TriSpec achieves up to 35\\% speedup over standard SD, with up to 50\\% fewer target model invocations while maintaining comparable accuracy.", "authors": ["Haoyun Jiang", "Junqi He", "Feng Hong", "Xinlong Yang", "Jianwei Zhang", "Zheng Li", "Zhengyang Zhuge", "Zhiyong Chen", "Bo Han", "Junyang Lin", "Jiangchao Yao"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-30", "url": "https://arxiv.org/abs/2601.23180", "pdf_url": "https://arxiv.org/pdf/2601.23180v1", "arxiv_id": "2601.23180", "doi": "10.48550/arXiv.2601.23180", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4583}
{"id": "e653377ac773be3448a90ab49ee5a50810cfdcb3c85d636dfe531e5ae857cf2a", "sources": ["arxiv", "semantic_scholar"], "title": "StarSD: One-for-Many Speculative Decoding", "abstract": "Speculative decoding accelerates autoregressive generation by separating token proposal from verification, but most existing approaches are designed for single-node execution and do not scale well to multi-accelerator clusters used for serving modern Large Language Models (LLMs). We present StarSD, a one-for-many speculative decoding framework that uses a single draft model to serve multiple target models across distributed nodes via a star topology. StarSD decouples drafting and verification, enabling effective sharing of draft computation, and preventing distributed accelerators from remaining idle under bursty workloads. We provide a system-level analysis that characterizes when and why a single draft model can remain fully utilized by multiple verifiers, yielding predictable latency and utilization gains. Extensive experiments in real-world distributed inference settings demonstrate that StarSD simplifies deployment and supports flexible resource allocation across heterogeneous accelerators, while maintaining output quality. These results indicate that StarSD is a practical and scalable framework for bringing speculative decoding to modern cloud and edge inference infrastructures.", "authors": ["Junhao He", "Feiran You", "Hongyang Du"], "categories": ["eess.SY"], "fields_of_study": ["Engineering", "Computer Science"], "published_date": "2026-01-29", "url": "https://arxiv.org/abs/2601.21622", "pdf_url": "https://arxiv.org/pdf/2601.21622v1", "arxiv_id": "2601.21622", "doi": "10.48550/arXiv.2601.21622", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4572}
{"id": "ce489f83ca124233083ae64ef35a607f69743de5cee2a10a19c3535b29f0d3ac", "sources": ["arxiv", "semantic_scholar"], "title": "TABED: Test-Time Adaptive Ensemble Drafting for Robust Speculative Decoding in LVLMs", "abstract": "Speculative decoding (SD) has proven effective for accelerating LLM inference by quickly generating draft tokens and verifying them in parallel. However, SD remains largely unexplored for Large Vision-Language Models (LVLMs), which extend LLMs to process both image and text prompts. To address this gap, we benchmark existing inference methods with small draft models on 11 datasets across diverse input scenarios and observe scenario-specific performance fluctuations. Motivated by these findings, we propose Test-time Adaptive Batched Ensemble Drafting (TABED), which dynamically ensembles multiple drafts obtained via batch inference by leveraging deviations from past ground truths available in the SD setting. The dynamic ensemble method achieves an average robust walltime speedup of 1.74x over autoregressive decoding and a 5% improvement over single drafting methods, while remaining training-free and keeping ensembling costs negligible through parameter sharing. With its plug-and-play compatibility, we further enhance TABED by integrating advanced verification and alternative drafting methods. Code and custom-trained models are available at https://github.com/furiosa-ai/TABED.", "authors": ["Minjae Lee", "Wonjun Kang", "Byeongkeun Ahn", "Christian Classen", "Kevin Galim", "Seunghyuk Oh", "Minghao Yan", "Hyung Il Koo", "Kangwook Lee"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-28", "url": "https://arxiv.org/abs/2601.20357", "pdf_url": "https://arxiv.org/pdf/2601.20357v1", "arxiv_id": "2601.20357", "doi": "10.48550/arXiv.2601.20357", "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/furiosa-ai/TABED", "venue": "Conference of the European Chapter of the Association for Computational Linguistics", "quality_score": 0.7048}
{"id": "307d3d85a4d5d45ddcfa6a0e380c060eeca2db027fe36c0aa0d6a19e4db304f8", "sources": ["arxiv", "semantic_scholar"], "title": "DART: Diffusion-Inspired Speculative Decoding for Fast LLM Inference", "abstract": "Speculative decoding is an effective and lossless approach for accelerating LLM inference. However, existing widely adopted model-based draft designs, such as EAGLE3, improve accuracy at the cost of multi-step autoregressive inference, resulting in high drafting latency and ultimately rendering the drafting stage itself a performance bottleneck. Inspired by diffusion-based large language models (dLLMs), we propose DART, which leverages parallel generation to reduce drafting latency. DART predicts logits for multiple future masked positions in parallel within a single forward pass based on hidden states of the target model, thereby eliminating autoregressive rollouts in the draft model while preserving a lightweight design. Based on these parallel logit predictions, we further introduce an efficient tree pruning algorithm that constructs high-quality draft token trees with N-gram-enforced semantic continuity. DART substantially reduces draft-stage overhead while preserving high draft accuracy, leading to significantly improved end-to-end decoding speed. Experimental results demonstrate that DART achieves a 2.03x--3.44x wall-clock time speedup across multiple datasets, surpassing EAGLE3 by 30% on average and offering a practical speculative decoding framework. Code is released at https://github.com/fvliang/DART.", "authors": ["Fuliang Liu", "Xue Li", "Ketai Zhao", "Yinxi Gao", "Ziyan Zhou", "Zhonghui Zhang", "Zhibin Wang", "Wanchun Dou", "Sheng Zhong", "Chen Tian"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-27", "url": "https://arxiv.org/abs/2601.19278", "pdf_url": "https://arxiv.org/pdf/2601.19278v1", "arxiv_id": "2601.19278", "doi": "10.48550/arXiv.2601.19278", "citation_count": 11, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/fvliang/DART", "venue": "arXiv.org", "quality_score": 0.703}
{"id": "ec1cffae932cd51406ec96fbfe04ce78dd9a2df67f2b282cc6fde30174c0a006", "sources": ["arxiv", "semantic_scholar"], "title": "LLM-42: Enabling Determinism in LLM Inference with Verified Speculation", "abstract": "In LLM inference, the same prompt may yield different outputs across different runs. At the system level, this non-determinism arises from floating-point non-associativity combined with dynamic batching and GPU kernels whose reduction orders vary with batch size. A straightforward way to eliminate non-determinism is to disable dynamic batching during inference, but doing so severely degrades throughput. Another approach is to make kernels batch-invariant; however, this tightly couples determinism to kernel design, requiring new implementations. This coupling also imposes fixed runtime overheads, regardless of how much of the workload actually requires determinism. Inspired by ideas from speculative decoding, we present LLM-42, a scheduling-based approach to enable determinism in LLM inference. Our key observation is that if a sequence is in a consistent state, the next emitted token is likely to be consistent even with dynamic batching. Moreover, most GPU kernels use shape-consistent reductions. Leveraging these insights, LLM-42 decodes tokens using a non-deterministic fast path and enforces determinism via a lightweight verify-rollback loop. The verifier replays candidate tokens under a fixed-shape reduction schedule, commits those that are guaranteed to be consistent across runs, and rolls back those violating determinism. LLM-42 mostly re-uses existing kernels unchanged and incurs overhead only in proportion to the traffic that requires determinism.", "authors": ["Raja Gond", "Aditya K Kamath", "Ramachandran Ramjee", "Ashish Panwar"], "categories": ["cs.LG", "cs.AI", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-25", "url": "https://arxiv.org/abs/2601.17768", "pdf_url": "https://arxiv.org/pdf/2601.17768v2", "arxiv_id": "2601.17768", "doi": "10.48550/arXiv.2601.17768", "citation_count": 7, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/microsoft/llm-42", "venue": "arXiv.org", "quality_score": 0.6995}
{"id": "440d3337d6a8a10ae8eabd7da0d34d6b7c5575cc4fa33589931467ba36e790fd", "sources": ["arxiv", "semantic_scholar"], "title": "MARS: Unleashing the Power of Speculative Decoding via Margin-Aware Verification", "abstract": "Speculative Decoding (SD) accelerates autoregressive large language model (LLM) inference by decoupling generation and verification. While recent methods improve draft quality by tightly coupling the drafter with the target model, the verification mechanism itself remains largely unchanged, relying on strict token-level rejection sampling. In practice, modern LLMs frequently operate in low-margin regimes where the target model exhibits weak preference among top candidates. In such cases, rejecting plausible runner-up tokens yields negligible information gain while incurring substantial rollback cost, leading to a fundamental inefficiency in verification. We propose Margin-Aware Speculative Verification, a training-free and domain-agnostic verification strategy that adapts to the target model's local decisiveness. Our method conditions verification on decision stability measured directly from the target logits and relaxes rejection only when strict verification provides minimal benefit. Importantly, the approach modifies only the verification rule and is fully compatible with existing target-coupled speculative decoding frameworks. Extensive experiments across model scales ranging from 8B to 235B demonstrate that our method delivers consistent and significant inference speedups over state-of-the-art baselines while preserving generation quality across diverse benchmarks. The code is available at https://github.com/5SSjw/MARS.", "authors": ["Jingwei Song", "Xinyu Wang", "Hanbin Wang", "Xiaoxuan Lei", "Bill Shi", "Shixin Han", "Eric Yang", "Xiao-Wen Chang", "Lynn Ai"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-21", "url": "https://arxiv.org/abs/2601.15498", "pdf_url": "https://arxiv.org/pdf/2601.15498v2", "arxiv_id": "2601.15498", "doi": "10.48550/arXiv.2601.15498", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/5SSjw/MARS", "venue": "arXiv.org", "quality_score": 0.6924}
{"id": "d389a309b0d387da23092e00ff5101132f9df70d303a178d43587fb6f595264b", "sources": ["arxiv", "semantic_scholar"], "title": "Stabilizer-Assisted Inactivation Decoding of Quantum Error-Correcting Codes with Erasures", "abstract": "In this work, we develop a reduced complexity maximum likelihood (ML) decoder for quantum low-density parity-check (QLDPC) codes over erasures. Our decoder combines classical inactivation decoding, which integrates peeling with symbolic guessing, with a new dual peeling procedure. In the dual peeling stage, we perform row operations on the stabilizer matrix to efficiently reveal stabilizer generators and their linear combinations whose support lies entirely on the erased set. Each such stabilizer identified allows us to freely fix a bit in its support without affecting the logical state of the decoded result. This removes one degree of freedom that would otherwise require a symbolic guess, reducing the number of inactivated variables and decreasing the size of the final linear system that must be solved. We further show that dual peeling combined with standard peeling alone, without inactivation, is sufficient to achieve ML for erasure decoding of surface codes. Simulations across several QLDPC code families confirm that our decoder matches ML logical failure performance while significantly reducing the complexity of inactivation decoding, including more than a 20% reduction in symbolic guesses for the B1 lifted product code at high erasure rates.", "authors": ["Giulio Pech", "Mert Gökduman", "Hanwen Yao", "Henry D. Pfister"], "categories": ["cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2026-01-20", "url": "https://arxiv.org/abs/2601.14236", "pdf_url": "https://arxiv.org/pdf/2601.14236v1", "arxiv_id": "2601.14236", "doi": "10.48550/arXiv.2601.14236", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4469}
{"id": "aff354abaf6986106576eaed09db112bc85046fcf08c4a59ba54d6b5d51574aa", "sources": ["arxiv", "semantic_scholar"], "title": "WISP: Waste- and Interference-Suppressed Distributed Speculative LLM Serving at the Edge via Dynamic Drafting and SLO-Aware Batching", "abstract": "As Large Language Models (LLMs) become increasingly accessible to end users, an ever-growing number of inference requests are initiated from edge devices and computed on centralized GPU clusters. However, the resulting exponential growth in computation workload is placing significant strain on data centers, while edge devices remain largely underutilized, leading to imbalanced workloads and resource inefficiency across the network. Integrating edge devices into the LLM inference process via speculative decoding helps balance the workload between the edge and the cloud, while maintaining lossless prediction accuracy. In this paper, we identify and formalize two critical bottlenecks that limit the efficiency and scalability of distributed speculative LLM serving: Wasted Drafting Time and Verification Interference. To address these challenges, we propose WISP, an efficient and SLO-aware distributed LLM inference system that consists of an intelligent speculation controller, a verification time estimator, and a verification batch scheduler. These components collaboratively enhance drafting efficiency and optimize verification request scheduling on the server. Extensive numerical results show that WISP improves system capacity by up to 2.1x and 4.1x, and increases system goodput by up to 1.94x and 3.7x, compared to centralized serving and SLED, respectively.", "authors": ["Xiangchen Li", "Jiakun Fan", "Qingyuan Wang", "Dimitrios Spatharakis", "Saeid Ghafouri", "Hans Vandierendonck", "Deepu John", "Bo Ji", "Ali R. Butt", "Dimitrios S. Nikolopoulos"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-15", "url": "https://arxiv.org/abs/2601.11652", "pdf_url": "https://arxiv.org/pdf/2601.11652v2", "arxiv_id": "2601.11652", "doi": "10.48550/arXiv.2601.11652", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.2807}
{"id": "005c13f4225924a16d6ce92bd7dbd5537faca06c2ed425b20777f98bb15d6ff5", "sources": ["arxiv", "semantic_scholar"], "title": "Annealed Relaxation of Speculative Decoding for Faster Autoregressive Image Generation", "abstract": "Despite significant progress in autoregressive image generation, inference remains slow due to the sequential nature of AR models and the ambiguity of image tokens, even when using speculative decoding. Recent works attempt to address this with relaxed speculative decoding but lack theoretical grounding. In this paper, we establish the theoretical basis of relaxed SD and propose COOL-SD, an annealed relaxation of speculative decoding built on two key insights. The first analyzes the total variation (TV) distance between the target model and relaxed speculative decoding and yields an optimal resampling distribution that minimizes an upper bound of the distance. The second uses perturbation analysis to reveal an annealing behaviour in relaxed speculative decoding, motivating our annealed design. Together, these insights enable COOL-SD to generate images faster with comparable quality, or achieve better quality at similar latency. Experiments validate the effectiveness of COOL-SD, showing consistent improvements over prior methods in speed-quality trade-offs.", "authors": ["Xingyao Li", "Fengzhuo Zhang", "Cunxiao Du", "Hui Ji"], "categories": ["cs.CV", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-14", "url": "https://arxiv.org/abs/2601.09212", "pdf_url": "https://arxiv.org/pdf/2601.09212v1", "arxiv_id": "2601.09212", "doi": "10.48550/arXiv.2601.09212", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "AAAI Conference on Artificial Intelligence", "quality_score": 0.44}
{"id": "95ee200f10728e73faea9428052361304de5bd18298a44e8e342faf36c033172", "sources": ["arxiv", "semantic_scholar"], "title": "HIPPO: Accelerating Video Large Language Models Inference via Holistic-aware Parallel Speculative Decoding", "abstract": "Speculative decoding (SD) has emerged as a promising approach to accelerate LLM inference without sacrificing output quality. Existing SD methods tailored for video-LLMs primarily focus on pruning redundant visual tokens to mitigate the computational burden of massive visual inputs. However, existing methods do not achieve inference acceleration comparable to text-only LLMs. We observe from extensive experiments that this phenomenon mainly stems from two limitations: (i) their pruning strategies inadequately preserve visual semantic tokens, degrading draft quality and acceptance rates; (ii) even with aggressive pruning (e.g., 90% visual tokens removed), the draft model's remaining inference cost limits overall speedup. To address these limitations, we propose HIPPO, a general holistic-aware parallel speculative decoding framework. Specifically, HIPPO proposes (i) a semantic-aware token preservation method, which fuses global attention scores with local visual semantics to retain semantic information at high pruning ratios; (ii) a video parallel SD algorithm that decouples and overlaps draft generation and target verification phases. Experiments on four video-LLMs across six benchmarks demonstrate HIPPO's effectiveness, yielding up to 3.51x speedup compared to vanilla auto-regressive decoding.", "authors": ["Qitan Lv", "Tianyu Liu", "Wen Wu", "Xuenan Xu", "Bowen Zhou", "Feng Wu", "Chao Zhang"], "categories": ["cs.CV", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-13", "url": "https://arxiv.org/abs/2601.08273", "pdf_url": "https://arxiv.org/pdf/2601.08273v1", "arxiv_id": "2601.08273", "doi": "10.48550/arXiv.2601.08273", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4389}
{"id": "6d78267baef039f0be8c9a0a4c9abdc16090a78de7e7488b42ba5562f36586cd", "sources": ["arxiv", "semantic_scholar"], "title": "TALON: Confidence-Aware Speculative Decoding with Adaptive Token Trees", "abstract": "Speculative decoding (SD) has become a standard technique for accelerating LLM inference without sacrificing output quality. Recent advances in speculative decoding have shifted from sequential chain-based drafting to tree-structured generation, where the draft model constructs a tree of candidate tokens to explore multiple possible drafts in parallel. However, existing tree-based SD methods typically build a fixed-width, fixed-depth draft tree, which fails to adapt to the varying difficulty of tokens and contexts. As a result, the draft model cannot dynamically adjust the tree structure to early stop on difficult tokens and extend generation for simple ones. To address these challenges, we introduce TALON, a training-free, budget-driven adaptive tree expansion framework that can be plugged into existing tree-based methods. Unlike static methods, TALON constructs the draft tree iteratively until a fixed token budget is met, using a hybrid expansion strategy that adaptively allocates the node budget to each layer of the draft tree. This framework naturally shapes the draft tree into a \"deep-and-narrow\" form for deterministic contexts and a \"shallow-and-wide\" form for uncertain branches, effectively optimizing the trade-off between exploration width and generation depth under a given budget. Extensive experiments across 5 models and 6 datasets demonstrate that TALON consistently outperforms state-of-the-art EAGLE-3, achieving up to 5.16x end-to-end speedup over auto-regressive decoding.", "authors": ["Tianyu Liu", "Qitan Lv", "Yuhao Shen", "Xiao Sun", "Xiaoyan Sun"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-12", "url": "https://arxiv.org/abs/2601.07353", "pdf_url": "https://arxiv.org/pdf/2601.07353v1", "arxiv_id": "2601.07353", "doi": "10.48550/arXiv.2601.07353", "citation_count": 8, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4377}
{"id": "5351116365b1b6f228e89daffd95a548d2b485d68f4a15f38abca686c6cf73b9", "sources": ["arxiv", "semantic_scholar"], "title": "Overcoming Joint Intractability with Lossless Hierarchical Speculative Decoding", "abstract": "Verification is a key bottleneck in improving inference speed while maintaining distribution fidelity in Speculative Decoding. Recent work has shown that sequence-level verification leads to a higher number of accepted tokens compared to token-wise verification. However, existing solutions often rely on surrogate approximations or are constrained by partial information, struggling with joint intractability. In this work, we propose Hierarchical Speculative Decoding (HSD), a provably lossless verification method that significantly boosts the expected number of accepted tokens and overcomes joint intractability by balancing excess and deficient probability mass across accessible branches. Our extensive large-scale experiments demonstrate that HSD yields consistent improvements in acceptance rates across diverse model families and benchmarks. Moreover, its strong explainability and generality make it readily integrable into a wide range of speculative decoding frameworks. Notably, integrating HSD into EAGLE-3 yields over a 12% performance gain, establishing state-of-the-art decoding efficiency without compromising distribution fidelity. Code is available at https://github.com/ZhouYuxuanYX/Hierarchical-Speculative-Decoding.", "authors": ["Yuxuan Zhou", "Fei Huang", "Heng Li", "Fengyi Wu", "Tianyu Wang", "Jianwei Zhang", "Junyang Lin", "Zhi-Qi Cheng"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-09", "url": "https://arxiv.org/abs/2601.05724", "pdf_url": "https://arxiv.org/pdf/2601.05724v2", "arxiv_id": "2601.05724", "doi": "10.48550/arXiv.2601.05724", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/ZhouYuxuanYX/Hierarchical-Speculative-Decoding", "venue": "arXiv.org", "quality_score": 0.6711}
{"id": "06f5a3e1424f3adfa47cfcf8890829cbc82a3d8d5ff2f3aebf5909ec5beaf65a", "sources": ["arxiv", "semantic_scholar"], "title": "Multi-Scale Local Speculative Decoding for Image Generation", "abstract": "Autoregressive (AR) models have achieved remarkable success in image synthesis, yet their sequential nature imposes significant latency constraints. Speculative Decoding offers a promising avenue for acceleration, but existing approaches are limited by token-level ambiguity and lack of spatial awareness. In this work, we introduce Multi-Scale Local Speculative Decoding (MuLo-SD), a novel framework that combines multi-resolution drafting with spatially informed verification to accelerate AR image generation. Our method leverages a low-resolution drafter paired with an up-sampling step to propose candidate image tokens, which are then verified in parallel by a high-resolution target model. Crucially, we incorporate a local rejection and resampling mechanism, enabling efficient correction of draft errors by focusing on spatial neighborhoods rather than raster-scan resampling after the first rejection. When integrated with parallel decoding resampling, MuLo-SD achieves substantial speedups -- up to $\\mathbf{5\\times}$ -- outperforming both speculative decoding and parallel decoding baselines in terms of acceleration, while maintaining comparable semantic alignment and perceptual quality. These results are validated using GenEval, DPG-Bench, and FID/HPSv2 on the MS-COCO 5k validation split. Extensive ablations highlight the impact of up-sampling design, probability pooling, and local rejection and resampling with neighborhood expansion. Our approach sets a new state-of-the-art in speculative decoding for image synthesis, bridging the gap between efficiency and fidelity. Project page is available at https://qualcomm-ai-research.github.io/mulo-sd-webpage/ .", "authors": ["Elia Peruzzo", "Guillaume Sautière", "Amirhossein Habibian"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-08", "url": "https://arxiv.org/abs/2601.05149", "pdf_url": "https://arxiv.org/pdf/2601.05149v2", "arxiv_id": "2601.05149", "doi": "10.48550/arXiv.2601.05149", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4331}
{"id": "d46349153a3709ab81f615bc8c6471466fdbf2a9fdd4931a6860ab4049b92ff2", "sources": ["arxiv", "semantic_scholar"], "title": "LoRA-Drop: Temporal LoRA Decoding for Efficient LLM Inference", "abstract": "Autoregressive large language models (LLMs) are bottlenecked by sequential decoding, where each new token typically requires executing all transformer layers. Existing dynamic-depth and layer-skipping methods reduce this cost, but often rely on auxiliary routing mechanisms or incur accuracy degradation when bypassed layers are left uncompensated. We present \\textbf{LoRA-Drop}, a plug-and-play inference framework that accelerates decoding by applying a \\emph{temporal compute schedule} to a fixed subset of intermediate layers: on most decoding steps, selected layers reuse the previous-token hidden state and apply a low-rank LoRA correction, while periodic \\emph{refresh} steps execute the full model to prevent drift. LoRA-Drop requires no routing network, is compatible with standard KV caching, and can reduce KV-cache footprint by skipping KV updates in droppable layers during LoRA steps and refreshing periodically. Across \\textbf{LLaMA2-7B}, \\textbf{LLaMA3-8B}, \\textbf{Qwen2.5-7B}, and \\textbf{Qwen2.5-14B}, LoRA-Drop achieves up to \\textbf{2.6$\\times$ faster decoding} and \\textbf{45--55\\% KV-cache reduction} while staying within \\textbf{0.5 percentage points (pp)} of baseline accuracy. Evaluations on reasoning (GSM8K, MATH, BBH), code generation (HumanEval, MBPP), and long-context/multilingual benchmarks (LongBench, XNLI, XCOPA) identify a consistent \\emph{safe zone} of scheduling configurations that preserves quality while delivering substantial efficiency gains, providing a simple path toward adaptive-capacity inference in LLMs. Codes are available at https://github.com/hosseinbv/LoRA-Drop.git.", "authors": ["Hossein Rajabzadeh", "Maryam Dialameh", "Chul B. Park", "Il-Min Kim", "Hyock Ju Kwon"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-05", "url": "https://arxiv.org/abs/2601.02569", "pdf_url": "https://arxiv.org/pdf/2601.02569v1", "arxiv_id": "2601.02569", "doi": "10.48550/arXiv.2601.02569", "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/hosseinbv/LoRA-Drop.git", "venue": "arXiv.org", "quality_score": 0.6641}
{"id": "b9dfa6ab794986bc558b147c423c668cee3f76d8115b4907b4751157b2b1a041", "sources": ["arxiv", "semantic_scholar"], "title": "FlexSpec: Frozen Drafts Meet Evolving Targets in Edge-Cloud Collaborative LLM Speculative Decoding", "abstract": "Deploying large language models (LLMs) in mobile and edge computing environments is constrained by limited on-device resources, scarce wireless bandwidth, and frequent model evolution. Although edge-cloud collaborative inference with speculative decoding (SD) can reduce end-to-end latency by executing a lightweight draft model at the edge and verifying it with a cloud-side target model, existing frameworks fundamentally rely on tight coupling between the two models. Consequently, repeated model synchronization introduces excessive communication overhead, increasing end-to-end latency, and ultimately limiting the scalability of SD in edge environments. To address these limitations, we propose FlexSpec, a communication-efficient collaborative inference framework tailored for evolving edge-cloud systems. The core design of FlexSpec is a shared-backbone architecture that allows a single and static edge-side draft model to remain compatible with a large family of evolving cloud-side target models. By decoupling edge deployment from cloud-side model updates, FlexSpec eliminates the need for edge-side retraining or repeated model downloads, substantially reducing communication and maintenance costs. Furthermore, to accommodate time-varying wireless conditions and heterogeneous device constraints, we develop a channel-aware adaptive speculation mechanism that dynamically adjusts the speculative draft length based on real-time channel state information and device energy budgets. Extensive experiments demonstrate that FlexSpec achieves superior performance compared to conventional SD approaches in terms of inference efficiency.", "authors": ["Yuchen Li", "Rui Kong", "Zhonghao Lyu", "Qiyang Li", "Xinran Chen", "Hengyi Cai", "Lingyong Yan", "Shuaiqiang Wang", "Jiashu Zhao", "Guangxu Zhu", "Linghe Kong", "Guihai Chen", "Haoyi Xiong", "Dawei Yin"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2026-01-02", "url": "https://arxiv.org/abs/2601.00644", "pdf_url": "https://arxiv.org/pdf/2601.00644v1", "arxiv_id": "2601.00644", "doi": "10.48550/arXiv.2601.00644", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE Transactions on Mobile Computing", "quality_score": 0.4263}
{"id": "6a37ecf101f70bd44f86bc1ec79283bf4ddba6aaab0277f37f8ea57d09725b37", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding: Performance or Illusion?", "abstract": "Speculative decoding (SD) has become a popular technique to accelerate Large Language Model (LLM) inference, yet its real-world effectiveness remains unclear as prior evaluations rely on research prototypes and unrealistically small batch sizes. We present, to our knowledge, the first systematic study of SD on a production-grade and widely deployed inference engine (vLLM), covering multiple SD variants ($n$-gram, EAGLE/EAGLE-3, Draft-Model, Multi-Token Prediction) across diverse workloads, model scales, and batch sizes. We analyze key factors governing SD performance, and quantify a theoretical upper bound on SD speedup. Our results show that verification by the target model dominates the execution, while acceptance length varies markedly across output token positions, requests, and datasets. Comparing measured performance with theoretical bounds reveals substantial gaps between observed and theoretical upper bounds, and we leverage this observation to highlight new research opportunities that our study opens up in improving SD.", "authors": ["Xiaoxuan Liu", "Jiaxiang Yu", "Jongseok Park", "Ion Stoica", "Alvin Cheung"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-31", "url": "https://arxiv.org/abs/2601.11580", "pdf_url": "https://arxiv.org/pdf/2601.11580v2", "arxiv_id": "2601.11580", "doi": "10.48550/arXiv.2601.11580", "citation_count": 8, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.424}
{"id": "f7fa9d9ad5bb6239527942757ebe4d044080c4c6b72e3feb5e53ca0658b899a9", "sources": ["arxiv", "semantic_scholar"], "title": "Entropy-Aware Speculative Decoding Toward Improved LLM Reasoning", "abstract": "Speculative decoding (SD) accelerates large language model (LLM) reasoning by using a small draft model to generate candidate tokens, which the target LLM either accepts directly or regenerates upon rejection. However, excessive alignment between the draft and target models constrains SD to the performance of the target LLM. To address this limitation, we propose Entropy-Aware Speculative Decoding (EASD), a training-free enhancement. Building on standard SD, EASD incorporates a dynamic entropy-based penalty. At each decoding step, we employ the entropy of the sampling distribution to quantify model uncertainty. When both models exhibit high entropy with substantial overlap among their top-N predictions, the corresponding token is rejected and re-sampled by the target LLM. This penalty prevents low-confidence errors from propagating. By incorporating draft-model verification, EASD enables the possibility of surpassing the target model's inherent performance. Experiments across multiple reasoning benchmarks demonstrate that EASD consistently outperforms existing SD methods and, in most cases, surpasses the target LLM itself. We further prove that the efficiency of EASD is comparable to that of SD. The code can be found in the Supplementary Materials.", "authors": ["Tiancheng Su", "Meicong Zhang", "Guoxiu He"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-29", "url": "https://arxiv.org/abs/2512.23765", "pdf_url": "https://arxiv.org/pdf/2512.23765v1", "arxiv_id": "2512.23765", "doi": "10.48550/arXiv.2512.23765", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4217}
{"id": "fea2f8148499c044a32dcd8138ca3c35c6bc6a8a61306f1861540e27e9d0e9a2", "sources": ["arxiv", "semantic_scholar"], "title": "Yggdrasil: Bridging Dynamic Speculation and Static Runtime for Latency-Optimal Tree-Based LLM Decoding", "abstract": "Speculative decoding improves LLM inference by generating and verifying multiple tokens in parallel, but existing systems suffer from suboptimal performance due to a mismatch between dynamic speculation and static runtime assumptions. We present Yggdrasil, a co-designed system that enables latency-optimal speculative decoding through context-aware tree drafting and compiler-friendly execution. Yggdrasil introduces an equal-growth tree structure for static graph compatibility, a latency-aware optimization objective for draft selection, and stage-based scheduling to reduce overhead. Yggdrasil supports unmodified LLMs and achieves up to $3.98\\times$ speedup over state-of-the-art baselines across multiple hardware setups.", "authors": ["Yue Guan", "Changming Yu", "Shihan Fang", "Weiming Hu", "Zaifeng Pan", "Zheng Wang", "Zihan Liu", "Yangjie Zhou", "Yufei Ding", "Minyi Guo", "Jingwen Leng"], "categories": ["cs.LG", "cs.PL"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-29", "url": "https://arxiv.org/abs/2512.23858", "pdf_url": "https://arxiv.org/pdf/2512.23858v1", "arxiv_id": "2512.23858", "doi": "10.48550/arXiv.2512.23858", "citation_count": 3, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4217}
{"id": "ec7402bcbd806d51eeaa663fb73a85956c2e5fde1105175c8b980600fc939c97", "sources": ["arxiv", "semantic_scholar"], "title": "Nightjar: Dynamic Adaptive Speculative Decoding for Large Language Models Serving", "abstract": "Speculative decoding (SD) accelerates LLM inference by verifying draft tokens in parallel. However, this method presents a critical trade-off: it improves throughput in low-load, memory-bound systems but degrades performance in high-load, compute-bound environments due to verification overhead. Existing speculative decoding methods use fixed lengths and cannot adapt to workload changes or decide when to stop speculation. The cost of restarting speculative inference also remains unquantified. Under high load, the benefit of speculation diminishes, while retaining the draft model reduces KV cache capacity, limiting batch size and degrading throughput. To overcome this, we propose Nightjar, a resource-aware adaptive speculative framework. It first adjusts to the request load by dynamically selecting the optimal speculative length for different batch sizes. Crucially, Nightjar proactively disables speculative decoding when the MAB planner determines that speculation is no longer beneficial, and during the disabled phase, offloads the draft model to the CPU only under GPU memory pressure. This reclaims memory for the KV cache, thereby facilitating larger batch sizes and maximizing overall system throughput. Experiments show that Nightjar achieves up to 14.76% higher throughput than standard speculative decoding and up to 20.18% lower latency in the main benchmark suite under dynamic request arrival rates for real-time LLM serving scenarios.", "authors": ["Rui Li", "Zhaoning Zhang", "Libo Zhang", "Huaimin Wang", "Xiang Fu", "Zhiquan Lai"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-27", "url": "https://arxiv.org/abs/2512.22420", "pdf_url": "https://arxiv.org/pdf/2512.22420v5", "arxiv_id": "2512.22420", "doi": "10.48550/arXiv.2512.22420", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Journal of systems architecture", "quality_score": 0.4194}
{"id": "eedb793a0825b82ee5f0476441917a0bf7f9d4bf4543530592c1d806d8bafcca", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerate Speculative Decoding with Sparse Computation in Verification", "abstract": "Speculative decoding accelerates autoregressive language model inference by verifying multiple draft tokens in parallel. However, the verification stage often becomes the dominant computational bottleneck, especially for long-context inputs and mixture-of-experts (MoE) models. Existing sparsification methods are designed primarily for standard token-by-token autoregressive decoding to remove substantial computational redundancy in LLMs. This work systematically adopts different sparse methods on the verification stage of the speculative decoding and identifies structured redundancy across multiple dimensions. Based on these observations, we propose a sparse verification framework that jointly sparsifies attention, FFN, and MoE components during the verification stage to reduce the dominant computation cost. The framework further incorporates an inter-draft token and inter-layer retrieval reuse strategy to further reduce redundant computation without introducing additional training. Extensive experiments across summarization, question answering, and mathematical reasoning datasets demonstrate that the proposed methods achieve favorable efficiency-accuracy trade-offs, while maintaining stable acceptance length.", "authors": ["Jikai Wang", "Jianchao Tan", "Yuxuan Hu", "Jiayu Qin", "Yerui Sun", "Yuchen Xie", "Xunliang Cai", "Juntao Li", "Min Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-26", "url": "https://arxiv.org/abs/2512.21911", "pdf_url": "https://arxiv.org/pdf/2512.21911v1", "arxiv_id": "2512.21911", "doi": "10.48550/arXiv.2512.21911", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4182}
{"id": "74a85d93819e9e43dbac2ac476bb113996d249e0cf4fdfe7e9e202e9e91bc4ba", "sources": ["arxiv", "semantic_scholar"], "title": "Fail Fast, Win Big: Rethinking the Drafting Strategy in Speculative Decoding via Diffusion LLMs", "abstract": "Diffusion Large Language Models (dLLMs) offer fast, parallel token generation, but their standalone use is plagued by an inherent efficiency-quality tradeoff. We show that, if carefully applied, the attributes of dLLMs can actually be a strength for drafters in speculative decoding with autoregressive (AR) verifiers. Our core insight is that dLLM's speed from parallel decoding drastically lowers the risk of costly rejections, providing a practical mechanism to effectively realize the (elusive) lengthy drafts that lead to large speedups with speculative decoding. We present FailFast, a dLLM-based speculative decoding framework that realizes this approach by dynamically adapting its speculation length. It \"fails fast\" by spending minimal compute in hard-to-speculate regions to shrink speculation latency and \"wins big\" by aggressively extending draft lengths in easier regions to reduce verification latency (in many cases, speculating and accepting 70 tokens at a time!). Without any fine-tuning, FailFast delivers lossless acceleration of AR LLMs and achieves up to 4.9$\\times$ speedup over vanilla decoding, 1.7$\\times$ over the best naive dLLM drafter, and 1.7$\\times$ over EAGLE-3 across diverse models and workloads. We open-source FailFast at https://github.com/ruipeterpan/failfast.", "authors": ["Rui Pan", "Zhuofu Chen", "Hongyi Liu", "Arvind Krishnamurthy", "Ravi Netravali"], "categories": ["cs.LG", "cs.AI", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-23", "url": "https://arxiv.org/abs/2512.20573", "pdf_url": "https://arxiv.org/pdf/2512.20573v3", "arxiv_id": "2512.20573", "doi": "10.48550/arXiv.2512.20573", "citation_count": 4, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/ruipeterpan/failfast", "venue": "arXiv.org", "quality_score": 0.641}
{"id": "c8c14eaa0db49706c434427c1ad212c0b3bdb8f610474766adc2115a0846be5a", "sources": ["arxiv", "semantic_scholar"], "title": "Fast Collaborative Inference via Distributed Speculative Decoding", "abstract": "Speculative decoding accelerates large language model (LLM) inference by allowing a small draft model to predict multiple future tokens for verification by a larger target model. In AI-native radio access networks (AI-RAN), this enables device-edge collaborative inference but introduces significant uplink overhead, as existing distributed speculative decoding schemes transmit full vocabulary logits at every step. We propose a sparsify-then-sample strategy, Truncated Sparse Logits Transmission (TSLT), which transmits only the logits and indices of a truncated candidate set. We provide theoretical guarantees showing that the acceptance rate is preserved under TSLT. TSLT is further extended to multi-candidate case, where multiple draft candidates per step increase acceptance probability. Experiments show that TSLT significantly reduces uplink communication while maintaining end-to-end inference latency and model quality, demonstrating its effectiveness for scalable, communication-efficient distributed LLM inference in future AI-RAN systems.", "authors": ["Ce Zheng", "Ke Zhang", "Chen Sun", "Wenqi Zhang", "Qiong Liu", "Angesom Ataklity Tesfay"], "categories": ["eess.SP"], "fields_of_study": ["Engineering"], "published_date": "2025-12-18", "url": "https://arxiv.org/abs/2512.16273", "pdf_url": "https://arxiv.org/pdf/2512.16273v2", "arxiv_id": "2512.16273", "doi": "10.1016/j.jiixd.2025.12.008", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Journal of Information and Intelligence", "quality_score": 0.4091}
{"id": "499684e1a913c91000964b15c55641304481c53ac2a6d836d928a4bc0ca45e41", "sources": ["arxiv", "semantic_scholar"], "title": "Efficient Adaptive Rejection Sampling for Accelerating Speculative Decoding in Large Language Models", "abstract": "Speculative Decoding is a prominent technique for accelerating the autoregressive inference of large language models (LLMs) by employing a fast draft model to propose candidate token sequences and a large target model to verify them in parallel. However, its core component -- the rejection sampling mechanism -- relies on a fixed, context-independent random threshold. This leads to a significant \"random rejection\" problem in high-uncertainty generation scenarios, where plausible candidate tokens are frequently rejected due to random chance, undermining inference efficiency. This paper introduces Efficient Adaptive Rejection Sampling (EARS), a novel method that dynamically adjusts the acceptance threshold by incorporating the target model's own predictive uncertainty, measured as 1 - max(P_target). By introducing a tolerance term proportional to this uncertainty, EARS intelligently relaxes the acceptance criterion when the model is uncertain, effectively reducing random rejections while maintaining strict standards when the model is confident. Experiments on creative writing and open-domain QA tasks demonstrate that EARS significantly enhances the efficiency of speculative decoding, achieving up to an 18.12% increase in throughput with a negligible 0.84% accuracy drop on the GSM8K benchmark. The method requires no modifications to model architectures and can be seamlessly integrated into existing speculative decoding frameworks.", "authors": ["Chendong Sun", "Ali Mao", "Lei Xu", "mingmin Chen"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-15", "url": "https://arxiv.org/abs/2512.13194", "pdf_url": "https://arxiv.org/pdf/2512.13194v3", "arxiv_id": "2512.13194", "doi": "10.48550/arXiv.2512.13194", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4056}
{"id": "30507c9f738661addfb8d8658d3369800663d30373e2958894fc3615fac4659d", "sources": ["arxiv", "semantic_scholar"], "title": "TS-DP: Reinforcement Speculative Decoding For Temporal Adaptive Diffusion Policy Acceleration", "abstract": "Diffusion Policy (DP) excels in embodied control but suffers from high inference latency and computational cost due to multiple iterative denoising steps. The temporal complexity of embodied tasks demands a dynamic and adaptable computation mode. Static and lossy acceleration methods, such as quantization, fail to handle such dynamic embodied tasks, while speculative decoding offers a lossless and adaptive yet underexplored alternative for DP. However, it is non-trivial to address the following challenges: how to match the base model's denoising quality at lower cost under time-varying task difficulty in embodied settings, and how to dynamically and interactively adjust computation based on task difficulty in such environments. In this paper, we propose Temporal-aware Reinforcement-based Speculative Diffusion Policy (TS-DP), the first framework that enables speculative decoding for DP with temporal adaptivity. First, to handle dynamic environments where task difficulty varies over time, we distill a Transformer-based drafter to imitate the base model and replace its costly denoising calls. Second, an RL-based scheduler further adapts to time-varying task difficulty by adjusting speculative parameters to maintain accuracy while improving efficiency. Extensive experiments across diverse embodied environments demonstrate that TS-DP achieves up to 4.17 times faster inference with over 94% accepted drafts, reaching an inference frequency of 25 Hz and enabling real-time diffusion-based control without performance degradation.", "authors": ["Ye Li", "Jiahe Feng", "Yuan Meng", "Kangye Ji", "Chen Tang", "Xinwan Wen", "Shutao Xia", "Zhi Wang", "Wenwu Zhu"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-13", "url": "https://arxiv.org/abs/2512.15773", "pdf_url": "https://arxiv.org/pdf/2512.15773v1", "arxiv_id": "2512.15773", "doi": "10.48550/arXiv.2512.15773", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4033}
{"id": "59a03ddf2e16daeaa14dc6cc178590dca8c3d72eca7829aa1b8c19168da8b458", "sources": ["arxiv", "semantic_scholar"], "title": "AdaSD: Adaptive Speculative Decoding for Efficient Language Model Inference", "abstract": "Large language models (LLMs) have achieved remarkable performance across a wide range of tasks, but their increasing parameter sizes significantly slow down inference. Speculative decoding mitigates this issue by leveraging a smaller draft model to predict candidate tokens, which are then verified by a larger target model. However, existing approaches often require additional training, extensive hyperparameter tuning, or prior analysis of models and tasks before deployment. In this paper, we propose Adaptive Speculative Decoding (AdaSD), a hyperparameter-free decoding scheme that dynamically adjusts generation length and acceptance criteria during inference. AdaSD introduces two adaptive components: one to determine when to stop candidate token generation and the other to decide token acceptance, updated in real time based on token entropy and Jensen-Shannon distance. This approach eliminates the need for pre-analysis or fine-tuning and is compatible with off-the-shelf models. Experiments on benchmark datasets demonstrate that AdaSD achieves up to 1.46x speedup over vanilla speculative decoding while limiting accuracy degradation to under 1.8%, making it a practical solution for efficient and adaptive LLM inference.", "authors": ["Kuan-Wei Lu", "Ding-Yong Hong", "Pangfeng Liu", "Jan-Jan Wu"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-12", "url": "https://arxiv.org/abs/2512.11280", "pdf_url": "https://arxiv.org/pdf/2512.11280v2", "arxiv_id": "2512.11280", "doi": "10.48550/arXiv.2512.11280", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4022}
{"id": "67e315d660d5f1469f1263b315c90fdec539f557aa3179f8a111ba7052ab0cea", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding Speed-of-Light: Optimal Lower Bounds via Branching Random Walks", "abstract": "Speculative generation has emerged as a promising technique to accelerate inference in large language models (LLMs) by leveraging parallelism to verify multiple draft tokens simultaneously. However, the fundamental limits on the achievable speedup remain poorly understood. In this work, we establish the first ``tight'' lower bounds on the runtime of any deterministic speculative generation algorithm. This is achieved by drawing a parallel between the token generation process and branching random walks, which allows us to analyze the optimal draft tree selection problem. We prove, under basic assumptions, that the expected number of tokens successfully predicted per speculative iteration is bounded as $\\mathbb{E}[X] \\leq (μ+ μ_{(2)})\\log(P )/μ^2 + O(1)$, where $P$ is the verifier's capacity, $μ$ is the expected entropy of the verifier's output distribution, and $μ_{(2)}$ is the expected second log-moment. This result provides new insights into the limits of parallel token generation, and could guide the design of future speculative decoding systems. Empirical evaluations on Llama models validate our theoretical predictions, confirming the tightness of our bounds in practical settings.", "authors": ["Sergey Pankratov", "Dan Alistarh"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-12", "url": "https://arxiv.org/abs/2512.11718", "pdf_url": "https://arxiv.org/pdf/2512.11718v1", "arxiv_id": "2512.11718", "doi": "10.48550/arXiv.2512.11718", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Conference of the European Chapter of the Association for Computational Linguistics", "quality_score": 0.4022}
{"id": "670a39eec13d5f66aa9e877ad2f1d84fbcb03bf2c49eea0cf0014f0004c268c8", "sources": ["arxiv", "semantic_scholar"], "title": "CXL-SpecKV: A Disaggregated FPGA Speculative KV-Cache for Datacenter LLM Serving", "abstract": "Large Language Models (LLMs) have revolutionized natural language processing tasks, but their deployment in datacenter environments faces significant challenges due to the massive memory requirements of key-value (KV) caches. During the autoregressive decoding process, KV caches consume substantial GPU memory, limiting batch sizes and overall system throughput. To address these challenges, we propose \\textbf{CXL-SpecKV}, a novel disaggregated KV-cache architecture that leverages Compute Express Link (CXL) interconnects and FPGA accelerators to enable efficient speculative execution and memory disaggregation. Our approach introduces three key innovations: (i) a CXL-based memory disaggregation framework that offloads KV-caches to remote FPGA memory with low latency, (ii) a speculative KV-cache prefetching mechanism that predicts and preloads future tokens' cache entries, and (iii) an FPGA-accelerated KV-cache compression and decompression engine that reduces memory bandwidth requirements by up to 4$\\times$. When evaluated on state-of-the-art LLM models, CXL-SpecKV achieves up to 3.2$\\times$ higher throughput compared to GPU-only baselines, while reducing memory costs by 2.8$\\times$ and maintaining accuracy. Our system demonstrates that intelligent memory disaggregation combined with speculative execution can effectively address the memory wall challenge in large-scale LLM serving. Our code implementation has been open-sourced at https://github.com/FastLM/CXL-SpecKV.", "authors": ["Dong Liu", "Yanxuan Yu"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-11", "url": "https://arxiv.org/abs/2512.11920", "pdf_url": "https://arxiv.org/pdf/2512.11920v1", "arxiv_id": "2512.11920", "doi": "10.1145/3748173.3779188", "citation_count": 7, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/FastLM/CXL-SpecKV", "venue": "Symposium on Field Programmable Gate Arrays", "quality_score": 0.6198}
{"id": "2e80d9b6fbf019bf647f2f04a06f3066ac60b83b23be734ec2b284fe908d6065", "sources": ["arxiv", "semantic_scholar"], "title": "GoodSpeed: Optimizing Fair Goodput with Adaptive Speculative Decoding in Distributed Edge Inference", "abstract": "Large language models (LLMs) have revolutionized natural language processing, yet their high computational demands pose significant challenges for real-time inference, especially in multi-user server speculative decoding and resource-constrained environments. Speculative decoding has emerged as a promising technique to accelerate LLM inference by using lightweight draft models to generate candidate tokens, which are subsequently verified by a larger, more accurate model. However, ensuring both high goodput (the effective rate of accepted tokens) and fairness across multiple draft servers cooperating with a central verification server remains an open challenge. This paper introduces GOODSPEED, a novel distributed inference framework that optimizes goodput through adaptive speculative decoding. GOODSPEED employs a central verification server that coordinates a set of heterogeneous draft servers, each running a small language model to generate speculative tokens. To manage resource allocation effectively, GOODSPEED incorporates a gradient scheduling algorithm that dynamically assigns token verification tasks, maximizing a logarithmic utility function to ensure proportional fairness across servers. By processing speculative outputs from all draft servers in parallel, the framework enables efficient collaboration between the verification server and distributed draft generators, streamlining both latency and throughput. Through rigorous fluid sample path analysis, we show that GOODSPEED converges to the optimal goodput allocation in steady-state conditions and maintains near-optimal performance with provably bounded error under dynamic workloads. These results demonstrate that GOODSPEED provides a scalable, fair and efficient solution for multi-server speculative decoding in distributed LLM inference systems.", "authors": ["Phuong Tran", "Tzu-Hao Liu", "Long Tan Le", "Tung-Anh Nguyen", "Van Quan La", "Eason Yu", "Han Shu", "Choong Seon Hong", "Nguyen H. Tran"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-10", "url": "https://arxiv.org/abs/2512.09963", "pdf_url": "https://arxiv.org/pdf/2512.09963v2", "arxiv_id": "2512.09963", "doi": "10.48550/arXiv.2512.09963", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3999}
{"id": "0557546589750fcbeab1a35e6bccf23929d4f4507b4d9454decdd8811ca5aefd", "sources": ["arxiv", "semantic_scholar"], "title": "A scalable and real-time neural decoder for topological quantum codes", "abstract": "Fault-tolerant quantum computing will require error rates far below those achievable with physical qubits. Quantum error correction (QEC) bridges this gap, but depends on decoders being simultaneously fast, accurate, and scalable. This combination of requirements remains unmet by a machine-learning decoder, nor by any decoder for promising resource-efficient codes such as the color code. Here we introduce AlphaQubit 2, a neural-network decoder that achieves near-optimal logical error rates for both surface and color codes at scale under realistic noise. For the color code, it is orders of magnitude faster than other high-accuracy decoders. We demonstrate real-time decoding faster than 1μs per cycle on commercial accelerators: for the surface code to distance 11, with better accuracy than leading real-time decoders; and the first real-time decoding of the color code to distance 9. These results support the practical application of a wider class of promising QEC codes, and establish a credible path towards high-accuracy, real-time neural decoding at the scales required for fault-tolerant quantum computation.", "authors": ["Andrew W. Senior", "Thomas Edlich", "Francisco J. H. Heras", "Lei M. Zhang", "Oscar Higgott", "James S. Spencer", "Taylor Applebaum", "Sam Blackwell", "Justin Ledford", "Akvilė Žemgulytė", "Augustin Žídek", "Noah Shutty", "Andrew Cowie", "Yin Li", "George Holland", "Peter Brooks", "Charlie Beattie", "Michael Newman", "Alex Davies", "Cody Jones", "Sergio Boixo", "Hartmut Neven", "Pushmeet Kohli", "Johannes Bausch"], "categories": ["quant-ph", "cs.LG"], "fields_of_study": ["Computer Science", "Physics"], "published_date": "2025-12-08", "url": "https://arxiv.org/abs/2512.07737", "pdf_url": "https://arxiv.org/pdf/2512.07737v2", "arxiv_id": "2512.07737", "doi": "10.48550/arXiv.2512.07737", "citation_count": 19, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3976}
{"id": "442c163ae3ab78445c11b565f4c68cbc558b01e1bf82b7e0aa51bd34d7a50687", "sources": ["arxiv", "semantic_scholar"], "title": "SJD++: Improved Speculative Jacobi Decoding for Training-free Acceleration of Discrete Auto-regressive Text-to-Image Generation", "abstract": "Large autoregressive models can generate high-quality, high-resolution images but suffer from slow generation speed, because these models require hundreds to thousands of sequential forward passes for next-token prediction during inference. To accelerate autoregressive text-to-image generation, we propose Speculative Jacobi Decoding++ (SJD++), a training-free probabilistic parallel decoding algorithm. Unlike traditional next-token prediction, SJD++ performs multi-token prediction in each forward pass, drastically reducing generation steps. Specifically, it integrates the iterative multi-token prediction mechanism from Jacobi decoding, with the probabilistic drafting-and-verification mechanism from speculative sampling. More importantly, for further acceleration, SJD++ reuses high-confidence draft tokens after each verification phase instead of resampling them all. We conduct extensive experiments on several representative autoregressive text-to-image generation models and demonstrate that SJD++ achieves $2\\times$ to $3\\times$ inference latency reduction and $2\\times$ to $7\\times$ step compression, while preserving visual quality with no observable degradation.", "authors": ["Yao Teng", "Zhihuan Jiang", "Han Shi", "Xian Liu", "Xuefei Ning", "Guohao Dai", "Yu Wang", "Zhenguo Li", "Xihui Liu"], "categories": ["cs.CV"], "fields_of_study": ["Medicine", "Computer Science"], "published_date": "2025-12-08", "url": "https://arxiv.org/abs/2512.07503", "pdf_url": "https://arxiv.org/pdf/2512.07503v1", "arxiv_id": "2512.07503", "doi": "10.48550/arXiv.2512.07503", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE Transactions on Pattern Analysis and Machine Intelligence", "quality_score": 0.3976}
{"id": "be5a16eca63b9912582b7752969dbd11d89e3cda9093c17715c66ab5cd39ff9c", "sources": ["arxiv", "semantic_scholar"], "title": "SpecPV: Improving Self-Speculative Decoding for Long-Context Generation via Partial Verification", "abstract": "Growing demands from tasks like code generation, deep reasoning, and long-document understanding have made long-context generation a crucial capability for large language models (LLMs). Speculative decoding is one of the most direct and effective approaches for accelerating generation. It follows a draft-verify paradigm, where a lightweight draft model proposes several candidate tokens and the target model verifies them. However, we find that as the context length grows, verification becomes the dominant bottleneck. To further accelerate speculative decoding in long-context generation, we introduce SpecPV, a self-speculative decoding approach that performs fast verification using partial key-value states (KV) and periodically applies full verification to eliminate accumulated errors. We validate SpecPV across multiple long-context benchmarks and models, including LLaMA-3.1-8B-Instruct and Qwen3-series. Experimental results show that SpecPV achieves up to 6x decoding speedup over standard autoregressive decoding with minor degradation.", "authors": ["Zhendong Tan", "Xingjun Zhang", "Chaoyi Hu", "Junjie Peng", "Kun Xia"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-02", "url": "https://arxiv.org/abs/2512.02337", "pdf_url": "https://arxiv.org/pdf/2512.02337v1", "arxiv_id": "2512.02337", "doi": "10.48550/arXiv.2512.02337", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3907}
{"id": "bac73db2cf10fc389dccbcb151df7fbb6920d13207327a04aa286b38972c4e3b", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Large-Scale Reasoning Model Inference with Sparse Self-Speculative Decoding", "abstract": "Reasoning language models have demonstrated remarkable capabilities on challenging tasks by generating elaborate chain-of-thought (CoT) solutions. However, such lengthy generation shifts the inference bottleneck from compute-bound to memory-bound. To generate each token, the model applies full attention to all previously generated tokens, requiring memory access to an increasingly large KV-Cache. Consequently, longer generations demand more memory access for every step, leading to substantial pressure on memory bandwidth. To address this, we introduce SparseSpec, a speculative decoding framework that reuses the same model as the draft and target models (i.e., self-speculation). SparseSpec features a novel sparse attention mechanism, PillarAttn, as the draft model, which accurately selects critical tokens via elegantly reusing information from the verification stage. Furthermore, SparseSpec co-designs self-speculation with three system innovations: (1) a unified scheduler to batch token drafting and verification, (2) delayed verification for CPU/GPU overlap, and (3) dynamic KV-Cache management to maximize memory utilization. Across various models and datasets, SparseSpec outperforms state-of-the-art solutions, with an up to 2.13x throughput speedup.", "authors": ["Yilong Zhao", "Jiaming Tang", "Kan Zhu", "Zihao Ye", "Chi-Chih Chang", "Chaofan Lin", "Jongseok Park", "Guangxuan Xiao", "Mohamed S. Abdelfattah", "Mingyu Gao", "Baris Kasikci", "Song Han", "Ion Stoica"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-12-01", "url": "https://arxiv.org/abs/2512.01278", "pdf_url": "https://arxiv.org/pdf/2512.01278v1", "arxiv_id": "2512.01278", "doi": "10.48550/arXiv.2512.01278", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3896}
{"id": "1812932d4c0c2e914ed10c58122b86492a9f2bfc70356a76770aae5a07324c65", "sources": ["arxiv", "semantic_scholar"], "title": "Training-Free Loosely Speculative Decoding: Accepting Semantically Correct Drafts Beyond Exact Match", "abstract": "Large language models (LLMs) achieve strong performance across diverse tasks but suffer from high inference latency due to their autoregressive generation. Speculative Decoding (SPD) mitigates this issue by verifying candidate tokens in parallel from a smaller draft model, yet its strict exact-match verification discards many semantically valid continuations. Moreover, existing training-based SPD methods often suffer from performance degradation on out-of-distribution (OOD) tasks. To this end, we propose Training-Free Loosely Speculative Decoding (FLy), a novel method that loosens the rigid verification criterion by leveraging the target model's self-corrective behavior to judge whether a draft-target mismatch remains semantically valid. FLy introduces a two-tier mechanism: an entropy-level gate that identifies whether the current token allows multiple plausible alternatives or is nearly deterministic, and a token-level deferred window that distinguishes genuine errors from differently worded yet semantically correct variants. To further reduce latency, we design a multi-level acceleration strategy that accelerates not only the target model but also the drafter itself. Owing to its training-free design, FLy composes seamlessly with arbitrary draft-target pairs and generalizes across models and domains without hyperparameter re-tuning. Experiments show that FLy preserves more than 99% of the target model's accuracy while achieving an average 2.81x speedup on Llama-3.1-70B-Instruct and 5.07x speedup on the 405B variant. Notably, on out-of-domain datasets, our method remains highly effective and outperforms the training-based method EAGLE-3 by 1.62x.", "authors": ["Jinze Li", "Yixing Xu", "Guanchen Li", "Shuo Yang", "Jinfeng Xu", "Xuanwu Yin", "Dong Li", "Edith C. H. Ngai", "Emad Barsoum"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-28", "url": "https://arxiv.org/abs/2511.22972", "pdf_url": "https://arxiv.org/pdf/2511.22972v3", "arxiv_id": "2511.22972", "doi": "10.48550/arXiv.2511.22972", "citation_count": 8, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3861}
{"id": "7c109b21d12fa51179d73bf274895c52d4ac5a25433627e6e32fe929aa2f80c7", "sources": ["arxiv", "semantic_scholar"], "title": "DSD: A Distributed Speculative Decoding Solution for Edge-Cloud Agile Large Model Serving", "abstract": "Large language model (LLM) inference often suffers from high decoding latency and limited scalability across heterogeneous edge-cloud environments. Existing speculative decoding (SD) techniques accelerate token generation but remain confined to single-node execution. We propose DSD, a distributed speculative decoding framework that extends SD to multi-device deployments through coordinated draft-target execution. Given the lack of prior work on simulating this paradigm, we first introduce DSD-Sim, a discrete-event simulator that captures network, batching, and scheduling dynamics. Building on insights from DSD-Sim, we further design an Adaptive Window Control (AWC) policy that dynamically adjusts speculation window size to optimize throughput. Experiments across diverse workloads show that DSD achieves up to 1.1x speedup and 9.7% higher throughput over existing SD baselines, enabling agile and scalable LLM serving across edge and cloud.", "authors": ["Fengze Yu", "Leshu Li", "Brad McDanel", "Sai Qian Zhang"], "categories": ["cs.LG", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-26", "url": "https://arxiv.org/abs/2511.21669", "pdf_url": "https://arxiv.org/pdf/2511.21669v2", "arxiv_id": "2511.21669", "doi": "10.48550/arXiv.2511.21669", "citation_count": 4, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3839}
{"id": "c1e7923edff546035821afb67aea2c6cea26aa2ea99eea9cceceaaef82c18b42", "sources": ["arxiv", "semantic_scholar"], "title": "Scaling LLM Speculative Decoding: Non-Autoregressive Forecasting in Large-Batch Scenarios", "abstract": "Speculative decoding accelerates LLM inference by utilizing otherwise idle computational resources during memory-to-chip data transfer. Current speculative decoding methods typically assume a considerable amount of available computing power, then generate a complex and massive draft tree using a small autoregressive language model to improve overall prediction accuracy. However, methods like batching have been widely applied in mainstream model inference systems as a superior alternative to speculative decoding, as they compress the available idle computing power. Therefore, performing speculative decoding with low verification resources and low scheduling costs has become an important research problem. We believe that more capable models that allow for parallel generation on draft sequences are what we truly need. Recognizing the fundamental nature of draft models to only generate sequences of limited length, we propose SpecFormer, a novel architecture that integrates unidirectional and bidirectional attention mechanisms. SpecFormer combines the autoregressive model's ability to extract information from the entire input sequence with the parallel generation benefits of non-autoregressive models. This design eliminates the reliance on large prefix trees and achieves consistent acceleration, even in large-batch scenarios. Through lossless speculative decoding experiments across models of various scales, we demonstrate that SpecFormer sets a new standard for scaling LLM inference with lower training demands and reduced computational costs.", "authors": ["Luohe Shi", "Zuchao Li", "Lefei Zhang", "Baoyuan Qi", "Guoming Liu", "Hai Zhao"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-25", "url": "https://arxiv.org/abs/2511.20340", "pdf_url": "https://arxiv.org/pdf/2511.20340v1", "arxiv_id": "2511.20340", "doi": "10.48550/arXiv.2511.20340", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "AAAI Conference on Artificial Intelligence", "quality_score": 0.3827}
{"id": "4b9f8426458debdbea1d191d159c55a38276c6df91da7a3f3883f67f4fb88a40", "sources": ["arxiv", "semantic_scholar"], "title": "Reducing Latency of LLM Search Agent via Speculation-based Algorithm-System Co-Design", "abstract": "LLM-based search agents achieve strong performance but suffer from severe latency, as each step requires serialized LLM reasoning followed by action of tool execution. We revisit this bottleneck through the lens of speculation. While traditional predict-verify speculation paradigm can break serial execution, its benefit remains limited, as it retains the full original workload and adds extra inference overhead. We observe that early agent steps often involve simple evidence-gathering, where correct actions can often be predicted without full reasoning. Building on these observations, we present SPAgent, an algorithm-system co-design framework that expands the role of speculation in search agents to reduce latency. Algorithmically, SPAgent introduces a two-phase adaptive speculation mechanism that selectively omits verification when safe. System-wise, a two-level scheduler regulates speculative requests based on engine load to ensure speculation remains beneficial. We implement SPAgent in real-world systems. Across extensive experimental settings, SPAgent achieves up to $1.65\\times$ end-to-end speedup while maintaining same or even achieving higher accuracy, enabling practical deployment of multi-step search agents.", "authors": ["Zixiao Huang", "Wen Zeng", "Tianyu Fu", "Tengxuan Liu", "Yizhou Sun", "Ke Hong", "Xinhao Yang", "Chengchun Liu", "Yan Li", "Quanlu Zhang", "Guohao Dai", "Zhenhua Zhu", "Yu Wang"], "categories": ["cs.AI", "cs.LG", "cs.PF"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-25", "url": "https://arxiv.org/abs/2511.20048", "pdf_url": "https://arxiv.org/pdf/2511.20048v1", "arxiv_id": "2511.20048", "doi": "10.48550/arXiv.2511.20048", "citation_count": 1, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3827}
{"id": "0669f1c82d8e4cf5424ef9f2ab9e08268c702e0692afc844fa2d2e12c07c7380", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Time Series Foundation Models with Speculative Decoding", "abstract": "Modern web applications--from real-time content recommendation and dynamic pricing to CDN optimization--increasingly rely on time-series forecasting to deliver personalized experiences to billions of users. Large-scale Transformer-based models have achieved state-of-the-art performance in time-series forecasting but suffer from high computational costs, limiting their deployment in latency-sensitive web applications. To address this challenge, we propose a general inference acceleration framework that adapts speculative decoding to autoregressive time-series models. Our approach employs a smaller \"draft\" model to propose future time-series patches, which are then verified in parallel by a larger \"target\" model, reducing the number of sequential forward passes required. We address key technical challenges in adapting this technique from discrete language tokens to continuous time-series distributions, including the design of acceptance criteria for multivariate Gaussian patches and practical variants that balance efficiency with accuracy. Through experiments on time series forecasting benchmarks relevant to web applications, we demonstrate significant inference speedups while maintaining competitive accuracy. The framework requires no architectural modifications to existing foundation models, making it immediately applicable to accelerate deployed time-series forecasting systems. Our implementation can be found at https://github.com/PranavSubbaraman/STRIDE", "authors": ["Pranav Subbaraman", "Fang Sun", "Yue Yao", "Huacong Tang", "Xiao Luo", "Yizhou Sun"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-22", "url": "https://arxiv.org/abs/2511.18191", "pdf_url": "https://arxiv.org/pdf/2511.18191v1", "arxiv_id": "2511.18191", "doi": "10.48550/arXiv.2511.18191", "citation_count": 1, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/PranavSubbaraman/STRIDE", "venue": "arXiv.org", "quality_score": 0.5861}
{"id": "42300a828df305c43b757c35800a31c6e2f426c43adb8614194d31602809506c", "sources": ["arxiv", "semantic_scholar"], "title": "Fast LLM Post-training via Decoupled and Fastest-of-N Speculation", "abstract": "Rollout dominates the training time in large language model (LLM) post-training, where the trained model is used to generate tokens given a batch of prompts. This work, SpecActor, achieves fast rollout with speculative decoding that deploys a fast draft path to accelerate the unparallelizable generation, while the correctness is guaranteed by fast parallel verification of the outputs with the original model. SpecActor addresses two foundational challenges that hinder speculation efficiency: (1) a Decoupled speculation method that overcomes the computation inefficiency issue when executing speculative decoding with relative large per-worker batch size -- a common configuration in training but unfriendly to speculation, and (2) a Fastest-of-N speculation method that selects and combines different draft methods according to the rollout progress to approximate the optimal draft method even when the best one is unknown a priori. Extensive evaluations on production traces show that SpecActor accelerates mean rollout speed by 2.0--2.4x, with up to 2.7x speedup, over common post-training baselines. The results are consistent across both dense and MoE models and across different RL algorithms. Notably, SpecActor is 1.1--2.6x faster compared to vanilla speculative rollout in different traces. The accelerated rollout achieves 1.4--2.3x faster end-to-end training time.", "authors": ["Rongxin Cheng", "Kai Zhou", "Xingda Wei", "Siyuan Liu", "Mingcong Han", "Mingjing Ai", "Yeju Zhou", "Baoquan Zhong", "Wencong Xiao", "Rong Chen", "Haibo Chen"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-20", "url": "https://arxiv.org/abs/2511.16193", "pdf_url": "https://arxiv.org/pdf/2511.16193v3", "arxiv_id": "2511.16193", "doi": "10.48550/arXiv.2511.16193", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.377}
{"id": "4868fcf1f532f7e9defcefad5b634557cd8b43275d1cf72889cc0bfb8b6bd24b", "sources": ["arxiv", "semantic_scholar"], "title": "Comparative Analysis of Large Language Model Inference Serving Systems: A Performance Study of vLLM and HuggingFace TGI", "abstract": "The deployment of Large Language Models (LLMs) in production environments requires efficient inference serving systems that balance throughput, latency, and resource utilization. This paper presents a comprehensive empirical evaluation of two prominent open-source LLM serving frameworks: vLLM and HuggingFace Text Generation Inference (TGI). We benchmark these systems across multiple dimensions including throughput performance, end-to-end latency, GPU memory utilization, and scalability characteristics using LLaMA-2 models ranging from 7B to 70B parameters. Our experiments reveal that vLLM achieves up to 24x higher throughput than TGI under high-concurrency workloads through its novel PagedAttention mechanism, while TGI demonstrates lower tail latencies for interactive single-user scenarios. We provide detailed performance profiles for different deployment scenarios and offer practical recommendations for system selection based on workload characteristics. Our findings indicate that the choice between these frameworks should be guided by specific use-case requirements: vLLM excels in high-throughput batch processing scenarios, while TGI is better suited for latency-sensitive interactive applications with moderate concurrency.", "authors": ["Saicharan Kolluru"], "categories": ["cs.LG", "cs.DC", "cs.PF"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-17", "url": "https://arxiv.org/abs/2511.17593", "pdf_url": "https://arxiv.org/pdf/2511.17593v1", "arxiv_id": "2511.17593", "doi": "10.48550/arXiv.2511.17593", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5773}
{"id": "a47aaa8701ad8543fe841c4badb21ed10e029b487c238c6b4e48cdc2268d739b", "sources": ["arxiv", "semantic_scholar"], "title": "Beat the long tail: Distribution-Aware Speculative Decoding for RL Training", "abstract": "Reinforcement learning(RL) post-training has become essential for aligning large language models (LLMs), yet its efficiency is increasingly constrained by the rollout phase, where long trajectories are generated token by token. We identify a major bottleneck:the long-tail distribution of rollout lengths, where a small fraction of long generations dominates wall clock time and a complementary opportunity; the availability of historical rollouts that reveal stable prompt level patterns across training epochs. Motivated by these observations, we propose DAS, a Distribution Aware Speculative decoding framework that accelerates RL rollouts without altering model outputs. DAS integrates two key ideas: an adaptive, nonparametric drafter built from recent rollouts using an incrementally maintained suffix tree, and a length aware speculation policy that allocates more aggressive draft budgets to long trajectories that dominate makespan. This design exploits rollout history to sustain acceptance while balancing base and token level costs during decoding. Experiments on math and code reasoning tasks show that DAS reduces rollout time up to 50% while preserving identical training curves, demonstrating that distribution-aware speculative decoding can significantly accelerate RL post training without compromising learning quality.", "authors": ["Zelei Shao", "Vikranth Srivatsa", "Sanjana Srivastava", "Qingyang Wu", "Alpay Ariyak", "Xiaoxia Wu", "Ameen Patel", "Jue Wang", "Percy Liang", "Tri Dao", "Ce Zhang", "Yiying Zhang", "Ben Athiwaratkun", "Chenfeng Xu", "Junxiong Wang"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-17", "url": "https://arxiv.org/abs/2511.13841", "pdf_url": "https://arxiv.org/pdf/2511.13841v1", "arxiv_id": "2511.13841", "doi": "10.48550/arXiv.2511.13841", "citation_count": 9, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3735}
{"id": "0a5f3ed1432b1b847bb5be16f5ba398f6f9f26621ba4b1591441ee34c4518193", "sources": ["arxiv", "semantic_scholar"], "title": "Cacheback: Speculative Decoding With Nothing But Cache", "abstract": "We present Cacheback Decoding, a training-free and model-agnostic speculative decoding method that exploits the locality in language to accelerate Large Language Model (LLM) inference. Cacheback leverages only Least Recently Used (LRU) cache tables of token n-grams to generate draft sequences. Cacheback achieves state-of-the-art performance among comparable methods despite its minimalist design, and its simplicity allows easy integration into existing systems. Cacheback also shows potential for fast adaptation to new domains.", "authors": ["Zhiyao Ma", "In Gim", "Lin Zhong"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-15", "url": "https://arxiv.org/abs/2511.21699", "pdf_url": "https://arxiv.org/pdf/2511.21699v1", "arxiv_id": "2511.21699", "doi": "10.18653/v1/2025.emnlp-main.1581", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.3713}
{"id": "5c970964f67e211782934047ac1149eaec7e55165aee8215dcf3bbb1c8ff7eae", "sources": ["arxiv", "semantic_scholar"], "title": "Striking the Right Balance between Compute and Copy: Improving LLM Inferencing Under Speculative Decoding", "abstract": "With the skyrocketing costs of GPUs and their virtual instances in the cloud, there is a significant desire to use CPUs for large language model (LLM) inference. KV cache update, often implemented as allocation, copying, and in-place strided update for each generated token, incurs significant overhead. As the sequence length increases, the allocation and copy overheads dominate the performance. Alternate approaches may allocate large KV tensors upfront to enable in-place updates, but these matrices (with zero-padded rows) cause redundant computations. In this work, we propose a new KV cache allocation mechanism called Balancing Memory and Compute (BMC). BMC allocates, once every r iterations, KV tensors with r redundant rows, allowing in-place update without copy overhead for those iterations, but at the expense of a small amount of redundant computation. Second, we make an interesting observation that the extra rows allocated in the KV tensors and the resulting redundant computation can be repurposed for Speculative Decoding (SD) that improves token generation efficiency. Last, BMC represents a spectrum of design points with different values of r. To identify the best-performing design point(s), we derive a simple analytical model for BMC. The proposed BMC method achieves an average throughput acceleration of up to 3.2x over baseline HuggingFace (without SD). Importantly when we apply BMC with SD, it results in an additional speedup of up to 1.39x, over and above the speedup offered by SD. Further, BMC achieves a throughput acceleration of up to 1.36x and 2.29x over state-of-the-art inference servers vLLM and DeepSpeed, respectively. Although the BMC technique is evaluated extensively across different classes of CPUs (desktop and server class), we also evaluate the scheme with GPUs and demonstrate that it works well for GPUs.", "authors": ["Arun Ramachandran", "Ramaswamy Govindarajan", "Murali Annavaram", "Prakash Raghavendra", "Hossein Entezari Zarch", "Lei Gao", "Chaoyi Jiang"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-15", "url": "https://arxiv.org/abs/2511.12031", "pdf_url": "https://arxiv.org/pdf/2511.12031v1", "arxiv_id": "2511.12031", "doi": "10.48550/arXiv.2511.12031", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3713}
{"id": "f380d052435b23fff7004248d2e7f09ba440c028ec4121c72a7c957cdaa69fd2", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding in Decentralized LLM Inference: Turning Communication Latency into Computation Throughput", "abstract": "Speculative decoding accelerates large language model (LLM) inference by using a lightweight draft model to propose tokens that are later verified by a stronger target model. While effective in centralized systems, its behavior in decentralized settings, where network latency often dominates compute, remains under-characterized. We present Decentralized Speculative Decoding (DSD), a plug-and-play framework for decentralized inference that turns communication delay into useful computation by verifying multiple candidate tokens in parallel across distributed nodes. We further introduce an adaptive speculative verification strategy that adjusts acceptance thresholds by token-level semantic importance, delivering an additional 15% to 20% end-to-end speedup without retraining. In theory, DSD reduces cross-node communication cost by approximately (N-1)t1(k-1)/k, where t1 is per-link latency and k is the average number of tokens accepted per round. In practice, DSD achieves up to 2.56x speedup on HumanEval and 2.59x on GSM8K, surpassing the Eagle3 baseline while preserving accuracy. These results show that adapting speculative decoding for decentralized execution provides a system-level optimization that converts network stalls into throughput, enabling faster distributed LLM inference with no model retraining or architectural changes.", "authors": ["Jingwei Song", "Wanyi Chen", "Xinyuan Song", " Max", "Chris Tong", "Gufeng Chen", "Tianyi Zhao", "Eric Yang", "Bill Shi", "Lynn Ai"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-13", "url": "https://arxiv.org/abs/2511.11733", "pdf_url": "https://arxiv.org/pdf/2511.11733v1", "arxiv_id": "2511.11733", "doi": "10.48550/arXiv.2511.11733", "citation_count": 2, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.369}
{"id": "876b9d510c25384844b4e76cba4cda89cca09c751648aaefd4de6c661b391d10", "sources": ["arxiv", "semantic_scholar"], "title": "Steering Pretrained Drafters during Speculative Decoding", "abstract": "Speculative decoding accelerates language model inference by separating generation into fast drafting and parallel verification. Its main limitation is drafter-verifier misalignment, which limits token acceptance and reduces overall effectiveness. While small drafting heads trained from scratch compensate with speed, they struggle when verification dominates latency or when inputs are out of distribution. In contrast, pretrained drafters, though slower, achieve higher acceptance rates thanks to stronger standalone generation capabilities, making them competitive when drafting latency is negligible relative to verification or communication overhead. In this work, we aim to improve the acceptance rates of pretrained drafters by introducing a lightweight dynamic alignment mechanism: a steering vector computed from the verifier's hidden states and injected into the pretrained drafter. Compared to existing offline alignment methods such as distillation, our approach boosts the number of accepted tokens by up to 35\\% under standard sampling and 22\\% under greedy sampling, all while incurring negligible computational overhead. Importantly, our approach can be retrofitted to existing architectures and pretrained models, enabling rapid adoption.", "authors": ["Frédéric Berdoz", "Peer Rheinboldt", "Roger Wattenhofer"], "categories": ["cs.LG", "cs.PF"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-13", "url": "https://arxiv.org/abs/2511.09844", "pdf_url": "https://arxiv.org/pdf/2511.09844v1", "arxiv_id": "2511.09844", "doi": "10.48550/arXiv.2511.09844", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "AAAI Conference on Artificial Intelligence", "quality_score": 0.369}
{"id": "c253e46bf368d665c679cb9f5cee11730d06b8b6bdfdd8a9f7580f5523520550", "sources": ["arxiv", "semantic_scholar"], "title": "Principled Coarse-Grained Acceptance for Speculative Decoding in Speech", "abstract": "Speculative decoding accelerates autoregressive speech generation by letting a fast draft model propose tokens that a larger target model verifies. However, for speech LLMs that generate acoustic tokens, exact token matching is overly restrictive: many discrete tokens are acoustically or semantically interchangeable, reducing acceptance rates and limiting speedups. We introduce Principled Coarse-Graining (PCG), which verifies proposals at the level of Acoustic Similarity Groups (ASGs) derived from the target model's embedding space. By splitting each token's probability mass across the overlapping groups that contain it, we define an overlap-aware coarse-grained distribution and perform rejection sampling on the resulting group variable. This yields an exactness guarantee at the group level while allowing the accepted draft token to stand in for any member of the group in practice. On LibriTTS, PCG increases acceptance and throughput relative to standard speculative decoding and prior speech-specific relaxations while maintaining intelligibility and speaker similarity. These results suggest acoustically aware, group-level acceptance as a simple and general way to accelerate speech token generation while maintaining speech quality.", "authors": ["Moran Yanuka", "Paul Dixon", "Eyal Finkelshtein", "Daniel Rotman", "Raja Giryes"], "categories": ["eess.AS", "cs.LG"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2025-11-05", "url": "https://arxiv.org/abs/2511.13732", "pdf_url": "https://arxiv.org/pdf/2511.13732v4", "arxiv_id": "2511.13732", "doi": "10.48550/arXiv.2511.13732", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE International Conference on Acoustics, Speech, and Signal Processing", "quality_score": 0.3598}
{"id": "c3f745d68b303decafed4e49057542b350d1f5eb3477a1d7f8fa0e42d08911cf", "sources": ["arxiv", "semantic_scholar"], "title": "Collaborative Large Language Model Inference via Resource-Aware Parallel Speculative Decoding", "abstract": "The growing demand for on-device large language model (LLM) inference highlights the need for efficient mobile edge computing (MEC) solutions, especially in resource-constrained settings. Speculative decoding offers a promising solution by partitioning token generation between a lightweight draft model on mobile devices and a powerful target model on edge servers, but suffers from communication overhead and asynchronous delays. This paper is the first to propose a unified framework that jointly optimizes user association and resource allocation (UARA) to support efficient parallel speculative decoding. We solve the UARA problem using a multi-agent deep reinforcement learning algorithm. To evaluate our approach under realistic conditions, we conduct experiments using the Sionna simulator. Results show that our method achieves up to 28.0% and an average of 23.7% reduction in end-to-end latency without compromising inference accuracy, enabling scalable and low-latency LLM services in MEC systems.", "authors": ["Jungyeon Koh", "Hyun Jong Yang"], "categories": ["cs.LG", "eess.SP"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2025-11-03", "url": "https://arxiv.org/abs/2511.01695", "pdf_url": "https://arxiv.org/pdf/2511.01695v4", "arxiv_id": "2511.01695", "doi": "10.48550/arXiv.2511.01695", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3575}
{"id": "82e7aad187c8d1079c665a2cf8dc154252b87c5d07f3d01b64cb9697c3dc5aa4", "sources": ["arxiv", "semantic_scholar"], "title": "TapOut: A Bandit-Based Approach to Dynamic Speculative Decoding", "abstract": "Speculative decoding accelerates LLMs by using a lightweight draft model to generate tokens autoregressively before verifying them in parallel with a larger target model. However, determining the optimal number of tokens to draft remains a key challenge limiting the approach's effectiveness. Dynamic speculative decoding aims to intelligently decide how many tokens to draft to achieve maximum speedups. Existing methods often rely on hand-tuned, sensitive thresholds (e.g., token entropy), which are costly to set and generalize poorly across models and domains. We propose TapOut, an online, training-free, plug-and-play algorithm for dynamic speculation policy selection using multi-armed bandits. Our approach employs a meta-algorithm that selects among multiple parameter-free dynamic speculation strategies based on past reward and exploration. We conduct extensive experiments across diverse model pairs and datasets, showing that TapOut achieves competitive or superior speedups compared to well-established dynamic speculation baselines without any hyperparameter tuning.", "authors": ["Aditya Sridhar", "Nish Sinnadurai", "Sean Lie", "Vithursan Thangarasa"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-03", "url": "https://arxiv.org/abs/2511.02017", "pdf_url": "https://arxiv.org/pdf/2511.02017v1", "arxiv_id": "2511.02017", "doi": "10.48550/arXiv.2511.02017", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3575}
{"id": "fb1eb6085b163a354ea9cd43f252586e3dcac20d4d22b4f74bfd3776ae10f2aa", "sources": ["arxiv", "semantic_scholar"], "title": "When, What, and How: Rethinking Retrieval-Enhanced Speculative Decoding", "abstract": "Speculative decoding (SD) has emerged as an effective technique to accelerate large language model (LLM) inference without compromising output quality. However, the achievable speedup largely depends on the effectiveness of the drafting model. While model-based methods like EAGLE-2 are accurate but costly, retrieval-enhanced methods like SAM-Decoding rely on heuristic switching strategies that often trigger unnecessary retrievals. To address this, we propose ReSpec (\\textbf{Re}trieval-enhanced \\textbf{Spe}culative Decoding), a novel framework that transforms heuristic drafter switching into adaptive decision-making. ReSpec features three core innovations: 1) An \\textbf{entropy-guided adaptive trigger} quantifies contextual predictability to initiate retrieval only when uncertainty is low, avoiding costly low-quality speculations. 2) A \\textbf{feedback-driven candidate selection} leverages historical feedback to organize multiple high-quality candidates for parallel verification, maximizing retrieval utility. 3) A source-aware \\textbf{relaxed verification strategy} applies strict checks to model-generated drafts while using a relaxed verification for retrieved drafts, achieving a better balance between accuracy and efficiency. Extensive experiments on Spec-Bench demonstrate that ReSpec achieves state-of-the-art acceleration,outperforming EAGLE-2 and SAM-Decoding by over $33\\%$ and $25\\%$, respectively, while maintaining output quality.", "authors": ["Min Fang", "Zhihui Fu", "Qibin Zhao", "Jun Wang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-03", "url": "https://arxiv.org/abs/2511.01282", "pdf_url": "https://arxiv.org/pdf/2511.01282v1", "arxiv_id": "2511.01282", "doi": "10.48550/arXiv.2511.01282", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3575}
{"id": "8cb9b7be3c81f03ae0f0a4ca5244c7510af10ca872daa5ddfa281a43a50b1d79", "sources": ["arxiv", "semantic_scholar"], "title": "SpecDiff-2: Scaling Diffusion Drafter Alignment For Faster Speculative Decoding", "abstract": "Speculative decoding has become the standard approach for accelerating Large Language Model (LLM) inference. It exploits a lossless draft-then-verify procedure to circumvent the latency of autoregressive decoding, achieving impressive speed-ups. Yet, current speculative decoding approaches remain limited by two fundamental bottlenecks: (1) the autoregressive dependency during drafting which limits parallelism, and (2) frequent rejections of draft tokens caused by misalignment between the draft and verify models. This paper proposes SpecDiff-2, a novel framework to jointly address these two bottlenecks. It leverages discrete diffusion as a non-autoregressive drafter to address bottleneck (1) and develops novel techniques to calibrate discrete diffusion drafters with autoregressive verifiers, addressing bottleneck (2). Experimental results across a comprehensive benchmark suite show that SpecDiff-2 achieves a new state-of-the-art across reasoning, coding, and mathematical benchmarks, improving tokens-per-second by up to an average of +55% over previous baselines and obtaining up to 5.5x average speed-up over standard decoding, without any loss of accuracy.", "authors": ["Jameson Sandler", "Jacob K. Christopher", "Thomas Hartvigsen", "Ferdinando Fioretto"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-01", "url": "https://arxiv.org/abs/2511.00606", "pdf_url": "https://arxiv.org/pdf/2511.00606v2", "arxiv_id": "2511.00606", "doi": "10.48550/arXiv.2511.00606", "citation_count": 10, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3552}
{"id": "5fa7d9ee24b82123aeeb6c8d91cc117789bdd3d684949b7f811080298a5c308a", "sources": ["arxiv", "semantic_scholar"], "title": "Reject Only Critical Tokens: Pivot-Aware Speculative Decoding", "abstract": "Speculative Decoding (SD) ensures that the output matches the target model's distribution exactly. However, we argue that this distribution matching requirement is too stringent and results in unnecessarily low acceptance rates, limiting potential speedups. Instead, we advocate a reformulation of the decoding objective: the proposed decoding strategy should match the expected utility, i.e., the task-specific performance, of the target model. This perspective also aligns better with real-world use cases of LLMs, where utility (e.g., code correctness, factual accuracy) is often more important than sampling distribution. Based on this reformulation, we propose a novel decoding strategy: Pivot-Aware Speculative Decoding, which rejects only those tokens that would lead to a utility drop in the final output. We refer to these critical tokens as pivot tokens. We propose a method for labeling tokens as pivotal or non-pivotal and train a lightweight classifier to detect them. This method can be viewed as a relaxed version of standard SD, which offers much higher acceptance while preserving utility. We evaluate our method across various datasets, demonstrating that we can achieve up to $2.5\\times$ speedup with comparable utility. Source code is available at https://github.com/amir-zsh/PAD.", "authors": ["Amir Ziashahabi", "Yavuz Faruk Bakman", "Duygu Nur Yaldiz", "Mostafa El-Khamy", "Sai Praneeth Karimireddy", "Salman Avestimehr"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-11-01", "url": "https://arxiv.org/abs/2511.00351", "pdf_url": "https://arxiv.org/pdf/2511.00351v1", "arxiv_id": "2511.00351", "doi": "10.48550/arXiv.2511.00351", "citation_count": 2, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/amir-zsh/PAD", "venue": "arXiv.org", "quality_score": 0.549}
{"id": "8416dc812eaed8f798cbf438e74eadcf685191de4939aa6b78968bd7904f199e", "sources": ["arxiv", "semantic_scholar"], "title": "SpecAttn: Speculating Sparse Attention", "abstract": "Large Language Models (LLMs) face significant computational bottlenecks during inference due to the quadratic complexity of self-attention mechanisms, particularly as context lengths increase. We introduce SpecAttn, a novel training-free approach that seamlessly integrates with existing speculative decoding techniques to enable efficient sparse attention in pre-trained transformers. Our key insight is to exploit the attention weights already computed by the draft model during speculative decoding to identify important tokens for the target model, eliminating redundant computation while maintaining output quality. SpecAttn employs three core techniques: KL divergence-based layer alignment between draft and target models, a GPU-optimized sorting-free algorithm for top-p token selection from draft attention patterns, and dynamic key-value cache pruning guided by these predictions. By leveraging the computational work already performed in standard speculative decoding pipelines, SpecAttn achieves over 75% reduction in key-value cache accesses with a mere 15.29% increase in perplexity on the PG-19 dataset, significantly outperforming existing sparse attention methods. Our approach demonstrates that speculative execution can be enhanced to provide approximate verification without significant performance degradation.", "authors": ["Harsh Shah"], "categories": ["cs.CL", "cs.LG", "eess.SY"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2025-10-31", "url": "https://arxiv.org/abs/2510.27641", "pdf_url": "https://arxiv.org/pdf/2510.27641v1", "arxiv_id": "2510.27641", "doi": "10.48550/arXiv.2510.27641", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3541}
{"id": "2e174777c59543ca0d26699225763832092fc478ef07f85e86ed6def7ac342ba", "sources": ["arxiv", "semantic_scholar"], "title": "CAS-Spec: Cascade Adaptive Self-Speculative Decoding for On-the-Fly Lossless Inference Acceleration of LLMs", "abstract": "Speculative decoding has become a widely adopted as an effective technique for lossless inference acceleration when deploying large language models (LLMs). While on-the-fly self-speculative methods offer seamless integration and broad utility, they often fall short of the speed gains achieved by methods relying on specialized training. Cascading a hierarchy of draft models promises further acceleration and flexibility, but the high cost of training multiple models has limited its practical application. In this paper, we propose a novel Cascade Adaptive Self-Speculative Decoding (CAS-Spec) method which constructs speculative draft models by leveraging dynamically switchable inference acceleration (DSIA) strategies, including layer sparsity and activation quantization. Furthermore, traditional vertical and horizontal cascade algorithms are inefficient when applied to self-speculative decoding methods. We introduce a Dynamic Tree Cascade (DyTC) algorithm that adaptively routes the multi-level draft models and assigns the draft lengths, based on the heuristics of acceptance rates and latency prediction. Our CAS-Spec method achieves state-of-the-art acceleration compared to existing on-the-fly speculative decoding methods, with an average speedup from $1.1\\times$ to $2.3\\times$ over autoregressive decoding across various LLMs and datasets. DyTC improves the average speedup by $47$\\% and $48$\\% over cascade-based baseline and tree-based baseline algorithms, respectively. CAS-Spec can be easily integrated into most existing LLMs and holds promising potential for further acceleration as self-speculative decoding techniques continue to evolve.", "authors": ["Zhiyuan Ning", "Jiawei Shao", "Ruge Xu", "Xinfei Guo", "Jun Zhang", "Chi Zhang", "Xuelong Li"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-30", "url": "https://arxiv.org/abs/2510.26843", "pdf_url": "https://arxiv.org/pdf/2510.26843v1", "arxiv_id": "2510.26843", "doi": "10.48550/arXiv.2510.26843", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3529}
{"id": "a07d6ed6c705aa74327ac24e68fc95aaeab3d5422fecd79dbc0663cd379de51a", "sources": ["arxiv", "semantic_scholar"], "title": "Polybasic Speculative Decoding Through a Theoretical Perspective", "abstract": "Inference latency stands as a critical bottleneck in the large-scale deployment of Large Language Models (LLMs). Speculative decoding methods have recently shown promise in accelerating inference without compromising the output distribution. However, existing work typically relies on a dualistic draft-verify framework and lacks rigorous theoretical grounding. In this paper, we introduce a novel \\emph{polybasic} speculative decoding framework, underpinned by a comprehensive theoretical analysis. Specifically, we prove a fundamental theorem that characterizes the optimal inference time for multi-model speculative decoding systems, shedding light on how to extend beyond the dualistic approach to a more general polybasic paradigm. Through our theoretical investigation of multi-model token generation, we expose and optimize the interplay between model capabilities, acceptance lengths, and overall computational cost. Our framework supports both standalone implementation and integration with existing speculative techniques, leading to accelerated performance in practice. Experimental results across multiple model families demonstrate that our approach yields speedup ratios ranging from $3.31\\times$ to $4.01\\times$ for LLaMA2-Chat 7B, up to $3.87 \\times$ for LLaMA3-8B, up to $4.43 \\times$ for Vicuna-7B and up to $3.85 \\times$ for Qwen2-7B -- all while preserving the original output distribution. We release our theoretical proofs and implementation code to facilitate further investigation into polybasic speculative decoding.", "authors": ["Ruilin Wang", "Huixia Li", "Yuexiao Ma", "Xiawu Zheng", "Fei Chao", "Xuefeng Xiao", "Rongrong Ji"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-30", "url": "https://arxiv.org/abs/2510.26527", "pdf_url": "https://arxiv.org/pdf/2510.26527v1", "arxiv_id": "2510.26527", "doi": "10.48550/arXiv.2510.26527", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.3529}
{"id": "e74f829e608057929e2acaa671b6e92e04fb164c8bc6d45cfb82f7ff9ab0ab79", "sources": ["arxiv", "semantic_scholar"], "title": "ReSpec: Towards Optimizing Speculative Decoding in Reinforcement Learning Systems", "abstract": "Adapting large language models (LLMs) via reinforcement learning (RL) is often bottlenecked by the generation stage, which can consume over 75\\% of the training time. Speculative decoding (SD) accelerates autoregressive generation in serving systems, but its behavior under RL training remains largely unexplored. We identify three critical gaps that hinder the naive integration of SD into RL systems: diminishing speedups at large batch sizes, drafter staleness under continual actor updates, and drafter-induced policy degradation. To address these gaps, we present ReSpec, a system that adapts SD to RL through three complementary mechanisms: dynamically tuning SD configurations, evolving the drafter via knowledge distillation, and weighting updates by rollout rewards. On Qwen models (3B--14B), ReSpec achieves up to 4.5x speedup while preserving reward convergence and training stability, providing a practical solution for efficient RL-based LLM adaptation.", "authors": ["Qiaoling Chen", "Zijun Liu", "Peng Sun", "Shenggui Li", "Guoteng Wang", "Ziming Liu", "Yonggang Wen", "Siyuan Feng", "Tianwei Zhang"], "categories": ["cs.LG", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-30", "url": "https://arxiv.org/abs/2510.26475", "pdf_url": "https://arxiv.org/pdf/2510.26475v1", "arxiv_id": "2510.26475", "doi": "10.48550/arXiv.2510.26475", "citation_count": 12, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3529}
{"id": "6c5571a9a8c1a70f1d4d0a1a6b41025a1d1f8b78ac0ef4ef36ec34854191e76d", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Coupled Decoding for Training-Free Lossless Acceleration of Autoregressive Visual Generation", "abstract": "Autoregressive (AR) modeling has recently emerged as a promising new paradigm in visual generation, but its practical adoption is severely constrained by the slow inference speed of per-token generation, which often requires thousands of steps to produce a single sample. While several Speculative Decoding (SD)-based methods have been proposed to solve this problem by generating multiple tokens in a single forward step, they suffer from limited speedup, degraded quality, or require the training of a draft model. To solve these problems, we propose a new training-free, lossless SD framework, Speculative Coupled Decoding (SCD), by extending the recently proposed Speculative Jacobi Decoding (SJD). While SJD shows strong potential for accelerating AR generation by combining Jacobi iteration and SD, we found that its acceptance rate is still significantly limited due to the instability arising from the independent sampling process used during draft token generation. To overcome this, we introduce an information-theoretic approach, Coupling, which stabilizes the drafting trajectory of SJD by maximizing the probability of sampling identical draft tokens across consecutive iterations, significantly enhancing the acceptance rate while preserving its lossless property. Remarkably, this method requires only a single-line modification to the existing algorithm with almost zero overhead, yet achieves substantial performance gains, delivering up to a 4.2x speedup in image generation and 13.6x speedup in video generation compared to standard AR decoding, without any degradation or the need for additional training. The source code is available at https://github.com/junhyukso/SCD", "authors": ["Junhyuk So", "Hyunho Kook", "Chaeyeon Jang", "Eunhyeok Park"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-28", "url": "https://arxiv.org/abs/2510.24211", "pdf_url": "https://arxiv.org/pdf/2510.24211v2", "arxiv_id": "2510.24211", "doi": null, "citation_count": 2, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/junhyukso/SCD", "venue": null, "quality_score": 0.4144}
{"id": "3e1f76c61d3fd3eb0e72cb811966a9cc8143dcfe426fef547478ada559e87e39", "sources": ["arxiv", "semantic_scholar"], "title": "Batch Speculative Decoding Done Right", "abstract": "Speculative decoding must produce outputs distribution identical to standard autoregressive generation-this output equivalence is not an optimization target but the defining criterion of valid speculative decoding. We demonstrate that all existing batch speculative decoding implementations violate this fundamental requirement, producing corrupted outputs ranging from repetitive tokens to gibberish. These failures stem from the ragged tensor problem: sequences in the same batch accept different numbers of draft tokens, desynchronizing position IDs, attention masks, and KV-cache state. We present the first authentic batch speculative decoding framework. We (1) formalize the synchronization invariants that valid batch speculative decoding must satisfy, (2) present EQSPEC, the first algorithm that guarantees output equivalence, and analyze its cost structure to show that alignment overhead grows superlinearly and consumes up to 40\\% of computation, and (3) introduce EXSPEC, which reduces this overhead through cross-batch scheduling that dynamically groups same-length sequences. On SpecBench across Vicuna-7B/68M, Qwen3-8B/0.6B, and GLM-4-9B/0.6B pairs, our methods achieve up to 3x throughput improvement at batch size 8 while maintaining algorithmic correctness. Our methods achieve 95\\% decoding-equivalence, with residual divergence attributable to floating-point non-determinism in GPU inference, not the synchronization failures that cause near-zero equivalence of prior methods. Our code is available at https://github.com/eBay/spec_dec.", "authors": ["Ranran Haoran Zhang", "Soumik Dey", "Ashirbad Mishra", "Hansi Wu", "Binbin Li", "Rui Zhang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-26", "url": "https://arxiv.org/abs/2510.22876", "pdf_url": "https://arxiv.org/pdf/2510.22876v3", "arxiv_id": "2510.22876", "doi": "10.48550/arXiv.2510.22876", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/eBay/spec_dec", "venue": "arXiv.org", "quality_score": 0.5383}
{"id": "424cf2d40c53b479a9b54901307ef04f87102ad6f915e42b195b5e96f5c53a9e", "sources": ["arxiv", "semantic_scholar"], "title": "FastVLM: Self-Speculative Decoding for Fast Vision-Language Model Inference", "abstract": "Vision-language Models (VLMs) have made significant strides in visual understanding and query response generation, but often face challenges of high computational cost and inference latency due to autoregressive decoding. In this work, we introduce an imitation-learning-based Self-Speculative Decoding (SSD) framework, named FastVLM, to address these limitations. Our approach employs a lightweight draft model for token generation in an autoregressive manner, while a full model verifies these tokens non-autoregressively. Accepted tokens proceed seamlessly, while rejected tokens are corrected by the full model and used to guide the draft model's refinement. Through an imitation network, FastVLM enhances the draft model by integrating deeper level insights from the full model's architecture. Also, it maintains the performance integrity of the full model while training the draft model, achieving a balance between efficiency and accuracy. Our method speeds up the inference process by 1.55-1.85x as compared to the final layer with minimal loss in performance.", "authors": ["Divya Jyoti Bajpai", "Manjesh Kumar Hanawal"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-26", "url": "https://arxiv.org/abs/2510.22641", "pdf_url": "https://arxiv.org/pdf/2510.22641v1", "arxiv_id": "2510.22641", "doi": "10.48550/arXiv.2510.22641", "citation_count": 2, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.2217}
{"id": "9f10a6bfd30a7547fe28f913e5dbd921840f2eb08c48a7e6612bd15e92e9dc9b", "sources": ["arxiv", "semantic_scholar"], "title": "Language Ranker: A Lightweight Ranking framework for LLM Decoding", "abstract": "Conventional research on large language models (LLMs) has primarily focused on refining output distributions, while paying less attention to the decoding process that transforms these distributions into final responses. Recent advances, such as scaling the computation of inference time with reward models, have underscored the importance of decoding, but these methods often suffer from high computational costs and limited applicability. In this paper, we revisit LLM generation through the lens of recommender systems, conceptualizing the decoding process as analogous to the ranking stage in recommendation pipelines. From this perspective, we observe that both traditional decoding methods and reward models exhibit clear limitations such as redundancy. Motivated by this insight, we propose Language Ranker, a novel framework that introduces a lightweight module to rerank candidate responses using features extracted by the base model. Experiments across a wide range of tasks show that Language Ranker achieves performance comparable to large-scale reward models, while requiring only <0.5M additional parameters, significantly reducing the computational overhead during both training and inference stages. This highlights the efficiency and effectiveness of our method, showcasing its potential to fully unlock the capabilities of LLMs.", "authors": ["Chenheng Zhang", "Tianqi Du", "Jizhe Zhang", "Mingqing Xiao", "Yifei Wang", "Yisen Wang", "Zhouchen Lin"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-23", "url": "https://arxiv.org/abs/2510.21883", "pdf_url": "https://arxiv.org/pdf/2510.21883v1", "arxiv_id": "2510.21883", "doi": "10.48550/arXiv.2510.21883", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3449}
{"id": "3fa3ac9526ecf22e5b087fbac6759bdf31765418cf52dd9ce391e9af20dd8678", "sources": ["arxiv", "semantic_scholar"], "title": "Not-a-Bandit: Provably No-Regret Drafter Selection in Speculative Decoding for LLMs", "abstract": "Speculative decoding is widely used in accelerating large language model (LLM) inference. In this work, we focus on the online draft model selection problem in speculative decoding. We design an algorithm that provably competes with the best draft model in hindsight for each query in terms of either the token acceptance probability or expected acceptance length. In particular, we show that we can accurately evaluate all draft models, instead of only the chosen model without incurring additional queries to the target model, which allows us to improve exponentially over the existing bandit-based approach as the number of draft models increases. Our approach is generically applicable with any speculative decoding methods (single draft, multi-drafts and draft-trees). Moreover, we design system-efficient versions of online learners and demonstrate that the overhead in computation and latency can be substantially reduced. We conduct extensive experiments on open-source LLMs and diverse datasets, demonstrating that our methods substantially outperform the state-of-the-art EAGLE3 and the BanditSpec baseline in a variety of domains where specialized domain-expert drafters are available, especially when long reasoning chains are required.", "authors": ["Hongyi Liu", "Jiaji Huang", "Zhen Jia", "Youngsuk Park", "Yu-Xiang Wang"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-22", "url": "https://arxiv.org/abs/2510.20064", "pdf_url": "https://arxiv.org/pdf/2510.20064v2", "arxiv_id": "2510.20064", "doi": "10.48550/arXiv.2510.20064", "citation_count": 3, "influential_citation_count": 2, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5312}
{"id": "468aba3d5dbe2f421933777ca7432eb3797d420859add9220c4b914c583549fc", "sources": ["arxiv", "semantic_scholar"], "title": "Fast Inference via Hierarchical Speculative Decoding", "abstract": "Transformer language models generate text autoregressively, making inference latency proportional to the number of tokens generated. Speculative decoding reduces this latency without sacrificing output quality, by leveraging a small draft model to propose tokens that the larger target model verifies in parallel. In practice, however, there may exist a set of potential draft models- ranging from faster but less inaccurate, to slower yet more reliable. We introduce Hierarchical Speculative Decoding (HSD), an algorithm that stacks these draft models into a hierarchy, where each model proposes tokens, and the next larger model verifies them in a single forward pass, until finally the target model verifies tokens. We derive an expression for the expected latency of any such hierarchy and show that selecting the latency-optimal hierarchy can be done in polynomial time. Empirically, HSD gives up to 1.2x speed-up over the best single-draft baseline, demonstrating the practicality of our algorithm in reducing generation latency beyond previous techniques.", "authors": ["Clara Mohri", "Haim Kaplan", "Tal Schuster", "Yishay Mansour", "Amir Globerson"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-22", "url": "https://arxiv.org/abs/2510.19705", "pdf_url": "https://arxiv.org/pdf/2510.19705v2", "arxiv_id": "2510.19705", "doi": "10.48550/arXiv.2510.19705", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3438}
{"id": "568a9bc978c86633380376765bd736d91ae693e0ec7d9fb4cc3f9df7ef325065", "sources": ["arxiv", "semantic_scholar"], "title": "AdaSPEC: Selective Knowledge Distillation for Efficient Speculative Decoders", "abstract": "Speculative Decoding (SD) accelerates large language model inference by employing a small draft model to generate predictions, which are then verified by a larger target model. The effectiveness of SD hinges on the alignment between these models, which is typically enhanced by Knowledge Distillation (KD). However, conventional KD methods aim to minimize the KL divergence between the draft and target models across all tokens, a goal that is misaligned with the true objective of SD, which is to maximize token acceptance rate. Therefore, draft models often struggle to fully assimilate the target model's knowledge due to capacity constraints, leading to suboptimal performance. To address this challenge, we propose AdaSPEC, a novel method that incorporates selective token filtering into the KD process. AdaSPEC utilizes a reference model to identify and filter out difficult-to-fit tokens, enabling the distillation of a draft model that better aligns with the target model on simpler tokens. This approach improves the overall token acceptance rate without compromising generation quality. We evaluate AdaSPEC across diverse tasks, including arithmetic reasoning, instruction-following, coding, and summarization, using model configurations of 31M/1.4B and 350M/2.7B parameters. Our results demonstrate that AdaSPEC consistently outperforms the state-of-the-art DistillSpec method, achieving higher acceptance rates across all tasks (up to 15\\%). The code is publicly available at https://github.com/yuezhouhu/adaspec.", "authors": ["Yuezhou Hu", "Jiaxin Guo", "Xinyu Feng", "Tuo Zhao"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-22", "url": "https://arxiv.org/abs/2510.19779", "pdf_url": "https://arxiv.org/pdf/2510.19779v1", "arxiv_id": "2510.19779", "doi": "10.48550/arXiv.2510.19779", "citation_count": 7, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/yuezhouhu/adaspec", "venue": "arXiv.org", "quality_score": 0.5312}
{"id": "eccaa7e525efb5095c757c9e5121e1353b0ab0a04221e0e590a34958742a4267", "sources": ["arxiv", "semantic_scholar"], "title": "From Quarter to All: Accelerating Speculative LLM Decoding via Floating-Point Exponent Remapping and Parameter Sharing", "abstract": "Large language models achieve impressive performance across diverse tasks but exhibit high inference latency due to their large parameter sizes. While quantization reduces model size, it often leads to performance degradation compared to the full model. Speculative decoding remains lossless but typically incurs extra overheads. We propose SPEQ, an algorithm-hardware co-designed speculative decoding method that uses part of the full-model weight bits to form a quantized draft model, thereby eliminating additional training or storage overhead. A reconfigurable processing element array enables efficient execution of both the draft and verification passes. Experimental results across 15 LLMs and tasks demonstrate that SPEQ achieves speedups of 2.07x, 1.53x, and 1.45x compared over FP16, Olive, and Tender, respectively.", "authors": ["Yushu Zhao", "Yubin Qin", "Yang Wang", "Xiaolong Yang", "Huiming Han", "Shaojun Wei", "Yang Hu", "Shouyi Yin"], "categories": ["cs.AR"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-21", "url": "https://arxiv.org/abs/2510.18525", "pdf_url": "https://arxiv.org/pdf/2510.18525v1", "arxiv_id": "2510.18525", "doi": "10.48550/arXiv.2510.18525", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3426}
{"id": "07ba491326287450a2b5b2a096cc34c9cd80454793b44913b10dd0fc4e6c8b9b", "sources": ["arxiv", "semantic_scholar"], "title": "TokenTiming: A Dynamic Alignment Method for Universal Speculative Decoding Model Pairs", "abstract": "Accelerating the inference of large language models (LLMs) has been a critical challenge in generative AI. Speculative decoding (SD) substantially improves LLM inference efficiency. However, its utility is limited by a fundamental constraint: the draft and target models must share the same vocabulary, thus limiting the herd of available draft models and often necessitating the training of a new model from scratch. Inspired by Dynamic Time Warping (DTW), a classic algorithm for aligning time series, we propose the algorithm TokenTiming for universal speculative decoding. It operates by re-encoding the draft token sequence to get a new target token sequence, and then uses DTW to build a mapping to transfer the probability distributions for speculative sampling. Benefiting from this, our method accommodates mismatched vocabularies and works with any off-the-shelf models without retraining and modification. We conduct comprehensive experiments on various tasks, demonstrating 1.57x speedup. This work enables a universal approach for draft model selection, making SD a more versatile and practical tool for LLM acceleration.", "authors": ["Sibo Xiao", "Jinyuan Fu", "Zhongle Xie", "Lidan Shou"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-17", "url": "https://arxiv.org/abs/2510.15545", "pdf_url": "https://arxiv.org/pdf/2510.15545v4", "arxiv_id": "2510.15545", "doi": "10.48550/arXiv.2510.15545", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.338}
{"id": "298e73b7622084d7adc69fef8929deb05ecbcba4f87219495486c72d0b961898", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Mobile Language Model via Speculative Decoding and NPU-Coordinated Execution", "abstract": "Performing Retrieval-Augmented Generation (RAG) directly on mobile devices is promising for data privacy and responsiveness but is hindered by the architectural constraints of mobile NPUs. Specifically, current hardware struggles with the variable workloads intrinsic to RAG: the transition between processing extensive contexts and generating tokens incurs significant overhead due to static graph constraints, while the memory-bound generation phase leaves computational resources underutilized. In this work, we propose a holistic acceleration framework sd.npu, designed to maximize NPU efficiency for on-device RAG ecosystem. To address the latency caused by NPU graph switching during phase transitions, we introduce a pipelined execution strategy. This approach masks the overhead of model reconfiguration by parallelizing the loading of decoding graphs with the computation of partitioned context chunks (chunked prefill), thereby ensuring continuous execution flow. Furthermore, to mitigate low hardware utilization during the decoding phase, we develop an NPU-centric speculative decoding mechanism. By calibrating generation distributions and extending draft sequences, our method effectively converts idle NPU cycles into valid token throughput. Experiments on commercial smartphones show that our framework significantly outperforms existing baselines, delivering 1.06$\\times$--3.81$\\times$ speedups and 1.07$\\times$--4.71$\\times$ energy savings across various RAG tasks.", "authors": ["Zhiyang Chen", "Daliang Xu", "Haiyang Shen", "Chiheng Lou", "Mengwei Xu", "Shangguang Wang", "Xin Jin", "Yun Ma"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-17", "url": "https://arxiv.org/abs/2510.15312", "pdf_url": "https://arxiv.org/pdf/2510.15312v4", "arxiv_id": "2510.15312", "doi": "10.48550/arXiv.2510.15312", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.338}
{"id": "3eb2c2605a527ff6d7d0ed87ce7cb34d50ab97004508c8102709f00873bf9cbb", "sources": ["arxiv", "semantic_scholar"], "title": "Mirror Speculative Decoding: Breaking the Serial Barrier in LLM Inference", "abstract": "Speculative decoding accelerates LLM inference by using a draft model to look ahead, but gains are capped by the cost of autoregressive draft generation: increasing draft size elevates acceptance rates but introduces additional latency overhead exacerbating the speed-accuracy tradeoff. Prior methods (Medusa, Hydra, EAGLE) partially reduce draft cost but either degrade acceptance or introduce overheads that limit scaling. We present Mirror Speculative Decoding (Mirror-SD), an inference algorithm that breaks the latency-acceptance tradeoff. Mirror-SD launches branch-complete rollouts from early-exit signals in parallel with the target model's suffix and explicitly maps computation across heterogeneous accelerators (GPU and NPU) to exploit cross-device parallelism. The draft speculates forward continuations for the target to verify, while the target simultaneously speculates correction paths for the draft, converting speculation into two complementary execution pipelines. To further cut draft latency without weakening acceptance semantics, we add speculative streaming so the draft emits multiple tokens per step. This dual strategy of parallel heterogeneous execution plus multi-token speculative streaming pushes speculative decoding toward its ideal regime of high acceptance with low overhead. On SpecBench with server-scale models from 14B to 66B parameters, Mirror-SD delivers consistent end-to-end gains, achieving 2.8x-5.8x wall-time speedups across diverse tasks and a 30% average relative improvement over the strongest baseline, EAGLE3.", "authors": ["Nikhil Bhendawade", "Kumari Nishu", "Arnav Kundu", "Chris Bartels", "Minsik Cho", "Irina Belousova"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-15", "url": "https://arxiv.org/abs/2510.13161", "pdf_url": "https://arxiv.org/pdf/2510.13161v2", "arxiv_id": "2510.13161", "doi": "10.48550/arXiv.2510.13161", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3357}
{"id": "3612acd84e7b6b937ac87cfba9ca43fd383814c99bbb4b680cc37da8a185387d", "sources": ["arxiv", "semantic_scholar"], "title": "STAR: Decode-Phase Rescheduling for LLM Inference", "abstract": "Large Language Model (LLM) inference has emerged as a fundamental paradigm, however, variations in output length cause severe workload imbalance in the decode phase, particularly for long-output reasoning tasks. Existing systems, such as PD disaggregation architectures, rely on static prefill-to-decode scheduling, which often results in SLO violations and OOM failures under evolving decode workloads. In this paper, we propose STAR, a decode rescheduling system powered by length prediction to anticipate future workloads. Our core contributions include: (1) A lightweight and continuous LLM-native prediction method that leverages LLM hidden state to model remaining generation length with high precision (reducing MAE by 49.42%) and low overhead (cutting predictor parameters by 93.28%); (2) A rescheduling solution in decode phase with a dynamic balancing mechanism that integrates current and predicted workloads, reducing P99 TPOT by 75.1% and achieving 2.63 times higher goodput.", "authors": ["Zhibin Wang", "Zetao Hong", "Xue Li", "Zibo Wang", "Shipeng Li", "Qingkai Meng", "Qing Wang", "Chengying Huan", "Rong Gu", "Sheng Zhong", "Chen Tian"], "categories": ["cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-15", "url": "https://arxiv.org/abs/2510.13668", "pdf_url": "https://arxiv.org/pdf/2510.13668v2", "arxiv_id": "2510.13668", "doi": "10.1145/3806645.3807813", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.2136}
{"id": "677bbdf95edb9fb3fc816aad5a69d531d06ba2c3e79d82b452f45b4115b93fb9", "sources": ["arxiv", "semantic_scholar"], "title": "3-Model Speculative Decoding", "abstract": "Speculative Decoding (SD) accelerates inference in large language models by using a smaller draft model to propose tokens, which are then verified by a larger target model. However, the throughput gains of SD are fundamentally limited by a trade-off between draft model size and token acceptance: smaller draft models generate tokens more quickly but exhibit greater divergence from the target model, resulting in lower acceptance rates and reduced speedups. We introduce Pyramid Speculative Decoding (PyramidSD), an extension of SD that inserts an intermediate qualifier model between the draft and target to bridge the distributional gap in output predictions, allowing smaller model to be used for drafting. This hierarchical decoding strategy improves alignment across models, enabling higher acceptance rates and allowing the use of significantly smaller draft models without sacrificing overall performance. PyramidSD builds on fuzzy acceptance criteria to support relaxed divergence thresholds at each stage, improving throughput. In experiments, PyramidSD achieves up to 1.91x generation speed over standard SD, reaching 124 tokens per second on a consumer GPU (RTX 4090). In small-memory settings with a 1B-parameter draft model and an 8B target model, PyramidSD minimally trades target model quality for improved throughput. Overall, PyramidSD offers a practical approach to enhancing speculative decoding efficiency and can be readily applied to existing inference pipelines.", "authors": ["Sanghyun Byun", "Mohanad Odema", "Jung Ick Guack", "Baisub Lee", "Jacob Song", "Woo Seong Chung"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-14", "url": "https://arxiv.org/abs/2510.12966", "pdf_url": "https://arxiv.org/pdf/2510.12966v1", "arxiv_id": "2510.12966", "doi": "10.48550/arXiv.2510.12966", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3346}
{"id": "03d69efd712c80783e10e157380ccc82ce2725ec8d2c05a5391551d536a31933", "sources": ["arxiv", "semantic_scholar"], "title": "Efficient LLM Inference over Heterogeneous Edge Networks with Speculative Decoding", "abstract": "Large language model (LLM) inference at the network edge is a promising serving paradigm that leverages distributed edge resources to run inference near users and enhance privacy. Existing edge-based LLM inference systems typically adopt autoregressive decoding (AD), which only generates one token per forward pass. This iterative process, compounded by the limited computational resources of edge nodes, results in high serving latency and constrains the system's ability to support multiple users under growing demands.To address these challenges, we propose a speculative decoding (SD)-based LLM serving framework that deploys small and large models across heterogeneous edge nodes to collaboratively deliver inference services. Specifically, the small model rapidly generates draft tokens that the large model verifies in parallel, enabling multi-token generation per forward pass and thus reducing serving latency. To improve resource utilization of edge nodes, we incorporate pipeline parallelism to overlap drafting and verification across multiple inference tasks. Based on this framework, we analyze and derive a comprehensive latency model incorporating both communication and inference latency. Then, we formulate a joint optimization problem for speculation length, task batching, and wireless communication resource allocation to minimize total serving latency. To address this problem, we derive the closed-form solutions for wireless communication resource allocation, and develop a dynamic programming algorithm for joint batching and speculation control strategies. Experimental results demonstrate that the proposed framework achieves lower serving latency compared to AD-based serving systems. In addition,the proposed joint optimization method delivers up to 44.9% latency reduction compared to benchmark schemes.", "authors": ["Bingjie Zhu", "Zhixiong Chen", "Liqiang Zhao", "Hyundong Shin", "Arumugam Nallanathan"], "categories": ["eess.SY"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2025-10-13", "url": "https://arxiv.org/abs/2510.11331", "pdf_url": "https://arxiv.org/pdf/2510.11331v1", "arxiv_id": "2510.11331", "doi": "10.48550/arXiv.2510.11331", "citation_count": 6, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3334}
{"id": "a0952e40b1e21fa0f8ac6fbd30cf6e1b0f777b2b584f590ee412447baffdf733", "sources": ["arxiv", "semantic_scholar"], "title": "Conformal Sparsification for Bandwidth-Efficient Edge-Cloud Speculative Decoding", "abstract": "Edge-cloud speculative decoding (SD) accelerates inference by having a cloud-based large language model (LLM) that verifies draft tokens generated by a resource-constrained small language model (SLM) at the edge. A central bottleneck is the limited bandwidth of the edge-cloud link, which necessitates efficient compression of draft token distributions. We first derive an information-theoretic bound that decomposes the token rejection rate into contributions from SLM-LLM distribution mismatch and from quantization distortion. Guided by this analysis, we propose the Sparse Quantize-and-Sample SD (SQS-SD) framework, which exploits distributional sparsity through structured sparsification and lattice-based quantization. Within this framework, K-SQS applies fixed top-K truncation, while C-SQS adaptively adjusts the retained token set via online conformal prediction to ensure bounded deviation from the dense distribution. Empirical results confirm that both approaches improve end-to-end latency and rejection rates in complimentary operating regimes.", "authors": ["Payel Bhattacharjee", "Fengwei Tian", "Meiyu Zhong", "Guangyi Zhang", "Osvaldo Simeone", "Ravi Tandon"], "categories": ["cs.LG", "cs.AI", "cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2025-10-11", "url": "https://arxiv.org/abs/2510.09942", "pdf_url": "https://arxiv.org/pdf/2510.09942v1", "arxiv_id": "2510.09942", "doi": "10.48550/arXiv.2510.09942", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3311}
{"id": "2f9527dcc8cb49e9c70e043c19e102b010890bb1d59e93d8122eaa4d8c8eca76", "sources": ["arxiv", "semantic_scholar"], "title": "SP-MoE: Speculative Decoding and Prefetching for Accelerating MoE-based Model Inference", "abstract": "The Mixture-of-Experts (MoE) architecture has been widely adopted in large language models (LLMs) to reduce computation cost through model sparsity. Employing speculative decoding (SD) can further accelerate MoE inference by drafting multiple tokens per step and verifying them in parallel. However, combining MoE with SD inflates GPU memory and aggravates CPU-GPU bandwidth contention during multi-token verification. Existing MoE offloading systems are SD-agnostic and do not address this bottleneck. We present SP-MoE, the first SD-aware expert-offloading and compute-communication pipelining framework. SP-MoE introduces: (1) speculative expert prefetching that exploits structural correspondence between the draft and target models to prefetch likely experts ahead of verification; (2) a cutoff-layer policy that bounds per-layer prefetch depth based on empirical profiles and an analytical latency model, guaranteeing just-in-time availability without overfetch; and (3) a pipelined runtime with asynchronous prefetch threads and batched I/O to hide loading latency. Extensive experiments demonstrate that SP-MoE achieves a 1.07-3.5 times TPOT speedup over state-of-the-art methods across diverse datasets, environments, and MoE-based models.", "authors": ["Liangkun Chen", "Zijian Wen", "Tian Wu", "Xiaoxi Zhang", "Chuan Wu"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-11", "url": "https://arxiv.org/abs/2510.10302", "pdf_url": "https://arxiv.org/pdf/2510.10302v2", "arxiv_id": "2510.10302", "doi": "10.48550/arXiv.2510.10302", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3311}
{"id": "cce4ea6eecbe8e5b871c8896f310d805fba3867618365d3436a8e3f8d0df76ca", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Jacobi-Denoising Decoding for Accelerating Autoregressive Text-to-image Generation", "abstract": "As a new paradigm of visual content generation, autoregressive text-to-image models suffer from slow inference due to their sequential token-by-token decoding process, often requiring thousands of model forward passes to generate a single image. To address this inefficiency, we propose Speculative Jacobi-Denoising Decoding (SJD2), a framework that incorporates the denoising process into Jacobi iterations to enable parallel token generation in autoregressive models. Our method introduces a next-clean-token prediction paradigm that enables the pre-trained autoregressive models to accept noise-perturbed token embeddings and predict the next clean tokens through low-cost fine-tuning. This denoising paradigm guides the model towards more stable Jacobi trajectories. During inference, our method initializes token sequences with Gaussian noise and performs iterative next-clean-token-prediction in the embedding space. We employ a probabilistic criterion to verify and accept multiple tokens in parallel, and refine the unaccepted tokens for the next iteration with the denoising trajectory. Experiments show that our method can accelerate generation by reducing model forward passes while maintaining the visual quality of generated images.", "authors": ["Yao Teng", "Fuyun Wang", "Xian Liu", "Zhekai Chen", "Han Shi", "Yu Wang", "Zhenguo Li", "Weiyang Liu", "Difan Zou", "Xihui Liu"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-10", "url": "https://arxiv.org/abs/2510.08994", "pdf_url": "https://arxiv.org/pdf/2510.08994v1", "arxiv_id": "2510.08994", "doi": "10.48550/arXiv.2510.08994", "citation_count": 4, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.33}
{"id": "8d87ee13b01f1e8580143616019a08c09c2bfc308ef3bce5f74003b398aa8186", "sources": ["arxiv", "semantic_scholar"], "title": "SPAD: Specialized Prefill and Decode Hardware for Disaggregated LLM Inference", "abstract": "Large Language Models (LLMs) have gained popularity in recent years, driving up the demand for inference. LLM inference is composed of two phases with distinct characteristics: a compute-bound prefill phase followed by a memory-bound decode phase. To efficiently serve LLMs, prior work proposes prefill-decode disaggregation to run each phase on separate hardware. However, existing hardware poorly matches the different requirements of each phase. Current datacenter GPUs and TPUs follow a more-is-better design philosophy that maximizes compute and memory resources, causing memory bandwidth underutilization in the prefill phase and compute underutilization in the decode phase. Such underutilization directly translates into increased serving costs. This paper proposes SPAD (Specialized Prefill and Decode hardware), adopting a less-is-more methodology to design specialized chips tailored to the distinct characteristics of prefill and decode phases. The proposed Prefill Chips have larger systolic arrays and use cost-effective GDDR memory, whereas the proposed Decode Chips retain high memory bandwidth but reduce compute capacity. Compared to modeled H100s, simulations show that the proposed Prefill Chips deliver 8% higher prefill performance on average at 52% lower hardware cost, while the proposed Decode Chips achieve 97% of the decode performance with 28% lower TDP. End-to-end simulations on production traces show that SPAD reduces hardware cost by 19%-41% and TDP by 2%-17% compared to modeled baseline clusters while offering the same performance. Even when models and workloads change, SPAD can reallocate either type of chip to run either phase and still achieve 11%-43% lower hardware costs, demonstrating the longevity of the SPAD design.", "authors": ["Hengrui Zhang", "Pratyush Patel", "August Ning", "David Wentzlaff"], "categories": ["cs.AR", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-09", "url": "https://arxiv.org/abs/2510.08544", "pdf_url": "https://arxiv.org/pdf/2510.08544v1", "arxiv_id": "2510.08544", "doi": "10.48550/arXiv.2510.08544", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3289}
{"id": "7b5859c00f7474158aece9cbeaa91dd3e811b1ed862e9f5029de4480cd32954e", "sources": ["arxiv", "semantic_scholar"], "title": "OWL: Overcoming Window Length-Dependence in Speculative Decoding for Long-Context Inputs", "abstract": "Speculative decoding promises faster inference for large language models (LLMs), yet existing methods fail to generalize to real-world settings. Benchmarks typically assume short contexts (e.g., 2K tokens), whereas practical workloads involve long contexts. We find current approaches degrade severely with long contexts; for instance, EAGLE3 even slows down the generation speed by 0.81x. We address these limitations by releasing a new long-context benchmark (LongSpecBench) and introducing a novel model (OWL). OWL achieves about 5x higher acceptance length than EAGLE3 on long-context inputs through three innovations: (1) an LSTM-based drafter conditioned only on the last-token state, making it generalize to various lengths, (2) a special token [SPEC] in the verifier that produces richer representation for drafter, and (3) a hybrid algorithm combining both tree and non-tree decoding methods. We release all code and datasets to advance future research.", "authors": ["Jaeseong Lee", "seung-won hwang", "Aurick Qiao", "Gabriele Oliaro", "Ye Wang", "Samyam Rajbhandari"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-08", "url": "https://arxiv.org/abs/2510.07535", "pdf_url": "https://arxiv.org/pdf/2510.07535v1", "arxiv_id": "2510.07535", "doi": "10.48550/arXiv.2510.07535", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3277}
{"id": "346cd617ccf424b28e1e4eab0cb897f62ecd7b2ffc7deaa9c316d6e58cd43eab", "sources": ["arxiv", "semantic_scholar"], "title": "Draft, Verify, and Improve: Toward Training-Aware Speculative Decoding", "abstract": "Autoregressive (AR) decoding is a major latency bottleneck for large language models. Speculative decoding (SD) accelerates AR by letting a drafter propose multi-token blocks that a verifier accepts or rejects. However, many SD systems require heavy offline training or extra components. These choices raise data/compute cost and can yield brittle drafters under distribution drift. We introduce \\emph{Draft, Verify, \\& Improve (DVI)}, a training-aware self-speculative framework that combines inference with continual online learning. We partition an LLM into a drafter and a verifier, and during generation, verifier accept/reject decisions are converted into supervision signals and used to update the drafter head. A simple \\emph{KL$\\rightarrow$RL} schedule bootstraps calibration via online distillation and then adds reward-masked cross-entropy with a on-policy policy-gradient term, preserving lossless, single model deployment. On Spec-Bench, DVI achieves a $2.16\\times$ wall-time speedup, on par with SoTA approaches like EAGLE-2, while orders of magnitude less data for training, and ablations show that DVI outperforms KL-only online distillation. DVI demonstrates that \\emph{training-aware} self-speculation can deliver state-of-the-art, lossless speedups with minimal training overhead.", "authors": ["Shrenik Bhansali", "Larry Heck"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-06", "url": "https://arxiv.org/abs/2510.05421", "pdf_url": "https://arxiv.org/pdf/2510.05421v1", "arxiv_id": "2510.05421", "doi": "10.48550/arXiv.2510.05421", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3254}
{"id": "c984a01a98ad2856f3e307a10fb15d4261bcc017d099d8052df0df7a8fbecb20", "sources": ["arxiv", "semantic_scholar"], "title": "Self Speculative Decoding for Diffusion Large Language Models", "abstract": "Diffusion-based Large Language Models (dLLMs) have emerged as a competitive alternative to autoregressive models, offering unique advantages through bidirectional attention and parallel generation paradigms. However, the generation results of current parallel decoding methods deviate from stepwise decoding, introducing potential performance degradation, which limits their practical deployment. To address this problem, we propose \\textbf{S}elf \\textbf{S}peculative \\textbf{D}ecoding (SSD), a lossless inference acceleration method that leverages the dLLM itself as both speculative decoding drafter and verifier without auxiliary modules. SSD introduces a self-drafting mechanism where the model generates predictions for multiple positions, then verifies them through hierarchical verification trees in a single forward pass. Unlike traditional speculative decoding that requires separate draft models, SSD eliminates model redundancy and memory overhead by exploiting the dLLM's inherent parallel prediction capability for multiple positions. This self-speculative approach allows the model to progressively verify and accept multiple tokens in a single forward pass. Our experiments demonstrate that SSD achieves up to 3.46$\\times$ speedup while keeping the output identical to stepwise decoding on open source models such as LLaDA and Dream. Code will be made publicly available on GitHub.", "authors": ["Yifeng Gao", "Ziang Ji", "Yuxuan Wang", "Biqing Qi", "Hanlin Xu", "Linfeng Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-05", "url": "https://arxiv.org/abs/2510.04147", "pdf_url": "https://arxiv.org/pdf/2510.04147v1", "arxiv_id": "2510.04147", "doi": "10.48550/arXiv.2510.04147", "citation_count": 25, "influential_citation_count": 5, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5011}
{"id": "de9adfe6916a2b78d1d444f1e3a0bf5f0401e3ea6ebe9b23dbfb7ff81dd9e499", "sources": ["arxiv", "semantic_scholar"], "title": "The Disparate Impacts of Speculative Decoding", "abstract": "The practice of speculative decoding, whereby inference is probabilistically supported by a smaller, cheaper, ``drafter'' model, has become a standard technique for systematically reducing the decoding time of large language models. This paper conducts an analysis of speculative decoding through the lens of its potential disparate speed-up rates across tasks. Crucially, the paper shows that speed-up gained from speculative decoding is not uniformly distributed across tasks, consistently diminishing for under-fit, and often underrepresented tasks. To better understand this phenomenon, we derive an analysis to quantify this observed ``unfairness'' and draw attention to the factors that motivate such disparate speed-ups to emerge. Further, guided by these insights, the paper proposes a mitigation strategy designed to reduce speed-up disparities and validates the approach across several model pairs, revealing on average a 12% improvement in our fairness metric.", "authors": ["Jameson Sandler", "Ahmet Üstün", "Marco Romanelli", "Sara Hooker", "Ferdinando Fioretto"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-02", "url": "https://arxiv.org/abs/2510.02128", "pdf_url": "https://arxiv.org/pdf/2510.02128v1", "arxiv_id": "2510.02128", "doi": "10.48550/arXiv.2510.02128", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3208}
{"id": "6efead990ebdce075638f64911d7b96e7a2d66789cb4e07586b47875b8310f56", "sources": ["arxiv", "semantic_scholar"], "title": "HiSpec: Hierarchical Speculative Decoding for LLMs", "abstract": "Speculative decoding accelerates LLM inference by using a smaller draft model to speculate tokens that a larger target model verifies. Verification is often the bottleneck (e.g. verification is $4\\times$ slower than token generation when a 3B model speculates for a 70B target model), but most prior works focus only on accelerating drafting. $\\textit{``Intermediate\"}$ verification reduces verification time by discarding inaccurate draft tokens early, but existing methods incur substantial training overheads in incorporating the intermediate verifier, increase the memory footprint to orchestrate the intermediate verification step, and compromise accuracy by relying on approximate heuristics. We propose $\\underline{\\textit{Hi}}\\textit{erarchical }\\underline{\\textit{Spec}}\\textit{ulative Decoding (HiSpec)}$, a framework for high-throughput speculative decoding that exploits $\\textit{early-exit (EE) models}$ for low-overhead intermediate verification. EE models allow tokens to exit early by skipping layer traversal and are explicitly trained so that hidden states at selected layers can be interpreted, making them uniquely suited for intermediate verification without drastically increasing compute and memory overheads. To improve resource-efficiency even further, we design a methodology that enables HiSpec to re-use key-value caches and hidden states between the draft, intermediate verifier, and target models. To maintain accuracy, HiSpec periodically validates the draft tokens accepted by the intermediate verifier against the target model. Our evaluations using various representative benchmarks and models show that HiSpec improves throughput by 1.28$\\times$ on average and by up to 2.01$\\times$ compared to the baseline single-layer speculation without compromising accuracy.", "authors": ["Avinash Kumar", "Sujay Sanghavi", "Poulami Das"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-10-01", "url": "https://arxiv.org/abs/2510.01336", "pdf_url": "https://arxiv.org/pdf/2510.01336v2", "arxiv_id": "2510.01336", "doi": "10.48550/arXiv.2510.01336", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3197}
{"id": "6324420d09a0a05ef0eb7a32dc54aa143c05a99733a2f0a8fa9211292c0ab04e", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Verification: Exploiting Information Gain to Refine Speculative Decoding", "abstract": "LLMs have low GPU efficiency and high latency due to autoregressive decoding. Speculative decoding (SD) mitigates this using a small draft model to speculatively generate multiple tokens, which are then verified in parallel by a target model. However, when speculation accuracy is low, the overhead from rejected tokens can offset the benefits, limiting SD's effectiveness, especially at large batch sizes. To address this, we propose Speculative Verification (SV), an efficient augmentation to SD that dynamically predicts speculation accuracy and adapts the verification length to maximize throughput. SV introduces a companion model - a small auxiliary model similar in size to the draft model - to estimate the alignment between draft and target model distributions. By maximizing the information gain from quantifying this alignment, SV refines verification decisions, reducing wasted computation on rejected tokens and improving decoding efficiency. Moreover, SV requires no modifications to the draft or target models and is compatible with existing SD variants. We extensively evaluated SV on publicly available LLMs across three NLP tasks using nine combinations of draft, companion, and target models, including 13B-72B target models and three types of variations: base (no finetuning), instruction-tuned, and task fine-tuned. Across all experiments and batch sizes (4-80), SV consistently outperforms both SD and standard decoding with the target model. It improves SD performance by up to 2$\\times$, with an average speedup of 1.4 $\\times$ in large-batch settings (batch sizes 32-80). These results demonstrate SV's robustness, scalability, and practical utility for efficient LLM inference.", "authors": ["Sungkyun Kim", "Jaemin Kim", "Dogyung Yoon", "Jiho Shin", "Junyeol Lee", "Jiwon Seo"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-29", "url": "https://arxiv.org/abs/2509.24328", "pdf_url": "https://arxiv.org/pdf/2509.24328v2", "arxiv_id": "2509.24328", "doi": "10.48550/arXiv.2509.24328", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3174}
{"id": "88eebe9d27ee6615bf039d2474fe8fe49463ebac3a4fc2cc3374f8217b3a18f4", "sources": ["arxiv", "semantic_scholar"], "title": "HiViS: Hiding Visual Tokens from the Drafter for Speculative Decoding in Vision-Language Models", "abstract": "Speculative decoding has proven effective for accelerating inference in Large Language Models (LLMs), yet its extension to Vision-Language Models (VLMs) remains limited by the computational burden and semantic inconsistency introduced by visual tokens. Recent studies reveal that visual tokens in large VLMs are highly redundant, and most of them can be removed without compromising generation quality. Motivated by this observation, we propose HiViS (Hiding Visual Tokens from the Drafter for Speculative Decoding in Vision-Language Models), a framework that utilizes the target VLM as a semantic fusion model, allowing the drafter to obtain visual information without explicitly processing visual tokens, ensuring that the drafter's prefill sequence length matches that of the textual tokens. Furthermore, HiViS employs a time-step-aware aligned training scheme that allows the drafter to autonomously propagate and refine instructive visual-textual semantics during independent drafting, guided by step-dependent bias-correction residuals. Extensive experiments across representative VLMs and benchmarks demonstrate that HiViS achieves significant improvements in average acceptance length and speedup ratio.", "authors": ["Zhinan Xie", "Peisong Wang", "Shuang Qiu", "Jian Cheng"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-28", "url": "https://arxiv.org/abs/2509.23928", "pdf_url": "https://arxiv.org/pdf/2509.23928v2", "arxiv_id": "2509.23928", "doi": "10.48550/arXiv.2509.23928", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3162}
{"id": "bb334f7fd49d3b2f29971cd30b20b09e1cf88cc54f08bdf3490153c93a60e261", "sources": ["arxiv", "semantic_scholar"], "title": "DiffuSpec: Unlocking Diffusion Language Models for Speculative Decoding", "abstract": "As large language models (LLMs) scale up, accuracy improves, but the autoregressive (AR) nature of decoding increases latency since each token requires a serial forward pass. Speculative decoding addresses this by employing a fast drafter to propose multi-token drafts, which are then verified in parallel by the target model. However, many deployments still rely on AR drafters, where sequential passes limit wall-clock gains. We revisit the drafting stage and present DiffuSpec, a training-free drop-in framework that uses a pretrained diffusion language model (DLM) to produce multi-token drafts in a single forward pass, while remaining compatible with standard AR verifiers. Because DLM drafts are generated under bidirectional conditioning, parallel per-position candidates form a token lattice in which the locally highest-probability token at each position need not form a causal left-to-right path. Moreover, DLM drafting requires pre-specifying a draft length, inducing a speed-quality trade-off. To address these challenges, we introduce two practical components: (i) a causal-consistency path search (CPS) over this lattice that extracts a left-to-right path aligned with AR verification; and (ii) an adaptive draft-length (ADL) controller that adjusts next proposal size based on recent acceptance feedback and realized generated length. Across benchmarks, DiffuSpec yields up to 3x wall-clock speedup, establishing diffusion-based drafting as a robust alternative to autoregressive drafters for speculative decoding.", "authors": ["Guanghao Li", "Zhihui Fu", "Min Fang", "Qibin Zhao", "Ming Tang", "Chun Yuan", "Jun Wang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-28", "url": "https://arxiv.org/abs/2510.02358", "pdf_url": "https://arxiv.org/pdf/2510.02358v1", "arxiv_id": "2510.02358", "doi": "10.48550/arXiv.2510.02358", "citation_count": 21, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3356}
{"id": "3f8daa231589327dd1ca1e40074b63c58eb37c1d9102560caa85c1a6b4310e1c", "sources": ["arxiv", "semantic_scholar"], "title": "SelfJudge: Faster Speculative Decoding via Self-Supervised Judge Verification", "abstract": "Speculative decoding accelerates LLM inference by verifying candidate tokens from a draft model against a larger target model. Recent judge decoding boosts this process by relaxing verification criteria by accepting draft tokens that may exhibit minor discrepancies from target model output, but existing methods are restricted by their reliance on human annotations or tasks with verifiable ground truths, limiting generalizability across diverse NLP tasks. We propose SelfJudge, which trains judge verifiers via self-supervision of the target model. Our method measures semantic preservation by assessing whether token-substituted responses preserve the meaning of original responses, enabling automatic verifier training across diverse NLP tasks. Our experiments show SelfJudge achieves superior inference-accuracy trade-offs than judge decoding baselines, offering a broadly applicable solution for faster LLM inference.", "authors": ["Kanghoon Yoon", "Minsub Kim", "Sungjae Lee", "Joonhyung Lee", "Sunghyeon Woo", "Yeonjun In", "Se Jung Kwon", "Chanyoung Park", "Dongsoo Lee"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-26", "url": "https://arxiv.org/abs/2510.02329", "pdf_url": "https://arxiv.org/pdf/2510.02329v2", "arxiv_id": "2510.02329", "doi": "10.48550/arXiv.2510.02329", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.314}
{"id": "179a0e42cc9b324bef67d7efee9db40d2a6f7f241d7d97a3730063a1142cb1de", "sources": ["arxiv", "semantic_scholar"], "title": "Self-Speculative Biased Decoding for Faster Re-Translation", "abstract": "Large language models achieve strong machine translation quality but incur high inference cost and latency, posing challenges for simultaneous translation. Re-translation provides a practical solution for off-the-shelf LLMs by repeatedly regenerating the target output as the source input grows, but it suffers from substantial redundant computation. We propose Self-Speculative Biased Decoding (SSBD), a simple and tuning-free inference method that accelerates re-translation by exploiting temporal coherence in streaming translation. SSBD reuses the model's previous output as a speculative draft for the updated input, verifies the draft efficiently in a single forward pass with a lightweight bias, and resumes autoregressive decoding only from the first divergence. We further introduce a display-only masking strategy that hides unstable suffixes from the user interface while retaining them in the draft for verification and potential acceptance. Experiments show that SSBD achieves substantial speedup over standard re-translation while maintaining comparable translation quality, without architectural changes, auxiliary models, or extra fine-tuning.", "authors": ["Linxiao Zeng", "Haoyun Deng", "Kangyuan Shu", "Shizhen Wang"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-26", "url": "https://arxiv.org/abs/2509.21740", "pdf_url": "https://arxiv.org/pdf/2509.21740v2", "arxiv_id": "2509.21740", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.1998}
{"id": "9afae801d6aac95feff6025a3697734664b926c472fb09096026d6a08ddc1eeb", "sources": ["arxiv", "semantic_scholar"], "title": "Bridging Draft Policy Misalignment: Group Tree Optimization for Speculative Decoding", "abstract": "Speculative decoding accelerates large language model (LLM) inference by letting a lightweight draft model propose multiple tokens that the target model verifies in parallel. Yet existing training objectives optimize only a single greedy draft path, while decoding follows a tree policy that re-ranks and verifies multiple branches. This draft policy misalignment limits achievable speedups. We introduce Group Tree Optimization (GTO), which aligns training with the decoding-time tree policy through two components: (i) Draft Tree Reward, a sampling-free objective equal to the expected acceptance length of the draft tree under the target model, directly measuring decoding performance; (ii) Group-based Draft Policy Training, a stable optimization scheme that contrasts trees from the current and a frozen reference draft model, forming debiased group-standardized advantages and applying a PPO-style surrogate along the longest accepted sequence for robust updates. We further prove that increasing our Draft Tree Reward provably improves acceptance length and speedup. Across dialogue (MT-Bench), code (HumanEval), and math (GSM8K), and multiple LLMs (e.g., LLaMA-3.1-8B, LLaMA-3.3-70B, Vicuna-1.3-13B, DeepSeek-R1-Distill-LLaMA-8B, Qwen3-8B), GTO increases acceptance length by (7.4%) and yields an additional (7.7%) speedup over prior state-of-the-art EAGLE-3. By bridging draft policy misalignment, GTO offers a practical, general solution for efficient LLM inference. Code and draft models are available at https://github.com/hsj576/GTO.", "authors": ["Shijing Hu", "Jingyang Li", "Zhihui Lu", "Pan Zhou"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-26", "url": "https://arxiv.org/abs/2509.22134", "pdf_url": "https://arxiv.org/pdf/2509.22134v2", "arxiv_id": "2509.22134", "doi": "10.48550/arXiv.2509.22134", "citation_count": 5, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/hsj576/GTO", "venue": "arXiv.org", "quality_score": 0.4852}
{"id": "67ab3e7150b2cc3e9d9a8baddf77e7a1fcb7305099a5866674099a9a01d94896", "sources": ["arxiv", "semantic_scholar"], "title": "FastGRPO: Accelerating Policy Optimization via Concurrency-aware Speculative Decoding and Online Draft Learning", "abstract": "Group relative policy optimization (GRPO) has demonstrated significant potential in improving the reasoning capabilities of large language models (LLMs) via reinforcement learning. However, its practical deployment is impeded by an excessively slow training process, primarily attributed to the computationally intensive autoregressive generation of multiple responses per query, which makes the generation phase the primary performance bottleneck. Although speculative decoding presents a promising direction for acceleration, its direct application in GRPO achieves limited speedup under high-concurrency training conditions. To overcome this limitation, we propose a concurrency-aware speculative decoding framework that dynamically adjusts the drafting and verification strategy according to real-time concurrency levels, thereby maximizing the acceleration of the generation process. Furthermore, to address performance degradation arising from distributional drift between the evolving target model and the fixed draft model during training, we introduce an online draft learning mechanism that enables the draft model to continuously adapt using feedback signals from the target model. Experimental results across multiple mathematical reasoning datasets and models demonstrate that the proposed method achieves end-to-end speedups of 2.35x to 2.72x, significantly surpassing baseline approaches in efficiency. The code is available at https://github.com/yedaotian9/GRPO_speculative.", "authors": ["Yizhou Zhang", "Ning Lv", "Teng Wang", "Jisheng Dang"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-26", "url": "https://arxiv.org/abs/2509.21792", "pdf_url": "https://arxiv.org/pdf/2509.21792v1", "arxiv_id": "2509.21792", "doi": "10.48550/arXiv.2509.21792", "citation_count": 4, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/yedaotian9/GRPO_speculative", "venue": "arXiv.org", "quality_score": 0.4852}
{"id": "16a83b33cd2e7a6fb6a32499669a7656b00c00d685df9c4410e31d3b14ae4f32", "sources": ["arxiv", "semantic_scholar"], "title": "SpecMER: Fast Protein Generation with K-mer Guided Speculative Decoding", "abstract": "Autoregressive models have transformed protein engineering by enabling the generation of novel protein sequences beyond those found in nature. However, their sequential inference introduces significant latency, limiting their utility in high-throughput protein screening. Speculative decoding accelerates generation by employing a lightweight draft model to sample tokens, which a larger target model then verifies and refines. Yet, in protein sequence generation, draft models are typically agnostic to the structural and functional constraints of the target protein, leading to biologically implausible outputs and a shift in the likelihood distribution of generated sequences. We introduce SpecMER (Speculative Decoding via k-mer Guidance), a novel framework that incorporates biological, structural, and functional priors using k-mer motifs extracted from multiple sequence alignments. By scoring candidate sequences in parallel and selecting those most consistent with known biological patterns, SpecMER significantly improves sequence plausibility while retaining the efficiency of speculative decoding. SpecMER achieves 24-32% speedup over standard autoregressive decoding, along with higher acceptance rates and improved sequence likelihoods.", "authors": ["Thomas Walton", "Darin Tsui", "Aryan Musharaf", "Amirali Aghazadeh"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-25", "url": "https://arxiv.org/abs/2509.21689", "pdf_url": "https://arxiv.org/pdf/2509.21689v1", "arxiv_id": "2509.21689", "doi": "10.48550/arXiv.2509.21689", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3128}
{"id": "215ab202dca4ce7f7f2df32aebc83b69958af1530c743a54610fd951c29ae036", "sources": ["arxiv", "semantic_scholar"], "title": "FastEagle: Cascaded Drafting for Accelerating Speculative Decoding", "abstract": "Speculative decoding accelerates generation by drafting candidates and verifying them in parallel, yet state-of-the-art drafters (e.g., EAGLE) still require N sequential passes to propose N tokens. We present FastEagle, a non-autoregressive cascaded drafter that emits an entire draft in a single forward pass. FastEagle replaces temporal steps with a lightweight layer cascade and trains with layer-wise supervision to mitigate error accumulation. Coupled with a constrained draft tree that preserves lossless verification cost, FastEagle delivers substantial wall-clock speedups over strong autoregressive drafters while maintaining competitive acceptance behavior. Across multiple LLMs (Vicuna-13B, LLaMA-Instruct 3.x, and DeepSeek-R1-Distill-LLaMA) and tasks (MT-Bench, HumanEval, GSM8K, CNN/DM, Alpaca), FastEagle consistently outperforms EAGLE-3 in speedup under both greedy and stochastic decoding, with comparable average acceptance lengths. These results indicate that removing sequential dependencies in drafting is a practical path toward lossless LLM inference acceleration.", "authors": ["Haiduo Huang", "Jiangcheng Song", "Wenzhe Zhao", "Pengju Ren"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-24", "url": "https://arxiv.org/abs/2509.20416", "pdf_url": "https://arxiv.org/pdf/2509.20416v1", "arxiv_id": "2509.20416", "doi": "10.48550/arXiv.2509.20416", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE International Conference on Acoustics, Speech, and Signal Processing", "quality_score": 0.3117}
{"id": "ad9c0d380cd671f34f669c66ce4e8491b51ea016c658e69440c0729be00810de", "sources": ["arxiv", "semantic_scholar"], "title": "SpecMamba: Accelerating Mamba Inference on FPGA with Speculative Decoding", "abstract": "The growing demand for efficient long-sequence modeling on edge devices has propelled widespread adoption of State Space Models (SSMs) like Mamba, due to their superior computational efficiency and scalability. As its autoregressive generation process remains memory-bound, speculative decoding has been proposed that incorporates draft model generation and target model verification. However, directly applying speculative decoding to SSMs faces three key challenges: (1) hidden state backtracking difficulties, (2) tree-based parallel verification incompatibility, and (3) hardware workload mismatch. To address these challenges, we propose SpecMamba, the first FPGA-based accelerator for Mamba with speculative decoding, which features system, algorithm, and hardware co-design. At the system level, we present a memory-aware hybrid backtracking strategy to coordinate both models. At the algorithm level, we propose first-in-first-out (FIFO)-based tree verification with tiling to minimize memory access. At the hardware level, we customize a dataflow that computes linear layers in parallel and SSM layers in series to enable maximal overlapping. Implemented on AMD FPGA platforms (VHK158 and VCK190), SpecMamba achieves a 2.27x speedup over GPU baselines and a 2.85x improvement compared to prior FPGA solutions, while demonstrating 5.41x and 1.26x higher energy efficiency, respectively.", "authors": ["Linfeng Zhong", "Songqiang Xu", "Huifeng Wen", "Tong Xie", "Qingyu Guo", "Yuan Wang", "Meng Li"], "categories": ["cs.AR"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-24", "url": "https://arxiv.org/abs/2509.19873", "pdf_url": "https://arxiv.org/pdf/2509.19873v1", "arxiv_id": "2509.19873", "doi": "10.1109/ICCAD66269.2025.11240945", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.1983}
{"id": "793293c4df76a6ccaa97b98fe6b9e18f2de67267134acfe7dccfe7d73c7bada5", "sources": ["arxiv", "semantic_scholar"], "title": "Steering Multimodal Large Language Models Decoding for Context-Aware Safety", "abstract": "Multimodal Large Language Models (MLLMs) are increasingly deployed in real-world applications, yet their ability to make context-aware safety decisions remains limited. Existing methods often fail to balance oversensitivity (unjustified refusals of benign queries) and undersensitivity (missed detection of visually grounded risks), leaving a persistent gap in safety alignment. To address this issue, we introduce Safety-aware Contrastive Decoding (SafeCoDe), a lightweight and model-agnostic decoding framework that dynamically adjusts token generation based on multimodal context. SafeCoDe operates in two stages: (1) a contrastive decoding mechanism that highlights tokens sensitive to visual context by contrasting real and Gaussian-noised images, and (2) a global-aware token modulation strategy that integrates scene-level reasoning with token-level adjustment to adapt refusals according to the predicted safety verdict. Extensive experiments across diverse MLLM architectures and safety benchmarks, covering undersensitivity, oversensitivity, and general safety evaluations, show that SafeCoDe consistently improves context-sensitive refusal behaviors while preserving model helpfulness.", "authors": ["Zheyuan Liu", "Zhangchen Xu", "Guangyao Dou", "Xiangchi Yuan", "Zhaoxuan Tan", "Radha Poovendran", "Meng Jiang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-23", "url": "https://arxiv.org/abs/2509.19212", "pdf_url": "https://arxiv.org/pdf/2509.19212v1", "arxiv_id": "2509.19212", "doi": "10.48550/arXiv.2509.19212", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3105}
{"id": "15fe8618c8fc53ca1fe892f6ff252d7ec1b9775da3493ce59408dc27b77ab19b", "sources": ["arxiv", "semantic_scholar"], "title": "Structuring The Future: Diffusion LLM Speculative Decoding via Calibrated Draft Graphs", "abstract": "Diffusion LLMs (dLLMs) have recently emerged as a powerful alternative to autoregressive LLMs (AR-LLMs) with the potential to operate at significantly higher token-generation rates. To unlock this potential, we present Spiffy, a speculative decoding algorithm to accelerate dLLM inference while provably preserving the model's output distribution. This work addresses the unique challenges involved in applying ideas from speculative decoding of AR-LLMs to dLLMs. Spiffy performs auto-speculation to eliminate the overheads of an independent draft model, structuring draft states in the form of a novel directed draft graph to take advantage of the bidirectional, blockwise nature of dLLM generation. These draft graphs are calibrated offline to maximize acceptance rates and are dynamically pruned during inference for improved computational efficiency. We present a detailed formulation of Spiffy and demonstrate its ability to accelerate LLaDA, Dream, and SDAR models in combination with KV caching and threshold-based dynamic unmasking leading to up to $8.6\\times$ reduction in model inferences and $6.3\\times$ acceleration in token rate.", "authors": ["Sudhanshu Agrawal", "Risheek Garrepalli", "Raghavv Goel", "Christopher Lott", "Fatih Porikli", "Mingu Lee"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-22", "url": "https://arxiv.org/abs/2509.18085", "pdf_url": "https://arxiv.org/pdf/2509.18085v4", "arxiv_id": "2509.18085", "doi": null, "citation_count": 17, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3138}
{"id": "c2fba4d457deaf2bfba4a6d9493c8f177835aa264aa1d0ab71932ca7ab5ea63f", "sources": ["arxiv", "semantic_scholar"], "title": "Speculate Deep and Accurate: Lossless and Training-Free Acceleration for Offloaded LLMs via Substitute Speculative Decoding", "abstract": "The immense model sizes of large language models (LLMs) challenge deployment on memory-limited consumer GPUs. Although model compression and parameter offloading are common strategies to address memory limitations, compression can degrade quality, and offloading maintains quality but suffers from slow inference. Speculative decoding presents a promising avenue to accelerate parameter offloading, utilizing a fast draft model to propose multiple draft tokens, which are then verified by the target LLM in parallel with a single forward pass. This method reduces the time-consuming data transfers in forward passes that involve offloaded weight transfers. Existing methods often rely on pretrained weights of the same family, but require additional training to align with custom-trained models. Moreover, approaches that involve draft model training usually yield only modest speedups. This limitation arises from insufficient alignment with the target model, preventing higher token acceptance lengths. To address these challenges and achieve greater speedups, we propose SubSpec, a plug-and-play method to accelerate parameter offloading that is lossless and training-free. SubSpec constructs a highly aligned draft model by generating low-bit quantized substitute layers from offloaded target LLM portions. Additionally, our method shares the remaining GPU-resident layers and the KV-Cache, further reducing memory overhead and enhance alignment. SubSpec achieves a high average acceptance length, delivering 9.1x speedup for Qwen2.5 7B on MT-Bench (8GB VRAM limit) and an average of 12.5x speedup for Qwen2.5 32B on popular generation benchmarks (24GB VRAM limit).", "authors": ["Pei-Shuo Wang", "Jian-Jia Chen", "Chun-Che Yang", "Chi-Chih Chang", "Ning-Chi Huang", "Mohamed S. Abdelfattah", "Kai-Chiang Wu"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-22", "url": "https://arxiv.org/abs/2509.18344", "pdf_url": "https://arxiv.org/pdf/2509.18344v2", "arxiv_id": "2509.18344", "doi": "10.48550/arXiv.2509.18344", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3094}
{"id": "365425d46ade5ed71301cc1ba85a2c3fee56538c7158dbb811c009a6308f407c", "sources": ["arxiv", "semantic_scholar"], "title": "Pipeline Parallelism is All You Need for Optimized Early-Exit Based Self-Speculative Decoding", "abstract": "Large language models (LLMs) deliver impressive generation quality, but incur very high inference cost because each output token is generated auto-regressively through all model layers. Early-exit based self-speculative decoding (EESD) has emerged to mitigate this cost. However, in practice, many approaches struggle to achieve the expected acceleration in such draft-then-verify paradigm even with a well-aligned early-exit head and selected exit position. Our analysis reveals that EESD only pays off when the vast majority of draft tokens are accepted by the LLM. Otherwise, the draft cost may overcome the acceleration gain and lead to a negative speedup. To mitigate this, we propose Pipeline-Parallel Self-Speculative Decoding (PPSD) that fully pipelines the draft and verification work so that no effort is wasted on failed predictions. It has two key innovations. We configure the model layers as a pipeline in which early-exit (draft) computations and remaining-layer (verification) computations overlap. We interleave drafting and verification per token. While the LLM is verifying the current token in its final layers, the early-exit path simultaneously drafts the next token. Such a verify-while-draft scheme keeps all units busy and validates tokens on-the-fly analogous to pipelining the speculation and verification stages. Empirical results confirm that PPSD achieves state-of-the-art acceleration in self-speculative LLM inference. On diverse benchmarks, PPSD achieves speedup ratios in the range of 2.01x~3.81x, which gains almost the optimal acceleration at the fixed acceptance rate and exit position, showcasing its advancement in providing efficient self-speculation.", "authors": ["Ruanjun Li", "Ziheng Liu", "Yuanming Shi", "Jiawei Shao", "Chi Zhang", "Xuelong Li"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-19", "url": "https://arxiv.org/abs/2509.19368", "pdf_url": "https://arxiv.org/pdf/2509.19368v1", "arxiv_id": "2509.19368", "doi": "10.48550/arXiv.2509.19368", "citation_count": 1, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3059}
{"id": "2f5c5a74ee7ee207824de4bd86f71705e6030b2a36960756cb9f72dfe50d7115", "sources": ["arxiv", "semantic_scholar"], "title": "ViSpec: Accelerating Vision-Language Models with Vision-Aware Speculative Decoding", "abstract": "Speculative decoding is a widely adopted technique for accelerating inference in large language models (LLMs), yet its application to vision-language models (VLMs) remains underexplored, with existing methods achieving only modest speedups (<1.5x). This gap is increasingly significant as multimodal capabilities become central to large-scale models. We hypothesize that large VLMs can effectively filter redundant image information layer by layer without compromising textual comprehension, whereas smaller draft models struggle to do so. To address this, we introduce Vision-Aware Speculative Decoding (ViSpec), a novel framework tailored for VLMs. ViSpec employs a lightweight vision adaptor module to compress image tokens into a compact representation, which is seamlessly integrated into the draft model's attention mechanism while preserving original image positional information. Additionally, we extract a global feature vector for each input image and augment all subsequent text tokens with this feature to enhance multimodal coherence. To overcome the scarcity of multimodal datasets with long assistant responses, we curate a specialized training dataset by repurposing existing datasets and generating extended outputs using the target VLM with modified prompts. Our training strategy mitigates the risk of the draft model exploiting direct access to the target model's hidden states, which could otherwise lead to shortcut learning when training solely on target model outputs. Extensive experiments validate ViSpec, achieving, to our knowledge, the first substantial speedup in VLM speculative decoding. Code is available at https://github.com/KangJialiang/ViSpec.", "authors": ["Jialiang Kang", "Han Shu", "Wenshuo Li", "Yingjie Zhai", "Xinghao Chen"], "categories": ["cs.CV", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-17", "url": "https://arxiv.org/abs/2509.15235", "pdf_url": "https://arxiv.org/pdf/2509.15235v5", "arxiv_id": "2509.15235", "doi": "10.48550/arXiv.2509.15235", "citation_count": 15, "influential_citation_count": 4, "has_code": true, "code_url": "https://github.com/KangJialiang/ViSpec", "venue": "arXiv.org", "quality_score": 0.4693}
{"id": "861ef963718f14257628c3a92f01282981a171aae5b147d92221f8ef465dde3f", "sources": ["arxiv", "semantic_scholar"], "title": "FastMTP: Accelerating LLM Inference with Enhanced Multi-Token Prediction", "abstract": "As large language models (LLMs) become increasingly powerful, the sequential nature of autoregressive generation creates a fundamental throughput bottleneck that limits the practical deployment. While Multi-Token Prediction (MTP) has demonstrated remarkable benefits for model training efficiency and performance, its inherent potential for inference acceleration remains largely unexplored. This paper introduces FastMTP, a simple yet effective method that improves multi-step draft quality by aligning MTP training with its inference pattern, significantly enhancing speculative decoding performance. Our approach fine-tunes a single MTP head with position-shared weights on self-distilled data, enabling it to capture dependencies among consecutive future tokens and maintain high acceptance rates across multiple recursive draft steps. By integrating language-aware dynamic vocabulary compression into the MTP head, we further reduce computational overhead in the drafting process. Experimental results across seven diverse benchmarks demonstrate that FastMTP achieves an average of 2.03x speedup compared to standard next token prediction with lossless output quality, outperforming vanilla MTP by 82%. FastMTP requires only lightweight training and seamlessly integrates with existing inference frameworks, offering a practical and rapidly deployable solution for accelerating LLM inference.", "authors": ["Yuxuan Cai", "Xiaozhuan Liang", "Xinghua Wang", "Jin Ma", "Haijin Liang", "Jinwen Luo", "Xinyu Zuo", "Lisheng Duan", "Yuyang Yin", "Xi Chen"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-16", "url": "https://arxiv.org/abs/2509.18362", "pdf_url": "https://arxiv.org/pdf/2509.18362v1", "arxiv_id": "2509.18362", "doi": "10.48550/arXiv.2509.18362", "citation_count": 9, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3025}
{"id": "5637bbb97ebe7f7da09af4390909e01a3877beed82e4a714cbf0fe37066ddc7c", "sources": ["arxiv", "semantic_scholar"], "title": "Spec-LLaVA: Accelerating Vision-Language Models with Dynamic Tree-Based Speculative Decoding", "abstract": "Vision-Language Models (VLMs) enable powerful multimodal reasoning but suffer from slow autoregressive inference, limiting their deployment in real-time applications. We introduce Spec-LLaVA, a system that applies speculative decoding to accelerate VLMs without sacrificing output quality. Spec-LLaVA pairs a lightweight draft VLM with a large target model: the draft speculates future tokens, which the target verifies in parallel, allowing multiple tokens to be generated per step. To maximize efficiency, we design a dynamic tree-based verification algorithm that adaptively expands and prunes speculative branches using draft model confidence. On MS COCO out-of-domain images, Spec-LLaVA achieves up to 3.28$\\times$ faster decoding on LLaVA-1.5 (7B, 13B) with no loss in generation quality. This work presents a lossless acceleration framework for VLMs using dynamic tree-structured speculative decoding, opening a path toward practical real-time multimodal assistants. Importantly, the lightweight draft model design makes the framework amenable to resource-constrained or on-device deployment settings.", "authors": ["Mingxiao Huo", "Jiayi Zhang", "Hewei Wang", "Jinfeng Xu", "Zheyu Chen", "Huilin Tai", "Yijun Chen"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-15", "url": "https://arxiv.org/abs/2509.11961", "pdf_url": "https://arxiv.org/pdf/2509.11961v1", "arxiv_id": "2509.11961", "doi": "10.48550/arXiv.2509.11961", "citation_count": 5, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3014}
{"id": "894f7a3f734e1b05aa2fc523968b6e32e79bd6352c39ee554082de048d7992aa", "sources": ["arxiv", "semantic_scholar"], "title": "SpecVLM: Fast Speculative Decoding in Vision-Language Models", "abstract": "Speculative decoding is a powerful way to accelerate autoregressive large language models (LLMs), but directly porting it to vision-language models (VLMs) faces unique systems constraints: the prefill stage is dominated by visual tokens whose count scales with image resolution and video length, inflating both compute and memory, especially the key-value (KV) cache. We study speculative decoding for VLMs and introduce SpecVLM, a practical system that (1) establishes a strong EAGLE-2-style baseline, EagleVLM, delivering 1.5--2.3x end-to-end speedups over full autoregressive inference, and (2) further accelerates VLM inference with an elastic visual compressor that adaptively selects among pruning, pooling, convolution, and resampler primitives to balance FLOPs/parameters and accuracy per input. To avoid costly offline distillation corpora, we propose an online-logit distillation protocol that trains the draft model with on-the-fly teacher logits and penultimate features using a combined cross-entropy and Smooth L1 objective, eliminating storage and preprocessing while remaining compute-efficient. This protocol reveals a training-time scaling effect: longer online training monotonically increases the draft model's average accepted length, improving speculative efficiency. Empirically, SpecVLM achieves additional acceleration, culminating in 2.5--2.9x end-to-end speedups within 5 epochs across LLaVA and MMMU, consistently over resolutions and task difficulties, while preserving the target model's output distribution (lossless decoding). Our code is available at https://github.com/haiduo/SpecVLM.", "authors": ["Haiduo Huang", "Fuwei Yang", "Zhenhua Liu", "Xuanwu Yin", "Dong Li", "Pengju Ren", "Emad Barsoum"], "categories": ["cs.CV", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-15", "url": "https://arxiv.org/abs/2509.11815", "pdf_url": "https://arxiv.org/pdf/2509.11815v2", "arxiv_id": "2509.11815", "doi": "10.48550/arXiv.2509.11815", "citation_count": 4, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/haiduo/SpecVLM", "venue": "arXiv.org", "quality_score": 0.4657}
{"id": "260f3fc8beeb9cd64567ffe55ff4df7e5c33e1573922c3a460a9e3dc0b8fdf18", "sources": ["arxiv", "semantic_scholar"], "title": "Communication-Efficient Collaborative LLM Inference via Distributed Speculative Decoding", "abstract": "Speculative decoding is an emerging technique that accelerates large language model (LLM) inference by allowing a smaller draft model to predict multiple tokens in advance, which are then verified or corrected by a larger target model. In AI-native radio access networks (AI-RAN), this paradigm is well-suited for collaborative inference between resource-constrained end devices and more capable edge servers or base stations (BSs). However, existing distributed speculative decoding requires transmitting the full vocabulary probability distribution from the draft model on the device to the target model at the BS, which leads to prohibitive uplink communication overhead. To address this issue, we propose a ``Top-K Sparse Logits Transmission (TK-SLT)`` scheme, where the draft model transmits only the top-K token raw probabilities and the corresponding token indices instead of the entire distribution. This approach significantly reduces bandwidth consumption while maintaining inference performance. We further derive an analytical expression for the optimal draft length that maximizes inference throughput, and provide a theoretical analysis of the achievable speedup ratio under TK-SLT. Experimental results validate both the efficiency and effectiveness of the proposed method.", "authors": ["Ce Zheng", "Tingting Yang"], "categories": ["eess.SP"], "fields_of_study": ["Engineering"], "published_date": "2025-09-04", "url": "https://arxiv.org/abs/2509.04576", "pdf_url": "https://arxiv.org/pdf/2509.04576v2", "arxiv_id": "2509.04576", "doi": "10.1109/WCSP68525.2025.1010651", "citation_count": 6, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "International Conference on Wireless Communications and Signal Processing", "quality_score": 0.2888}
{"id": "887a9689b0b953df920246b4e40c7c8527c9c53738df7c46de2c0135b29dca2b", "sources": ["arxiv", "semantic_scholar"], "title": "Can LLMs Lie? Investigation beyond Hallucination", "abstract": "Large language models (LLMs) have demonstrated impressive capabilities across a variety of tasks, but their increasing autonomy in real-world applications raises concerns about their trustworthiness. While hallucinations-unintentional falsehoods-have been widely studied, the phenomenon of lying, where an LLM knowingly generates falsehoods to achieve an ulterior objective, remains underexplored. In this work, we systematically investigate the lying behavior of LLMs, differentiating it from hallucinations and testing it in practical scenarios. Through mechanistic interpretability techniques, we uncover the neural mechanisms underlying deception, employing logit lens analysis, causal interventions, and contrastive activation steering to identify and control deceptive behavior. We study real-world lying scenarios and introduce behavioral steering vectors that enable fine-grained manipulation of lying tendencies. Further, we explore the trade-offs between lying and end-task performance, establishing a Pareto frontier where dishonesty can enhance goal optimization. Our findings contribute to the broader discourse on AI ethics, shedding light on the risks and potential safeguards for deploying LLMs in high-stakes environments. Code and more illustrations are available at https://llm-liar.github.io/", "authors": ["Haoran Huan", "Mihir Prabhudesai", "Mengning Wu", "Shantanu Jaiswal", "Deepak Pathak"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-09-03", "url": "https://arxiv.org/abs/2509.03518", "pdf_url": "https://arxiv.org/pdf/2509.03518v1", "arxiv_id": "2509.03518", "doi": "10.48550/arXiv.2509.03518", "citation_count": 6, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2876}
{"id": "bb7f67abf4e114b9bdedb911990b4d6d39f22f8a60ac6048b1d98f7f167376df", "sources": ["arxiv", "semantic_scholar"], "title": "DSDE: Dynamic Speculative Decoding with KLD Stability for Real-World Serving", "abstract": "Speculative decoding accelerates large language model inference, but its reliance on a fixed speculation length is suboptimal in large-batch serving environments with diverse requests. This paper explores a new direction for dynamic adaptation by investigating a novel class of post-hoc, diagnostic signals. We propose Dynamic Speculative Decoding Engine (DSDE), a training-free framework built on two primary components: (1) a predictive signal based on the variance of the Kullback-Leibler (KLD) divergence, which diagnoses the generation's regional stability, and (2) an adaptive speculation length cap to mitigate the straggler problem in per-sequence decoding. Experiments demonstrate the potential of using KLD-based stability signals for dynamic adaptation. An algorithm guided by these signals achieves end-to-end latency competitive with leading baselines and exhibits superior robustness across diverse workloads. This robustness is particularly valuable in challenging low-acceptance-rate regimes, where the proposed signal maintains its diagnostic utility. Collectively, these findings validate post-hoc signals as a valuable component for building more robust and intelligent LLM inference systems, and highlight a promising direction for future research on dynamic speculation length adaptation.", "authors": ["Mingyu Yang", "Jae-Young Choi", "Kihyo Moon", "Minsung Jang", "Eunjoo Jeon"], "categories": ["cs.DC", "cs.AI", "cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2025-09-01", "url": "https://arxiv.org/abs/2509.01083", "pdf_url": "https://arxiv.org/pdf/2509.01083v3", "arxiv_id": "2509.01083", "doi": "10.1109/BigData66926.2025.11400767", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "BigData Congress [Services Society]", "quality_score": 0.2853}
{"id": "0c7bd14df643412aeb3523fe539bdd2c708cd6a197acad42833a7fe4134a4c34", "sources": ["arxiv", "semantic_scholar"], "title": "Scaling Up, Speeding Up: A Benchmark of Speculative Decoding for Efficient LLM Test-Time Scaling", "abstract": "Test-time scaling has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models (LLMs) by allocating additional computational resources during inference. However, this paradigm is inherently inefficient due to the generation of redundant and repetitive reasoning traces, leading to significant computational overhead. Speculative decoding offers a promising avenue for mitigating this inefficiency, yet its efficacy in the structured, repetition-rich context of test-time scaling remains largely unexplored. To bridge this gap, we introduce the first comprehensive benchmark designed to evaluate speculative decoding methods for accelerating LLM test-time scaling. Our benchmark provides consistent experimental protocols across representative test-time scaling paradigms (e.g., Best-of-N sampling and multi-round thinking), enabling a fair comparison of three major categories of speculative decoding: model-based, training-based, and n-gram-based methods. Extensive experiments reveal that simple n-gram-based methods effectively capture repetitive patterns, demonstrating unique potential in accelerating test-time scaling. This phenomenon demonstrates the value of integrating n-gram-based methods with model-based or training-based approaches to balance acceleration for both repetitive and diverse reasoning in test-time scaling. We hope this benchmark spurs further research on speculative decoding for test-time scaling, enabling faster and more practical reasoning in LLMs through better handling of repetitive and diverse reasoning paths.", "authors": ["Shengyin Sun", "Yiming Li", "Xing Li", "Yingzhao Lian", "Weizhe Lin", "Hui-Ling Zhen", "Zhiyuan Yang", "Chen Chen", "Xianzhi Yu", "Mingxuan Yuan", "Chen Ma"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-30", "url": "https://arxiv.org/abs/2509.04474", "pdf_url": "https://arxiv.org/pdf/2509.04474v1", "arxiv_id": "2509.04474", "doi": "10.48550/arXiv.2509.04474", "citation_count": 5, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.283}
{"id": "15d6a2f927d4c2ebd145bd467769306cfdc6d0ccd385a856fcf02f73aae45288", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Mixture-of-Experts Inference by Hiding Offloading Latency with Speculative Decoding", "abstract": "Recent advancements in Mixture of Experts (MoE) models have significantly increased their parameter scale as well as model performance. Extensive offloading techniques have been proposed to address the GPU memory limitations of MoE inference. However, due to the I/O bottleneck and sparse computation of MoE models, existing offloading techniques still suffer from low hardware utilization. To fully utilize the hardware resources, we propose SpecMoEOff, which employs the speculative decoding technique to enlarge the workload of each expert. SpecMoEOff orchestrates the GPU and CPU by both theoretical and empirical roofline analysis. In addition, we develop a dedicated CPU chunked attention verification kernel to fit the speculative decoding in offloading scenarios as well as minimizing the additional overhead led by draft models. SpecMoEOff further integrates an optimizer to automatically tune the hyperparameters of speculative decoding under given hardware and workload. Experimental results show that SpecMoEOff achieves up to 2.5x decode throughput improvement over the state-of-the-art MoE offloading techniques.", "authors": ["Zhibin Wang", "Zhonghui Zhang", "Yuhang Zhou", "Zibo Wang", "Mo Zhou", "Peng Jiang", "Weilin Cai", "Chengying Huan", "Rong Gu", "Sheng Zhong", "Chen Tian"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-29", "url": "https://arxiv.org/abs/2508.21706", "pdf_url": "https://arxiv.org/pdf/2508.21706v2", "arxiv_id": "2508.21706", "doi": "10.48550/arXiv.2508.21706", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2819}
{"id": "f54025c03f3782e8ebc58ad19d73578c6a0901e475080f92d8f495e8e518ca06", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Safety-Aware Decoding", "abstract": "Despite extensive efforts to align Large Language Models (LLMs) with human values and safety rules, jailbreak attacks that exploit certain vulnerabilities continuously emerge, highlighting the need to strengthen existing LLMs with additional safety properties to defend against these attacks. However, tuning large models has become increasingly resource intensive and may have difficulty ensuring consistent performance. We introduce Speculative Safety-Aware Decoding (SSD), a lightweight decoding-time approach that equips LLMs with the desired safety property while accelerating inference. We assume that there exists a small language model that possesses this desired property. SSD integrates speculative sampling during decoding and leverages the match ratio between the small and composite models to quantify jailbreak risks. This enables SSD to dynamically switch between decoding schemes to prioritize utility or safety, to handle the challenge of different model capacities. The output token is then sampled from a new distribution that combines the distributions of the original and the small models. Experimental results show that SSD successfully equips the large model with the desired safety property, and also allows the model to remain helpful to benign queries. Furthermore, SSD accelerates the inference time, thanks to the speculative sampling design.", "authors": ["Xuekang Wang", "Shengyu Zhu", "Xueqi Cheng"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-25", "url": "https://arxiv.org/abs/2508.17739", "pdf_url": "https://arxiv.org/pdf/2508.17739v2", "arxiv_id": "2508.17739", "doi": "10.48550/arXiv.2508.17739", "citation_count": 2, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/k-k1w-w1x-x/Speculative-Safety-Aware-Decoding", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.4285}
{"id": "582310ee96001a9bd00e7ea681e1420d7ad51ece472ffc4e9d87a8f3056da3ee", "sources": ["arxiv", "semantic_scholar"], "title": "SpecVLM: Enhancing Speculative Decoding of Video LLMs via Verifier-Guided Token Pruning", "abstract": "Video large language models (Vid-LLMs) have shown strong capabilities in understanding video content. However, their reliance on dense video token representations introduces substantial memory and computational overhead in both prefilling and decoding. To mitigate the information loss of recent video token reduction methods and accelerate the decoding stage of Vid-LLMs losslessly, we introduce SpecVLM, a training-free speculative decoding (SD) framework tailored for Vid-LLMs that incorporates staged video token pruning. Building on our novel finding that the draft model's speculation exhibits low sensitivity to video token pruning, SpecVLM prunes up to 90% of video tokens to enable efficient speculation without sacrificing accuracy. To achieve this, we performs a two-stage pruning process: Stage I selects highly informative tokens guided by attention signals from the verifier (target model), while Stage II prunes remaining redundant ones in a spatially uniform manner. Extensive experiments on four video understanding benchmarks demonstrate the effectiveness and robustness of SpecVLM, which achieves up to 2.68$\\times$ decoding speedup for LLaVA-OneVision-72B and 2.11$\\times$ speedup for Qwen2.5-VL-32B. Code is available at https://github.com/zju-jiyicheng/SpecVLM.", "authors": ["Yicheng Ji", "Jun Zhang", "Heming Xia", "Jinpeng Chen", "Lidan Shou", "Gang Chen", "Huan Li"], "categories": ["cs.CV", "cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-22", "url": "https://arxiv.org/abs/2508.16201", "pdf_url": "https://arxiv.org/pdf/2508.16201v2", "arxiv_id": "2508.16201", "doi": "10.48550/arXiv.2508.16201", "citation_count": 18, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/zju-jiyicheng/SpecVLM", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.4232}
{"id": "df8a049eebf8bd12ee5724231258017456ae08970dde5a3d8e428415223a47b5", "sources": ["arxiv", "semantic_scholar"], "title": "Confidence-Modulated Speculative Decoding for Large Language Models", "abstract": "Speculative decoding has emerged as an effective approach for accelerating autoregressive inference by parallelizing token generation through a draft-then-verify paradigm. However, existing methods rely on static drafting lengths and rigid verification criteria, limiting their adaptability across varying model uncertainties and input complexities. This paper proposes an information-theoretic framework for speculative decoding based on confidence-modulated drafting. By leveraging entropy and margin-based uncertainty measures over the drafter's output distribution, the proposed method dynamically adjusts the number of speculatively generated tokens at each iteration. This adaptive mechanism reduces rollback frequency, improves resource utilization, and maintains output fidelity. Additionally, the verification process is modulated using the same confidence signals, enabling more flexible acceptance of drafted tokens without sacrificing generation quality. Experiments on machine translation and summarization tasks demonstrate significant speedups over standard speculative decoding while preserving or improving BLEU and ROUGE scores. The proposed approach offers a principled, plug-in method for efficient and robust decoding in large language models under varying conditions of uncertainty.", "authors": ["Jaydip Sen", "Subhasis Dasgupta", "Hetvi Waghela"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-21", "url": "https://arxiv.org/abs/2508.15371", "pdf_url": "https://arxiv.org/pdf/2508.15371v1", "arxiv_id": "2508.15371", "doi": "10.1109/INDISCON66021.2025.11254640", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.1735}
{"id": "8c765d98efbd5162ad8be764e6c51a4f0c79dc8246c332914d98bb9bdc8fcef7", "sources": ["arxiv", "semantic_scholar"], "title": "READER: Retrieval-Assisted Drafter for Efficient LLM Inference", "abstract": "Autoregressive Language Models instantiate a factorized likelihood over token sequences, yet their strictly sequential decoding process imposes an intrinsic lower bound on inference latency. This bottleneck has emerged as a central obstacle to the scalable deployment of large-scale generative models. Existing acceleration techniques partially mitigate token-level latency by relying on auxiliary draft models or introducing an additional training phase, but fail to address the dominant memory and communication costs. We present READER, a provably lossless speculative decoding framework that bypasses the training of the auxiliary draft model. READER formalizes speculative decoding as a stochastic tree construction problem and exploits the empirical redundancy structure of natural language to generate high-probability candidate continuations. Our method revisits the problem of constructing draft trees, establishing substantial statistical improvements over stochastic draft-tree methods and providing a complexity-theoretic analysis that characterizes the optimality frontier of speculative decoding under bounded computation and memory resources. Beyond the single-sequence regime traditionally considered in prior work, we introduce a memory-optimal key-value cache-serving strategy that guarantees amortized sublinear overhead in the batch dimension, allowing READER to scale to realistic inference workloads. Comprehensive experiments demonstrate up to 6.13x wall-clock speedup on single-prompt inference and up to 5.92x on batched inference, consistently surpassing prior speculative decoding baselines, while preserving exact output equivalence, with even more pronounced gains in retrieval-augmented generation pipelines. Our results close a key gap between theoretical parallelism limits and practical LLM inference, suggesting a new standard for efficient deployment.", "authors": ["Maxim Divilkovskiy", "Vitaly Malygin", "Sergey Zlobin", "Stanislav Ilyushin", "Sultan Isali", "Vasily Kalugin", "Nuriza Aitassova", "Fei Yi", "Weidi Zeng"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-12", "url": "https://arxiv.org/abs/2508.09072", "pdf_url": "https://arxiv.org/pdf/2508.09072v2", "arxiv_id": "2508.09072", "doi": "10.48550/arXiv.2508.09072", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2624}
{"id": "f8812d9aa15cb01e05acf34eacaec0b9426cbdb76b6db7cd19cc7507d7a0cf6d", "sources": ["arxiv", "semantic_scholar"], "title": "ASPD: Unlocking Adaptive Serial-Parallel Decoding by Exploring Intrinsic Parallelism in LLMs", "abstract": "The increasing scale and complexity of large language models (LLMs) pose significant inference latency challenges, primarily due to their autoregressive decoding paradigm characterized by the sequential nature of next-token prediction. By re-examining the outputs of autoregressive models, we observed that some segments exhibit parallelizable structures, which we term intrinsic parallelism. Decoding each parallelizable branch simultaneously (i.e. parallel decoding) can significantly improve the overall inference speed of LLMs. In this paper, we propose an Adaptive Serial-Parallel Decoding (ASPD), which addresses two core challenges: automated construction of parallelizable data and efficient parallel decoding mechanism. More specifically, we introduce a non-invasive pipeline that automatically extracts and validates parallelizable structures from the responses of autoregressive models. To empower efficient adaptive serial-parallel decoding, we implement a Hybrid Decoding Engine which enables seamless transitions between serial and parallel decoding modes while maintaining a reusable KV cache, maximizing computational efficiency. Extensive evaluations across General Tasks, Retrieval-Augmented Generation, Mathematical Reasoning, demonstrate that ASPD achieves unprecedented performance in both effectiveness and efficiency. Notably, on Vicuna Bench, our method achieves up to 3.19x speedup (1.85x on average) while maintaining response quality within 1% difference compared to autoregressive models, realizing significant acceleration without compromising generation quality. Our framework sets a groundbreaking benchmark for efficient LLM parallel inference, paving the way for its deployment in latency-sensitive applications such as AI-powered customer service bots and answer retrieval engines.", "authors": ["Keyu Chen", "Zhifeng Shen", "Daohai Yu", "Haoqian Wu", "Wei Wen", "Jianfeng He", "Ruizhi Qiao", "Xing Sun"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-12", "url": "https://arxiv.org/abs/2508.08895", "pdf_url": "https://arxiv.org/pdf/2508.08895v2", "arxiv_id": "2508.08895", "doi": "10.48550/arXiv.2508.08895", "citation_count": 6, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2624}
{"id": "3e9fa77cdc3447792bc8baf19bbffe001e1efaf16379fe10d5dc1544e350cc6b", "sources": ["arxiv", "semantic_scholar"], "title": "Efficient Speculative Decoding for Llama at Scale: Challenges and Solutions", "abstract": "Speculative decoding is a standard method for accelerating the inference speed of large language models. However, scaling it for production environments poses several engineering challenges, including efficiently implementing different operations (e.g., tree attention and multi-round speculative decoding) on GPU. In this paper, we detail the training and inference optimization techniques that we have implemented to enable EAGLE-based speculative decoding at a production scale for Llama models. With these changes, we achieve a new state-of-the-art inference latency for Llama models. For example, Llama4 Maverick decodes at a speed of about 4 ms per token (with a batch size of one) on 8 NVIDIA H100 GPUs, which is 10% faster than the previously best known method. Furthermore, for EAGLE-based speculative decoding, our optimizations enable us to achieve a speed-up for large batch sizes between 1.4x and 2.0x at production scale.", "authors": ["Bangsheng Tang", "Carl Chengyan Fu", "Fei Kou", "Grigory Sizov", "Haoci Zhang", "Jason Park", "Jiawen Liu", "Jie You", "Qirui Yang", "Sachin Mehta", "Shengyong Cai", "Xiaodong Wang", "Xingyu Liu", "Yunlu Li", "Yanjun Zhou", "Wei Wei", "Zhiwei Zhao", "Zixi Qi", "Adolfo Victoria", "Aya Ibrahim", "Bram Wasti", "Changkyu Kim", "Daniel Haziza", "Fei Sun", "Giancarlo Delfin", "Emily Guo", "Jialin Ouyang", "Jaewon Lee", "Jianyu Huang", "Jeremy Reizenstein", "Lu Fang", "Quinn Zhu", "Ria Verma", "Vlad Mihailescu", "Xingwen Guo", "Yan Cui", "Ye Hu", "Yejin Lee"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-11", "url": "https://arxiv.org/abs/2508.08192", "pdf_url": "https://arxiv.org/pdf/2508.08192v1", "arxiv_id": "2508.08192", "doi": "10.48550/arXiv.2508.08192", "citation_count": 8, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2612}
{"id": "d56d5385f04fa3dd872f88fe9cdf2f251ffdcbf64156bff3310095f5975bc234", "sources": ["arxiv", "semantic_scholar"], "title": "Grouped Speculative Decoding for Autoregressive Image Generation", "abstract": "Recently, autoregressive (AR) image models have demonstrated remarkable generative capabilities, positioning themselves as a compelling alternative to diffusion models. However, their sequential nature leads to long inference times, limiting their practical scalability. In this work, we introduce Grouped Speculative Decoding (GSD), a novel, training-free acceleration method for AR image models. While recent studies have explored Speculative Decoding (SD) as a means to speed up AR image generation, existing approaches either provide only modest acceleration or require additional training. Our in-depth analysis reveals a fundamental difference between language and image tokens: image tokens exhibit inherent redundancy and diversity, meaning multiple tokens can convey valid semantics. However, traditional SD methods are designed to accept only a single most-likely token, which fails to leverage this difference, leading to excessive false-negative rejections. To address this, we propose a new SD strategy that evaluates clusters of visually valid tokens rather than relying on a single target token. Additionally, we observe that static clustering based on embedding distance is ineffective, which motivates our dynamic GSD approach. Extensive experiments show that GSD accelerates AR image models by an average of 3.7x while preserving image quality-all without requiring any additional training. The source code is available at https://github.com/junhyukso/GSD", "authors": ["Junhyuk So", "Juncheol Shin", "Hyunho Kook", "Eunhyeok Park"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-11", "url": "https://arxiv.org/abs/2508.07747", "pdf_url": "https://arxiv.org/pdf/2508.07747v1", "arxiv_id": "2508.07747", "doi": "10.1109/ICCV51701.2025.01426", "citation_count": 13, "influential_citation_count": 6, "has_code": true, "code_url": "https://github.com/junhyukso/GSD", "venue": "IEEE International Conference on Computer Vision", "quality_score": 0.4225}
{"id": "43d2589c16df6f1d0cdafaf9faa07e9245abc1198310a3e99ea6eae9fb251354", "sources": ["arxiv", "semantic_scholar"], "title": "LP-Spec: Leveraging LPDDR PIM for Efficient LLM Mobile Speculative Inference with Architecture-Dataflow Co-Optimization", "abstract": "LLM inference on mobile devices faces extraneous challenges due to limited memory bandwidth and computational resources. To address these issues, speculative inference and processing-in-memory (PIM) techniques have been explored at the algorithmic and hardware levels. However, speculative inference results in more compute-intensive GEMM operations, creating new design trade-offs for existing GEMV-accelerated PIM architectures. Furthermore, there exists a significant amount of redundant draft tokens in tree-based speculative inference, necessitating efficient token management schemes to minimize energy consumption. In this work, we present LP-Spec, an architecture-dataflow co-design leveraging hybrid LPDDR5 performance-enhanced PIM architecture with draft token pruning and dynamic workload scheduling to accelerate LLM speculative inference. A near-data memory controller is proposed to enable data reallocation between DRAM and PIM banks. Furthermore, a data allocation unit based on the hardware-aware draft token pruner is developed to minimize energy consumption and fully exploit parallel execution opportunities. Compared to end-to-end LLM inference on other mobile solutions such as mobile NPUs or GEMV-accelerated PIMs, our LP-Spec achieves 13.21x, 7.56x, and 99.87x improvements in performance, energy efficiency, and energy-delay-product (EDP). Compared with prior AttAcc PIM and RTX 3090 GPU, LP-Spec can obtain 12.83x and 415.31x EDP reduction benefits.", "authors": ["Siyuan He", "Zhantong Zhu", "Yandong He", "Tianyu Jia"], "categories": ["cs.AR"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-10", "url": "https://arxiv.org/abs/2508.07227", "pdf_url": "https://arxiv.org/pdf/2508.07227v3", "arxiv_id": "2508.07227", "doi": "10.1109/ICCAD66269.2025.11240889", "citation_count": 6, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.2113}
{"id": "041b82597f1ba8ca7a864e962a2dad0f6bcebaad68080ae301c91dc6af7b3b7f", "sources": ["arxiv", "semantic_scholar"], "title": "CARD: A Cache-Assisted Parallel Speculative Decoding Framework via Query-and-Correct Paradigm for Accelerating LLM Inference", "abstract": "Speculative decoding (SD), where a draft model provides multiple candidate tokens for the target model to verify in parallel, has demonstrated significant potential for accelerating LLM inference. Yet, existing SD approaches adhere to a strict draft-then-verify paradigm, enforcing a sequential process that hampers performance and constrains the draft model's capacity. Moreover, rejecting a token in the candidate sequence invalidates all subsequent tokens, leading to wasted computation during drafting. To overcome these limitations, we propose a cache-assisted parallel speculative decoding framework called CARD, which employs a novel query-and-correct paradigm. Our approach decouples drafting from verification: the draft model populates a shared cache with candidate tokens, while the target model concurrently refines the draft's trajectory. This enables inference at near-draft-speed, effectively leveraging the draft model's efficiency without additional fine-tuning. Experimental results show that CARD significantly outperforms existing state-of-the-art methods, achieving up to a 4.83x acceleration over vanilla autoregressive decoding, with no fine-tuning required for either models.", "authors": ["Enyu Zhou", "Kai Sheng", "Hao Chen", "Xin He"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-08-06", "url": "https://arxiv.org/abs/2508.04462", "pdf_url": "https://arxiv.org/pdf/2508.04462v2", "arxiv_id": "2508.04462", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.1626}
{"id": "c6c07b080e6989ff636f08bcf3b5c24573c2cd66272f03e7f1a517e50ae82fb3", "sources": ["arxiv", "semantic_scholar"], "title": "XSpecMesh: Quality-Preserving Auto-Regressive Mesh Generation Acceleration via Multi-Head Speculative Decoding", "abstract": "Current auto-regressive models can generate high-quality, topologically precise meshes; however, they necessitate thousands-or even tens of thousands-of next-token predictions during inference, resulting in substantial latency. We introduce XSpecMesh, a quality-preserving acceleration method for auto-regressive mesh generation models. XSpecMesh employs a lightweight, multi-head speculative decoding scheme to predict multiple tokens in parallel within a single forward pass, thereby accelerating inference. We further propose a verification and resampling strategy: the backbone model verifies each predicted token and resamples any tokens that do not meet the quality criteria. In addition, we propose a distillation strategy that trains the lightweight decoding heads by distilling from the backbone model, encouraging their prediction distributions to align and improving the success rate of speculative predictions. Extensive experiments demonstrate that our method achieves a 1.7x speedup without sacrificing generation quality. Our code will be released.", "authors": ["Dian Chen", "Yansong Qu", "Xinyang Li", "Ming Li", "Shengchuan Zhang"], "categories": ["cs.GR", "cs.CV", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-31", "url": "https://arxiv.org/abs/2507.23777", "pdf_url": "https://arxiv.org/pdf/2507.23777v2", "arxiv_id": "2507.23777", "doi": "10.48550/arXiv.2507.23777", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2486}
{"id": "3a2d2d259747232165d5fd7ac6603592bd111242e4cadacb951d2f7a58c63a1a", "sources": ["arxiv", "semantic_scholar"], "title": "Hierarchical Verification of Speculative Beams for Accelerating LLM Inference", "abstract": "Large language models (LLMs) have achieved remarkable success across diverse natural language processing tasks but face persistent challenges in inference efficiency due to their autoregressive nature. While speculative decoding and beam sampling offer notable improvements, traditional methods verify draft sequences sequentially without prioritization, leading to unnecessary computational overhead. This work proposes the Hierarchical Verification Tree (HVT), a novel framework that restructures speculative beam decoding by prioritizing high-likelihood drafts and enabling early pruning of suboptimal candidates. Theoretical foundations and a formal verification-pruning algorithm are developed to ensure correctness and efficiency. Integration with standard LLM inference pipelines is achieved without requiring retraining or architecture modification. Experimental evaluations across multiple datasets and models demonstrate that HVT consistently outperforms existing speculative decoding schemes, achieving substantial reductions in inference time and energy consumption while maintaining or enhancing output quality. The findings highlight the potential of hierarchical verification strategies as a new direction for accelerating large language model inference.", "authors": ["Jaydip Sen", "Harshitha Puvvala", "Subhasis Dasgupta"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-30", "url": "https://arxiv.org/abs/2508.03726", "pdf_url": "https://arxiv.org/pdf/2508.03726v1", "arxiv_id": "2508.03726", "doi": "10.1007/978-3-032-07735-6_19", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2475}
{"id": "85a18671c07272727909d6c3889128af62a1391015f10e7c52d404e7ac5c7d23", "sources": ["arxiv", "semantic_scholar"], "title": "Spec-VLA: Speculative Decoding for Vision-Language-Action Models with Relaxed Acceptance", "abstract": "Vision-Language-Action (VLA) models have made substantial progress by leveraging the robust capabilities of Visual Language Models (VLMs). However, VLMs' significant parameter size and autoregressive (AR) decoding nature impose considerable computational demands on VLA models. While Speculative Decoding (SD) has shown efficacy in accelerating Large Language Models (LLMs) by incorporating efficient drafting and parallel verification, allowing multiple tokens to be generated in one forward pass, its application to VLA models remains unexplored. This work introduces Spec-VLA, an SD framework designed to accelerate VLA models. Due to the difficulty of the action prediction task and the greedy decoding mechanism of the VLA models, the direct application of the advanced SD framework to the VLA prediction task yields a minor speed improvement. To boost the generation speed, we propose an effective mechanism to relax acceptance utilizing the relative distances represented by the action tokens of the VLA model. Empirical results across diverse test scenarios affirm the effectiveness of the Spec-VLA framework, and further analysis substantiates the impact of our proposed strategies, which enhance the acceptance length by 44%, achieving 1.42 times speedup compared with the OpenVLA baseline, without compromising the success rate. The success of the Spec-VLA framework highlights the potential for broader application of speculative execution in VLA prediction scenarios.", "authors": ["Songsheng Wang", "Rucheng Yu", "Zhihang Yuan", "Chao Yu", "Feng Gao", "Yu Wang", "Derek F. Wong"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-30", "url": "https://arxiv.org/abs/2507.22424", "pdf_url": "https://arxiv.org/pdf/2507.22424v2", "arxiv_id": "2507.22424", "doi": "10.48550/arXiv.2507.22424", "citation_count": 25, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.3537}
{"id": "28a30d0efea8f062a95654d87654af295152898cc6351bc8d3810006e5bc765f", "sources": ["arxiv", "semantic_scholar"], "title": "Model-free Speculative Decoding for Transformer-based ASR with Token Map Drafting", "abstract": "End-to-end automatic speech recognition (ASR) systems based on transformer architectures, such as Whisper, offer high transcription accuracy and robustness. However, their autoregressive decoding is computationally expensive, hence limiting deployment on CPU-based and resource-constrained devices. Speculative decoding (SD) mitigates this issue by using a smaller draft model to propose candidate tokens, which are then verified by the main model. However, this approach is impractical for devices lacking hardware accelerators like GPUs. To address this, we propose \\emph{Token Map Drafting}, a model-free SD technique that eliminates the need for a separate draft model. Instead, we leverage a precomputed n-gram token map derived from domain-specific training data, enabling efficient speculative decoding with minimal overhead. Our method significantly accelerates ASR inference in structured, low-perplexity domains without sacrificing transcription accuracy. Experimental results demonstrate decoding speed-ups of $1.27\\times$ on the CI-AVSR dataset and $1.37\\times$ on our internal dataset without degrading recognition accuracy. Additionally, our approach achieves a $10\\%$ absolute improvement in decoding speed over the Distill-spec baseline running on CPU, highlighting its effectiveness for on-device ASR applications.", "authors": ["Tuan Vu Ho", "Hiroaki Kokubo", "Masaaki Yamamoto", "Yohei Kawaguchi"], "categories": ["cs.CL", "cs.SD", "eess.AS"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2025-07-29", "url": "https://arxiv.org/abs/2507.21522", "pdf_url": "https://arxiv.org/pdf/2507.21522v1", "arxiv_id": "2507.21522", "doi": "10.48550/arXiv.2507.21522", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "European Signal Processing Conference", "quality_score": 0.2464}
{"id": "0b3c5255fbfb5f2b3587ffe13e9e62fd1ed173b4ca757e9de08b52c5cbf100d7", "sources": ["arxiv", "semantic_scholar"], "title": "Enhancing Jailbreak Attacks on LLMs via Persona Prompts", "abstract": "Jailbreak attacks aim to exploit large language models (LLMs) by inducing them to generate harmful content, thereby revealing their vulnerabilities. Understanding and addressing these attacks is crucial for advancing the field of LLM safety. Previous jailbreak approaches have mainly focused on direct manipulations of harmful intent, with limited attention to the impact of persona prompts. In this study, we systematically explore the efficacy of persona prompts in compromising LLM defenses. We propose a genetic algorithm-based method that automatically crafts persona prompts to bypass LLM's safety mechanisms. Our experiments reveal that: (1) our evolved persona prompts reduce refusal rates by 50-70% across multiple LLMs, and (2) these prompts demonstrate synergistic effects when combined with existing attack methods, increasing success rates by 10-20%. Our code and data are available at https://github.com/CjangCjengh/Generic_Persona.", "authors": ["Zheng Zhang", "Peilin Zhao", "Deheng Ye", "Hao Wang"], "categories": ["cs.CR", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-28", "url": "https://arxiv.org/abs/2507.22171", "pdf_url": "https://arxiv.org/pdf/2507.22171v3", "arxiv_id": "2507.22171", "doi": "10.48550/arXiv.2507.22171", "citation_count": 8, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/CjangCjengh/Generic_Persona", "venue": "arXiv.org", "quality_score": 0.379}
{"id": "d422d6c1cc9967ba5dc818d0636667a38fa381d0d7f215ee46de2d7a29bacec8", "sources": ["arxiv", "semantic_scholar"], "title": "SpecASR: Accelerating LLM-based Automatic Speech Recognition via Speculative Decoding", "abstract": "Large language model (LLM)-based automatic speech recognition (ASR) has recently attracted a lot of attention due to its high recognition accuracy and enhanced multi-dialect support. However, the high decoding latency of LLMs challenges the real-time ASR requirements. Although speculative decoding has been explored for better decoding efficiency, they usually ignore the key characteristics of the ASR task and achieve limited speedup. To further reduce the real-time ASR latency, in this paper, we propose a novel speculative decoding framework specialized for ASR, dubbed SpecASR. SpecASR is developed based on our core observation that ASR decoding is audio-conditioned, which results in high output alignment between small and large ASR models, even given output mismatches in intermediate decoding steps. Therefore, SpecASR features an adaptive draft sequence generation process that dynamically modifies the draft sequence length to maximize the token acceptance length. SpecASR further proposes a draft sequence recycling strategy that reuses the previously generated draft sequence to reduce the draft ASR model latency. Moreover, a two-pass sparse token tree generation algorithm is also proposed to balance the latency of draft and target ASR models. With extensive experimental results, we demonstrate SpecASR achieves 3.04x-3.79x and 1.25x-1.84x speedup over the baseline autoregressive decoding and speculative decoding, respectively, without any loss in recognition accuracy.", "authors": ["Linye Wei", "Shuzhang Zhong", "Songqiang Xu", "Runsheng Wang", "Ru Huang", "Meng Li"], "categories": ["eess.AS", "cs.SD"], "fields_of_study": ["Engineering", "Computer Science"], "published_date": "2025-07-24", "url": "https://arxiv.org/abs/2507.18181", "pdf_url": "https://arxiv.org/pdf/2507.18181v2", "arxiv_id": "2507.18181", "doi": "10.1109/DAC63849.2025.11132579", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Design Automation Conference", "quality_score": 0.2406}
{"id": "f26ab641d82fc6750345b54d32f8c8e45e634b5d5fe1c950b291d0bdade436bc", "sources": ["arxiv", "semantic_scholar"], "title": "DSSD: Efficient Edge-Device LLM Deployment and Collaborative Inference via Distributed Split Speculative Decoding", "abstract": "Large language models (LLMs) have transformed natural language processing but face critical deployment challenges in device-edge systems due to resource limitations and communication overhead. To address these issues, collaborative frameworks have emerged that combine small language models (SLMs) on devices with LLMs at the edge, using speculative decoding (SD) to improve efficiency. However, existing solutions often trade inference accuracy for latency or suffer from high uplink transmission costs when verifying candidate tokens. In this paper, we propose Distributed Split Speculative Decoding (DSSD), a novel architecture that not only preserves the SLM-LLM split but also partitions the verification phase between the device and edge. In this way, DSSD replaces the uplink transmission of multiple vocabulary distributions with a single downlink transmission, significantly reducing communication latency while maintaining inference quality. Experiments show that our solution outperforms current methods, and codes are at: https://github.com/JasonNing96/DSSD-Efficient-Edge-Computing", "authors": ["Jiahong Ning", "Ce Zheng", "Tingting Yang"], "categories": ["eess.SP"], "fields_of_study": ["Engineering"], "published_date": "2025-07-16", "url": "https://arxiv.org/abs/2507.12000", "pdf_url": "https://arxiv.org/pdf/2507.12000v2", "arxiv_id": "2507.12000", "doi": null, "citation_count": 15, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/JasonNing96/DSSD-Efficient-Edge-Computing", "venue": null, "quality_score": 0.301}
{"id": "f9e6dd0ec9d17211d214885a175a019a55f6ebec66d0e2c187983e23c9a2083d", "sources": ["arxiv", "semantic_scholar"], "title": "TPP-SD: Accelerating Transformer Point Process Sampling with Speculative Decoding", "abstract": "We propose TPP-SD, a novel approach that accelerates Transformer temporal point process (TPP) sampling by adapting speculative decoding (SD) techniques from language models. By identifying the structural similarities between thinning algorithms for TPPs and speculative decoding for language models, we develop an efficient sampling framework that leverages a smaller draft model to generate multiple candidate events, which are then verified by the larger target model in parallel. TPP-SD maintains the same output distribution as autoregressive sampling while achieving significant acceleration. Experiments on both synthetic and real datasets demonstrate that our approach produces samples from identical distributions as standard methods, but with 2-6$\\times$ speedup. Our ablation studies analyze the impact of hyperparameters such as draft length and draft model size on sampling efficiency. TPP-SD bridges the gap between powerful Transformer TPP models and the practical need for rapid sequence sampling.", "authors": ["Shukai Gong", "Yiyang Fu", "Fengyuan Ran", "Quyu Kong", "Feng Zhou"], "categories": ["cs.LG", "stat.ML"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2025-07-12", "url": "https://arxiv.org/abs/2507.09252", "pdf_url": "https://arxiv.org/pdf/2507.09252v3", "arxiv_id": "2507.09252", "doi": "10.48550/arXiv.2507.09252", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2269}
{"id": "4d126d963477bc25ccf29dd451b567121cff15502148c062dcb405cdbcb2c70b", "sources": ["arxiv", "semantic_scholar"], "title": "On Evaluating Performance of LLM Inference Serving Systems", "abstract": "The rapid evolution of Large Language Model (LLM) inference systems has yielded significant efficiency improvements. However, our systematic analysis reveals that current evaluation methodologies frequently exhibit fundamental flaws, often manifesting as common evaluation anti-patterns that obscure true performance characteristics and impede scientific progress. Through a comprehensive examination of recent systems, we identify recurring anti-patterns across three key dimensions: Baseline Fairness, Evaluation Setup, and Metric Design. These anti-patterns are uniquely problematic for LLM inference due to its dual-phase nature combining distinct prefill and decode operations, its handling of highly heterogeneous workloads, and its strict temporal requirements for interactive use. We demonstrate how common anti-patterns -- such as inadequate baseline comparisons that conflate engineering effort with algorithmic novelty, workload selections that fail to represent production scenarios, and metric normalizations that hide substantial performance variability like generation stalls-lead to misleading conclusions. To address these challenges, we provide a comprehensive checklist derived from our analysis, establishing a framework for recognizing and avoiding these anti-patterns in favor of robust LLM inference evaluation. To demonstrate the practical application of our framework, we present a case study analyzing speculative decoding, a technique whose bursty, non-uniform token generation is easily misinterpreted when evaluated using approaches characteristic of these anti-patterns. Our work establishes a rigorous foundation for evaluation methodology, enabling meaningful comparisons, ensuring reproducible results, and ultimately accelerating genuine progress in LLM inference systems by moving beyond common anti-patterns to align evaluation with real-world requirements.", "authors": ["Amey Agrawal", "Nitin Kedia", "Anmol Agarwal", "Jayashree Mohan", "Nipun Kwatra", "Souvik Kundu", "Ramachandran Ramjee", "Alexey Tumanov"], "categories": ["cs.LG", "cs.AI", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-11", "url": "https://arxiv.org/abs/2507.09019", "pdf_url": "https://arxiv.org/pdf/2507.09019v1", "arxiv_id": "2507.09019", "doi": "10.48550/arXiv.2507.09019", "citation_count": 12, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2785}
{"id": "d98830d27b072526e589de5b88b174d3dea6db03e181cb63aecb1157d517a1b0", "sources": ["arxiv", "semantic_scholar"], "title": "On the Convergence Speed of Spatially Coupled LDPC Ensembles Under Window Decoding", "abstract": "It is known that windowed decoding (WD) can effectively balance the performance and complexity of spatially coupled low-density parity-check (LDPC) codes. In this study, we show that information can propagate in a wave-like manner at a constant speed under WD. Additionally, we provide an upper bound for the information propagation speed on the binary erasure channel, which can assist in designing the number of iterations required within each window.", "authors": ["Qingqing Peng", "Dongxu Chang", "Guanghui Wang", "Guiying Yan"], "categories": ["cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2025-07-09", "url": "https://arxiv.org/abs/2507.06635", "pdf_url": "https://arxiv.org/pdf/2507.06635v1", "arxiv_id": "2507.06635", "doi": "10.1109/ITW62417.2025.11240495", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Information Theory Workshop", "quality_score": 0.2234}
{"id": "fd59f8b8e506cc9333dfce85a86e5d3b2b6848c31954a0c821c378a20756df50", "sources": ["arxiv", "semantic_scholar"], "title": "Cascade: Token-Sharded Private LLM Inference", "abstract": "As LLMs continue to increase in parameter size, the computational resources required to run them are available to fewer parties. Therefore, third-party inference services -- where LLMs are hosted by third parties with significant computational resources -- are becoming increasingly popular. However, third party inference raises critical concerns about user data privacy. To mitigate these risks, privacy researchers have developed provably secure schemes for third-party inference, such as Secure Multi-Party Computation (SMPC). However, SMPC protocols have significant computational and communication overhead, and do not scale to large models. In this work, we propose a new multi-party inference protocol, Cascade, that avoids these punitive costs by leveraging sharding in the sequence dimension to maintain privacy, trading off cryptographic privacy guarantees for increased performance and scalability. We demonstrate that Cascade is resistant to a generalization of a recent attack that is highly effective against other statistical privacy schemes, and that it is further resistant to learning-based attacks. As Cascade is orders of magnitude faster than existing schemes, our findings offer practical solutions for secure deployment of modern state-of-the-art LLMs.", "authors": ["Rahul Thomas", "Louai Zahran", "Erica Choi", "Akilesh Potti", "Micah Goldblum", "Arka Pal"], "categories": ["cs.LG", "cs.CR"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-07", "url": "https://arxiv.org/abs/2507.05228", "pdf_url": "https://arxiv.org/pdf/2507.05228v1", "arxiv_id": "2507.05228", "doi": "10.48550/arXiv.2507.05228", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2211}
{"id": "acc11b4c8d216e52fc403a24e82d5b0663ad59410c0c58a7b715eb71805f270a", "sources": ["arxiv", "semantic_scholar"], "title": "FlowSpec: Continuous Pipelined Speculative Decoding for Efficient Distributed LLM Inference", "abstract": "Distributed inference serves as a promising approach to enabling the inference of large language models (LLMs) at the network edge. It distributes the inference process to multiple devices to ensure that the LLMs can fit into the device memory. Recent pipeline-based approaches have the potential to parallelize communication and computation, which helps reduce inference latency. However, the benefit diminishes when the inference request at the network edge is sparse, where pipeline is typically at low utilization. To enable efficient distributed LLM inference at the edge, we propose \\textbf{FlowSpec}, a pipeline-parallel tree-based speculative decoding framework. FlowSpec incorporates three key mechanisms to improve decoding efficiency: 1) score-based step-wise verification prioritizes more important draft tokens to bring earlier accepted tokens; 2) efficient draft management to prune invalid tokens while maintaining correct causal relationship during verification; 3) dynamic draft expansion strategies to supply high-quality speculative inputs. These techniques work in concert to enhance both pipeline utilization and speculative efficiency. We evaluate FlowSpec on a real-world testbed with other baselines. Experimental results demonstrate that our proposed framework significantly improves inference speed across diverse models and configurations, achieving speedup ratios 1.37$\\times$-1.73$\\times$ compared to baselines. Our code is publicly available at \\href{https://github.com/Leosang-lx/FlowSpec#}{https://github.com/Leosang-lx/FlowSpec\\#}.", "authors": ["Xing Liu", "Lizhuo Luo", "Ming Tang", "Chao Huang", "Xu Chen"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-03", "url": "https://arxiv.org/abs/2507.02620", "pdf_url": "https://arxiv.org/pdf/2507.02620v3", "arxiv_id": "2507.02620", "doi": "10.48550/arXiv.2507.02620", "citation_count": 3, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Leosang-lx/FlowSpec#}{https://github.com/Leosang-lx/FlowSpec\\#}", "venue": "arXiv.org", "quality_score": 0.3347}
{"id": "03c79b7304a4a68c2e95603bf14f6f657d890ae10004856802a230949e8e5ce8", "sources": ["arxiv", "semantic_scholar"], "title": "OmniDraft: A Cross-vocabulary, Online Adaptive Drafter for On-device Speculative Decoding", "abstract": "Speculative decoding generally dictates having a small, efficient draft model that is either pretrained or distilled offline to a particular target model series, for instance, Llama or Qwen models. However, within online deployment settings, there are two major challenges: 1) usage of a target model that is incompatible with the draft model; 2) expectation of latency improvements over usage and time. In this work, we propose OmniDraft, a unified framework that enables a single draft model to operate with any target model and adapt dynamically to user data. We introduce an online n-gram cache with hybrid distillation fine-tuning to address the cross-vocabulary mismatch across draft and target models; and further improve decoding speed by leveraging adaptive drafting techniques. OmniDraft is particularly suitable for on-device LLM applications where model cost, efficiency and user customization are the major points of contention. This further highlights the need to tackle the above challenges and motivates the \\textit{``one drafter for all''} paradigm. We showcase the proficiency of the OmniDraft framework by performing online learning on math reasoning, coding and text generation tasks. Notably, OmniDraft enables a single Llama-68M model to pair with various target models including Vicuna-7B, Qwen2-7B and Llama3-8B models for speculative decoding; and additionally provides up to 1.5-2x speedup.", "authors": ["Ramchalam Kinattinkara Ramakrishnan", "Zhaocong Yuan", "Shaojie Zhuo", "Chen Feng", "Yicheng Lin", "Chenzheng Su", "Xiaopeng Zhang"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-03", "url": "https://arxiv.org/abs/2507.02659", "pdf_url": "https://arxiv.org/pdf/2507.02659v3", "arxiv_id": "2507.02659", "doi": "10.48550/arXiv.2507.02659", "citation_count": 2, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2166}
{"id": "2709f59c07981dedde83e331f291c26d02eab73f5d5018db8144488ffb26dbcd", "sources": ["arxiv", "semantic_scholar"], "title": "LogitSpec: Accelerating Retrieval-based Speculative Decoding via Next Next Token Speculation", "abstract": "Speculative decoding (SD), where a small draft model is employed to propose draft tokens in advance and then the target model validates them in parallel, has emerged as a promising technique for LLM inference acceleration. Many endeavors to improve SD are to eliminate the need for a draft model and generate draft tokens in a retrieval-based manner in order to further alleviate the drafting overhead and significantly reduce the difficulty in deployment and applications. However, retrieval-based SD relies on a matching paradigm to retrieval the most relevant reference as the draft tokens, where these methods often fail to find matched and accurate draft tokens. To address this challenge, we propose LogitSpec to effectively expand the retrieval range and find the most relevant reference as drafts. Our LogitSpec is motivated by the observation that the logit of the last token can not only predict the next token, but also speculate the next next token. Specifically, LogitSpec generates draft tokens in two steps: (1) utilizing the last logit to speculate the next next token; (2) retrieving relevant reference for both the next token and the next next token. LogitSpec is training-free and plug-and-play, which can be easily integrated into existing LLM inference frameworks. Extensive experiments on a wide range of text generation benchmarks demonstrate that LogitSpec can achieve up to 2.61 $\\times$ speedup and 3.28 mean accepted tokens per decoding step. Our code is available at https://github.com/smart-lty/LogitSpec.", "authors": ["Tianyu Liu", "Qitan Lv", "Hao Li", "Xing Gao", "Xiao Sun", "Xiaoyan Sun"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-07-02", "url": "https://arxiv.org/abs/2507.01449", "pdf_url": "https://arxiv.org/pdf/2507.01449v3", "arxiv_id": "2507.01449", "doi": "10.48550/arXiv.2507.01449", "citation_count": 9, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/smart-lty/LogitSpec", "venue": "arXiv.org", "quality_score": 0.3329}
{"id": "10de40d14a89417d36b9abe9e64486d9c28a6ddb30f0fd56675e1db16a1a3cd2", "sources": ["arxiv", "semantic_scholar"], "title": "Quantize-Sample-and-Verify: LLM Acceleration via Adaptive Edge-Cloud Speculative Decoding", "abstract": "In edge-cloud speculative decoding (SD), edge devices equipped with small language models (SLMs) generate draft tokens that are verified by large language models (LLMs) in the cloud. A key bottleneck in such systems is the limited communication bandwidth between edge and cloud, which necessitates quantization of the information transmitted about generated tokens. In this work, we introduce a novel quantize-sample (Q-S) strategy that provably preserves the output distribution of the cloud-based model, ensuring that the verified tokens match the distribution of those that would have been generated directly by the LLM. We develop a throughput model for edge-cloud SD that explicitly accounts for communication latency. Leveraging this model, we propose an adaptive mechanism that optimizes token throughput by dynamically adjusting the draft length and quantization precision in response to both semantic uncertainty and channel conditions. Simulations demonstrate that the proposed Q-S approach significantly improves decoding efficiency in realistic edge-cloud deployment scenarios.", "authors": ["Guangyi Zhang", "Yunlong Cai", "Guanding Yu", "Petar Popovski", "Osvaldo Simeone"], "categories": ["eess.SP"], "fields_of_study": ["Engineering", "Computer Science"], "published_date": "2025-07-01", "url": "https://arxiv.org/abs/2507.00605", "pdf_url": "https://arxiv.org/pdf/2507.00605v3", "arxiv_id": "2507.00605", "doi": "10.1109/LCOMM.2026.3651580", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE Communications Letters", "quality_score": 0.2143}
{"id": "2c284cbcbe0ae757f1f882c4c5f4f452c14971af867075f9d3c6e700c46a2b72", "sources": ["arxiv", "semantic_scholar"], "title": "VOCABTRIM: Vocabulary Pruning for Efficient Speculative Decoding in LLMs", "abstract": "In this paper, we introduce a simple training-free technique to improve the performance of drafter-based speculative decoding (SpD) methods that incorporates language modeling head (LM head) during drafting process. A drafter-based speculative decoding leverages one or more smaller language models, a.k.a. drafters or draft models, to sample a draft sequence or tree consisting of multiple tokens, followed by verification by a base LLM, a target model, accepting a subset as its valid generation. As it is usually considered that the speculative decoding requires one-to-one mapping between vocabularies of the target model and the draft model, it has been natural to share the vocabulary between them, or even share the LM head as in EAGLE or Medusa. We first identify that this draft token sampling scheme inherently contains an unnecessary inference overhead in drafting, especially for some target LLMs with very large vocabularies. Then, we propose a simple technique, VocabTrim, to mitigate the drafting overhead to improve the generation speed in memory-bound environment. VocabTrim reconstructs the drafter LM head to contain only a limited set of tokens, selected by the most frequently sampled from the vocabulary of the target model. While limiting the vocabulary in drafting slightly degrades the acceptance rate, it significantly reduces the drafting latency in memory-bound process which is often the case on edge devices, resulting in higher memory-bound speed up (MBSU). We show that our method can boost the memory-bound speed-up for Llama-3 models on Spec-Bench, specifically by 16% for Llama-3.2-3B-Instruct.", "authors": ["Raghavv Goel", "Sudhanshu Agrawal", "Mukul Gagrani", "Junyoung Park", "Yifan Zao", "He Zhang", "Tian Liu", "Yiping Yang", "Xin Yuan", "Jiuyan Lu", "Chris Lott", "Mingu Lee"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-28", "url": "https://arxiv.org/abs/2506.22694", "pdf_url": "https://arxiv.org/pdf/2506.22694v2", "arxiv_id": "2506.22694", "doi": "10.48550/arXiv.2506.22694", "citation_count": 9, "influential_citation_count": 4, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3495}
{"id": "e6883442625a1a239ccbfe2556c5fe1042d67d1625a9172f85b01bf7d1ccc512", "sources": ["arxiv", "semantic_scholar"], "title": "Scaling Speculative Decoding with Lookahead Reasoning", "abstract": "Reasoning models excel by generating long chain-of-thoughts, but decoding the resulting thousands of tokens is slow. Token-level speculative decoding (SD) helps, but its benefit is capped, because the chance that an entire $γ$-token guess is correct falls exponentially as $γ$ grows. This means allocating more compute for longer token drafts faces an algorithmic ceiling -- making the speedup modest and hardware-agnostic. We raise this ceiling with Lookahead Reasoning, which exploits a second, step-level layer of parallelism. Our key insight is that reasoning models generate step-by-step, and each step needs only to be semantically correct, not exact token matching. In Lookahead Reasoning, a lightweight draft model proposes several future steps; the target model expands each proposal in one batched pass, and a verifier keeps semantically correct steps while letting the target regenerate any that fail. Token-level SD still operates within each reasoning step, so the two layers of parallelism multiply. We show Lookahead Reasoning lifts the peak speedup of SD both theoretically and empirically. Across GSM8K, AIME, and other benchmarks, Lookahead Reasoning improves the speedup of SD from 1.4x to 2.1x while preserving answer quality, and its speedup scales better with additional GPU throughput. Our code is available at https://github.com/hao-ai-lab/LookaheadReasoning", "authors": ["Yichao Fu", "Rui Ge", "Zelei Shao", "Zhijie Deng", "Hao Zhang"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-24", "url": "https://arxiv.org/abs/2506.19830", "pdf_url": "https://arxiv.org/pdf/2506.19830v1", "arxiv_id": "2506.19830", "doi": "10.48550/arXiv.2506.19830", "citation_count": 10, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/hao-ai-lab/LookaheadReasoning", "venue": "arXiv.org", "quality_score": 0.3187}
{"id": "59b9ab8a4339927323ccfcfe0e162a7eb92a2d95747ee79e4d9fca1dcd8d0800", "sources": ["arxiv", "semantic_scholar"], "title": "Utility-Driven Speculative Decoding for Mixture-of-Experts", "abstract": "GPU memory bandwidth is the main bottleneck for low-latency Large Language Model (LLM) inference. Speculative decoding leverages idle GPU compute by using a lightweight drafter to propose K tokens, which the LLM verifies in parallel, boosting token throughput. In conventional dense LLMs, all model weights are fetched each iteration, so speculation adds no latency overhead. Emerging Mixture of Experts (MoE) models activate only a subset of weights per token, greatly reducing data movement. However, we show that speculation is ineffective for MoEs: draft tokens collectively activate more weights, increasing data movement and verification time by 2-3x. When token throughput gains fail to offset this overhead, speculation causes slowdowns up to 1.5x, making it infeasible. Even when useful, the optimal K varies by task, model, and even between requests and iterations. Thus, despite widespread use in dense LLMs, speculation remains impractical in leading MoEs. We present Cascade, a utility-driven framework that selectively enables speculation to avoid slowdowns and dynamically tunes K to accelerate MoE serving. Cascade uses a lightweight metric, speculation utility, the ratio of token gains to verification cost, which shows iteration-level locality, enabling periodic decisions via short test and longer set phases. For each request, Cascade disables speculation if utility drops below one during testing, and when utility exceeds one, tests multiple K-values to choose the utility-maximizing K for the set phase. We implement Cascade in vLLM and evaluate it on five popular MoEs with workloads spanning code, math, extraction, and mixed tasks. Cascade limits slowdown to 5% (vs. 1.5x) and improves throughput by 7-14% over static K, making speculative decoding practical for MoEs.", "authors": ["Anish Saxena", "Po-An Tsai", "Hritvik Taneja", "Aamer Jaleel", "Moinuddin Qureshi"], "categories": ["cs.DC", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-17", "url": "https://arxiv.org/abs/2506.20675", "pdf_url": "https://arxiv.org/pdf/2506.20675v1", "arxiv_id": "2506.20675", "doi": "10.48550/arXiv.2506.20675", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1982}
{"id": "4cee255d37b52f67629ac5e9ba942bc4a74f9e1768cf2f12d21483cfcd9e03f9", "sources": ["arxiv", "semantic_scholar"], "title": "SwiftSpec: Ultra-Low Latency LLM Decoding by Scaling Asynchronous Speculative Decoding", "abstract": "Low-latency decoding for large language models (LLMs) is crucial for applications like chatbots and code assistants, yet generating long outputs remains slow in single-query settings. Prior work on speculative decoding (which combines a small draft model with a larger target model) and tensor parallelism has each accelerated decoding. However, conventional approaches fail to apply both simultaneously due to imbalanced compute requirements (between draft and target models), KV-cache inconsistencies, and communication overheads under small-batch tensor-parallelism. This paper introduces SwiftSpec, a system that targets ultra-low latency for LLM decoding. SwiftSpec redesigns the speculative decoding pipeline in an asynchronous and disaggregated manner, so that each component can be scaled flexibly and remove draft overhead from the critical path. To realize this design, SwiftSpec proposes parallel tree generation, tree-aware KV cache management, and fused, latency-optimized kernels to overcome the challenges listed above. Across 5 model families and 6 datasets, SwiftSpec achieves an average of 1.75x speedup over state-of-the-art speculative decoding systems and, as a highlight, serves Llama3-70B at 348 tokens/s on 8 Nvidia Hopper GPUs, making it the fastest known system for low-latency LLM serving at this scale.", "authors": ["Ziyi Zhang", "Ziheng Jiang", "Chengquan Jiang", "Menghan Yu", "Size Zheng", "Haibin Lin", "Henry Hoffmann", "Xin Liu"], "categories": ["cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-12", "url": "https://arxiv.org/abs/2506.11309", "pdf_url": "https://arxiv.org/pdf/2506.11309v1", "arxiv_id": "2506.11309", "doi": "10.48550/arXiv.2506.11309", "citation_count": 9, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.25}
{"id": "62644411d7e6043fc8613a92a5d0dd6817192671126ef3a4bcc2ab1e4fed6d8d", "sources": ["arxiv", "semantic_scholar"], "title": "SLED: A Speculative LLM Decoding Framework for Efficient Edge Serving", "abstract": "The growing gap between the increasing complexity of large language models (LLMs) and the limited computational budgets of edge devices poses a key challenge for efficient on-device inference, despite gradual improvements in hardware capabilities. Existing strategies, such as aggressive quantization, pruning, or remote inference, trade accuracy for efficiency or lead to substantial cost burdens. This position paper introduces a new framework that leverages speculative decoding, previously viewed primarily as a decoding acceleration technique for autoregressive generation of LLMs, as a promising approach specifically adapted for edge computing by orchestrating computation across heterogeneous devices. We propose \\acronym, a framework that allows lightweight edge devices to draft multiple candidate tokens locally using diverse draft models, while a single, shared edge server verifies the tokens utilizing a more precise target model. To further increase the efficiency of verification, the edge server batch the diverse verification requests from devices. This approach supports device heterogeneity and reduces server-side memory footprint by sharing the same upstream target model across multiple devices. Our initial experiments with Jetson Orin Nano, Raspberry Pi 4B/5, and an edge server equipped with 4 Nvidia A100 GPUs indicate substantial benefits: 2.2 more system throughput, 2.8 more system capacity, and better cost efficiency, all without sacrificing model accuracy.", "authors": ["Xiangchen Li", "Dimitrios Spatharakis", "Saeid Ghafouri", "Jiakun Fan", "Hans Vandierendonck", "Deepu John", "Bo Ji", "Dimitrios Nikolopoulos"], "categories": ["cs.DC", "cs.AI", "cs.LG", "cs.NI"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-11", "url": "https://arxiv.org/abs/2506.09397", "pdf_url": "https://arxiv.org/pdf/2506.09397v5", "arxiv_id": "2506.09397", "doi": "10.1145/3769102.3770608", "citation_count": 14, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IFIP International Information Security Conference", "quality_score": 0.294}
{"id": "2fca7375b69324286b366eeac3a3ac3a13d246e78ff08a497ff95adf7c868c56", "sources": ["arxiv", "semantic_scholar"], "title": "Tokenized Bandit for LLM Decoding and Alignment", "abstract": "We introduce the tokenized linear bandit (TLB) and multi-armed bandit (TMAB), variants of linear and stochastic multi-armed bandit problems inspired by LLM decoding and alignment. In these problems, at each round $t \\in [T]$, a user submits a query (context), and the decision maker (DM) sequentially selects a token irrevocably from a token set. Once the sequence is complete, the DM observes a random utility from the user, whose expectation is presented by a sequence function mapping the chosen token sequence to a nonnegative real value that depends on the query. In both problems, we first show that learning is impossible without any structure on the sequence function. We introduce a natural assumption, diminishing distance with more commons (DDMC), and propose algorithms with regret $\\tilde{O}(L\\sqrt{T})$ and $\\tilde{O}(L\\sqrt{T^{2/3}})$ for TLB and TMAB, respectively. As a side product, we obtain an (almost) optimality of the greedy decoding for LLM decoding algorithm under DDMC, which justifies the unresaonable effectiveness of greedy decoding in several tasks. This also has an immediate application to decoding-time LLM alignment, when the misaligned utility can be represented as the frozen LLM's utility and a linearly realizable latent function. We finally validate our algorithm's performance empirically as well as verify our assumptions using synthetic and real-world datasets.", "authors": ["Suho Shin", "Chenghao Yang", "Haifeng Xu", "Mohammad T. Hajiaghayi"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-08", "url": "https://arxiv.org/abs/2506.07276", "pdf_url": "https://arxiv.org/pdf/2506.07276v1", "arxiv_id": "2506.07276", "doi": "10.48550/arXiv.2506.07276", "citation_count": 4, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.1879}
{"id": "d8ef199644ac8edfa91b31da024db82fdecc3a2bbf7d67ad738ac90a15c4b566", "sources": ["arxiv", "semantic_scholar"], "title": "Saffron-1: Safety Inference Scaling", "abstract": "Existing safety assurance research has primarily focused on training-phase alignment to instill safe behaviors into LLMs. However, recent studies have exposed these methods' susceptibility to diverse jailbreak attacks. Concurrently, inference scaling has significantly advanced LLM reasoning capabilities but remains unexplored in the context of safety assurance. Addressing this gap, our work pioneers inference scaling for robust and effective LLM safety against emerging threats. We reveal that conventional inference scaling techniques, despite their success in reasoning tasks, perform poorly in safety contexts, even falling short of basic approaches like Best-of-N Sampling. We attribute this inefficiency to a newly identified challenge, the exploration--efficiency dilemma, arising from the high computational overhead associated with frequent process reward model (PRM) evaluations. To overcome this dilemma, we propose SAFFRON, a novel inference scaling paradigm tailored explicitly for safety assurance. Central to our approach is the introduction of a multifurcation reward model (MRM) that significantly reduces the required number of reward model evaluations. To operationalize this paradigm, we further propose: (i) a partial supervision training objective for MRM, (ii) a conservative exploration constraint to prevent out-of-distribution explorations, and (iii) a Trie-based key--value caching strategy that facilitates cache sharing across sequences during tree search. Extensive experiments validate the effectiveness of our method. Additionally, we publicly release our trained multifurcation reward model (Saffron-1) and the accompanying token-level safety reward dataset (Safety4M) to accelerate future research in LLM safety. Our code, model, and data are publicly available at https://github.com/q-rz/saffron , and our project homepage is at https://q-rz.github.io/p/saffron .", "authors": ["Ruizhong Qiu", "Gaotang Li", "Tianxin Wei", "Jingrui He", "Hanghang Tong"], "categories": ["cs.LG", "cs.AI", "cs.CR"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-06", "url": "https://arxiv.org/abs/2506.06444", "pdf_url": "https://arxiv.org/pdf/2506.06444v2", "arxiv_id": "2506.06444", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/q-rz/saffron", "venue": null, "quality_score": 0.2194}
{"id": "4698d1054b05f5c534029fd6af09c688963224b517d759940013f6d1a5511b9e", "sources": ["arxiv", "semantic_scholar"], "title": "Guided Speculative Inference for Efficient Test-Time Alignment of LLMs", "abstract": "We propose Guided Speculative Inference (GSI), a novel algorithm for efficient reward-guided decoding in large language models. GSI combines soft best-of-$n$ test-time scaling with a reward model $r(x,y)$ and speculative samples from a small auxiliary model $π_S(y\\mid x)$. We provably approximate both the optimal tilted policy $π_{β,B}(y\\mid x) \\propto π_B(y\\mid x)\\exp(β\\,r(x,y))$ of soft best-of-$n$ under the base model $π_B$, as well as the expected reward under the optimal policy. In experiments on reasoning benchmarks (MATH500, OlympiadBench, Minerva Math, MMLU-STEM, GSM8K) and across different model families, our method achieves higher accuracy than standard soft best-of-$n$ with $π_S$ and reward-guided speculative decoding (Liao et al., 2025), and in certain settings even outperforms soft best-of-$n$ with $π_B$, while reducing end-to-end latency by up to $28\\%$. The code is available at https://github.com/j-geuter/GSI .", "authors": ["Jonathan Geuter", "Youssef Mroueh", "David Alvarez-Melis"], "categories": ["cs.LG", "stat.ML"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2025-06-04", "url": "https://arxiv.org/abs/2506.04118", "pdf_url": "https://arxiv.org/pdf/2506.04118v3", "arxiv_id": "2506.04118", "doi": "10.48550/arXiv.2506.04118", "citation_count": 10, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/j-geuter/GSI", "venue": "arXiv.org", "quality_score": 0.2833}
{"id": "fe08fc4092c711017b17a89fb27ecc129373485c19cee6465d437b3a07667550", "sources": ["arxiv", "semantic_scholar"], "title": "AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism", "abstract": "Large language models (LLMs) are increasingly used for long-content generation (e.g., long Chain-of-Thought reasoning) where decoding efficiency becomes a critical bottleneck: Autoregressive decoding is inherently limited by its sequential token generation process, where each token must be generated before the next can be processed. This sequential dependency restricts the ability to fully leverage modern hardware's parallel processing capabilities. Existing methods like speculative decoding and layer skipping offer potential speedups but have notable drawbacks: speculative decoding relies on an auxiliary \"drafter\" model, which can be challenging to acquire and increases memory overhead, while layer skipping may introduce discrepancies in the outputs due to the missing key-value cache at skipped layers. In this work, we propose AdaDecode, which accelerates LLM decoding without requiring auxiliary models or changes to the original model parameters, while ensuring output consistency. AdaDecode leverages the insight that many tokens can accurately be generated at intermediate layers, as further layers often do not significantly alter predictions once the model reaches a certain confidence. By adaptively generating tokens at intermediate layers when confidence is high, AdaDecode enables the next token's computation to begin immediately. The remaining layer computations for early-predicted tokens are deferred and executed in parallel with subsequent tokens when needed, maximizing hardware utilization and reducing decoding latency. A final verification step ensures that early predictions match the results of standard autoregressive decoding, preserving output parity. Experiments across diverse generation tasks shows that AdaDecode consistently achieves superior decoding throughput with up to 1.73x speedup, while guaranteeing output parity with standard autoregressive decoding.", "authors": ["Zhepei Wei", "Wei-Lin Chen", "Xinyu Zhu", "Yu Meng"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-04", "url": "https://arxiv.org/abs/2506.03700", "pdf_url": "https://arxiv.org/pdf/2506.03700v1", "arxiv_id": "2506.03700", "doi": "10.48550/arXiv.2506.03700", "citation_count": 4, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/weizhepei/AdaDecode", "venue": "International Conference on Machine Learning", "quality_score": 0.2833}
{"id": "5cdbafd441b0b061c80b1cb46dd0e6430083407ed0a465b789cb1acc7f40b542", "sources": ["arxiv", "semantic_scholar"], "title": "POSS: Position Specialist Generates Better Draft for Speculative Decoding", "abstract": "Speculative decoding accelerates Large Language Model (LLM) inference by using a small draft model to predict multiple tokens, and a large target model to verify these tokens in parallel. Recent studies leverage the hidden state of the target model to enhance draft model prediction accuracy. However, existing methods suffer from the degrading quality of draft token predictions at later positions, due to error accumulation in draft model generated features. In this paper, we propose Position Specialists (PosS), which consist of multiple position-specialized draft layers to generate tokens at assigned position(s). Position specialists greatly improve token acceptance rate at later positions per drafting round, as each specialist only needs to focus on handling a certain level of draft model feature deviation. Experiment results on Llama-3-8B-Instruct and Llama-2-13B-chat across six datasets demonstrate that PosS effectively improves over baselines on average acceptance length and speed-up ratio. Our codebase is available at https://github.com/shrango/PosS.", "authors": ["Langlin Huang", "Chengsong Huang", "Jixuan Leng", "Di Huang", "Jiaxin Huang"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-04", "url": "https://arxiv.org/abs/2506.03566", "pdf_url": "https://arxiv.org/pdf/2506.03566v1", "arxiv_id": "2506.03566", "doi": "10.48550/arXiv.2506.03566", "citation_count": 2, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/shrango/PosS", "venue": "arXiv.org", "quality_score": 0.2833}
{"id": "4d5c930042067467b583f610a26655d7dadaf0962c6ce491f61a85f0fad218d5", "sources": ["arxiv", "semantic_scholar"], "title": "EfficientEdit: Accelerating Code Editing via Edit-Oriented Speculative Decoding", "abstract": "Large Language Models (LLMs) have demonstrated remarkable capabilities in code editing, substantially enhancing software development productivity. However, the inherent complexity of code editing tasks forces existing approaches to rely on LLMs' autoregressive end-to-end generation, where decoding speed plays a critical role in efficiency. While inference acceleration techniques like speculative decoding are applied to improve the decoding efficiency, these methods fail to account for the unique characteristics of code editing tasks where changes are typically localized and existing code segments are reused. To address this limitation, we propose EfficientEdit, a novel method that improves LLM-based code editing efficiency through two key mechanisms based on speculative decoding: (1) effective reuse of original code segments while identifying potential edit locations, and (2) efficient generate edit content via high-quality drafts from edit-oriented draft models and a dynamic verification mechanism that balances quality and acceleration. Experimental results show that EfficientEdit can achieve up to 10.38$\\times$ and 13.09$\\times$ speedup compared to standard autoregressive decoding in CanItEdit and CodeIF-Bench, respectively, outperforming state-of-the-art inference acceleration approaches by up to 90.6%. The code and data are available at https://github.com/zhu-zhu-ding/EfficientEdit.", "authors": ["Peiding Wang", "Li Zhang", "Fang Liu", "Yinghao Zhu", "Wang Xu", "Lin Shi", "Xiaoli Lian", "Minxiao Li", "Bo Shen", "An Fu"], "categories": ["cs.SE"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-03", "url": "https://arxiv.org/abs/2506.02780", "pdf_url": "https://arxiv.org/pdf/2506.02780v2", "arxiv_id": "2506.02780", "doi": "10.1109/ASE63991.2025.00215", "citation_count": 3, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/zhu-zhu-ding/EfficientEdit", "venue": "International Conference on Automated Software Engineering", "quality_score": 0.2816}
{"id": "da97645368cb2c9e6d05c6ecbaa7af3bd7dd6250ec14fa59144fb9fb14f6fba3", "sources": ["arxiv", "semantic_scholar"], "title": "Out-of-Vocabulary Sampling Boosts Speculative Decoding", "abstract": "Speculative decoding relies on fast and accurate drafters. Recent state-of-the-art language models employ larger and larger vocabularies, which significantly slows down drafters. One promising approach to boost the efficiency of speculative decoding is to use drafters with smaller vocabularies. However, existing sampling methods cannot draw out-of-vocabulary tokens, creating a tradeoff between drafters' vocabulary size and acceptance rates. This paper introduces Redistributing Drafter Kernels (RDK), the first out-of-vocabulary sampler that effectively recovers acceptance rates by virtually restoring pruned target tokens. RDK leverages token-affinity priors to reallocate drafter mass towards high-overlap regions. We prove mathematically that RDK can achieve higher acceptance rates than vanilla and state-of-the-art samplers. We provide an efficient first-order approximation of RDK and prove that it reduces redistribution times from $O(N^2)$ to $O(N)$, enabling lightweight implementations for large vocabularies. Our experiments demonstrate that this linear-time RDK significantly boosts acceptance rates even after extreme pruning (removing more than 75% of the drafter's vocabulary), where existing samplers fail. RDK opens the door to extremely pruned drafters, which were previously impractical.", "authors": ["Nadav Timor", "Jonathan Mamou", "Oren Pereg", "Hongyang Zhang", "David Harel"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-02", "url": "https://arxiv.org/abs/2506.03206", "pdf_url": "https://arxiv.org/pdf/2506.03206v1", "arxiv_id": "2506.03206", "doi": "10.48550/arXiv.2506.03206", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.181}
{"id": "874e52828838f468316cc8e891a356c6678045d5a9b238734eee6674c6a960db", "sources": ["arxiv", "semantic_scholar"], "title": "Mamba Drafters for Speculative Decoding", "abstract": "Speculative decoding has emerged as a promising approach to accelerating large language model (LLM) generation using a fast drafter while maintaining alignment with the target model's distribution. However, existing approaches face a trade-off: external drafters offer flexibility but can suffer from slower drafting, while self-speculation methods use drafters tailored to the target model but require re-training. In this paper, we introduce novel drafters based on Mamba, a state-of-the-art state space model (SSM), as a solution that combines the best aspects of both approaches. By leveraging the linear structure of SSMs, our approach avoids the quadratic complexity inherent in traditional Transformer-based methods, enabling faster drafting and lower memory usage while maintaining the flexibility to work across different target models. We further enhance efficiency with a novel test-time tree search algorithm for generating high-quality draft candidates. Our empirical evaluation demonstrates that Mamba-based drafters not only outperform existing external drafting methods but are also comparable to state-of-the-art self-speculation approaches while using less memory and maintaining their cross-model adaptability.", "authors": ["Daewon Choi", "Seunghyuk Oh", "Saket Dingliwal", "Jihoon Tack", "Kyuyoung Kim", "Woomin Song", "Seojin Kim", "Insu Han", "Jinwoo Shin", "Aram Galstyan", "Shubham Katiyar", "Sravan Babu Bodapati"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-06-01", "url": "https://arxiv.org/abs/2506.01206", "pdf_url": "https://arxiv.org/pdf/2506.01206v1", "arxiv_id": "2506.01206", "doi": "10.48550/arXiv.2506.01206", "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.1799}
{"id": "71084d20e540158f60ef7789548939cd9ad6b697d7471784241443e8f6482eb7", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Reward Model Boosts Decision Making Ability of LLMs Cost-Effectively", "abstract": "Effective decision-making in Large Language Models (LLMs) is essential for handling intricate tasks. However, existing approaches prioritize performance but often overlook the balance between effectiveness and computational cost. To address this, we first introduce the 3E Criteria to systematically assess the cost-effectiveness of search strategies, revealing that existing methods often trade significant efficiency for marginal performance gains. To improve LLM decision-making while maintaining efficiency, we propose the Speculative Reward Model (SRM), a plug-and-play framework that seamlessly integrates with existing search strategies. Specifically, SRM employs an external reward assigner to predict optimal actions, reducing reliance on LLMs' internal self-evaluation. And a speculative verification mechanism is used to prune suboptimal choices and guide the search toward more promising steps. We evaluate SRM on several complex decision-making tasks including mathematical reasoning, planning and numerical reasoning in specialized domains. Experimental results show that SRM reduces costs to 1/10 of the original search framework on average while maintaining effectiveness.", "authors": ["Jiawei Gu", "Shangsong Liang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-31", "url": "https://arxiv.org/abs/2506.00396", "pdf_url": "https://arxiv.org/pdf/2506.00396v1", "arxiv_id": "2506.00396", "doi": "10.48550/arXiv.2506.00396", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.1788}
{"id": "74bcc2ccd109bbc84516624ed182d6ba9ba5baeb7bca4ec8a5a9cb01e88bb445", "sources": ["arxiv", "semantic_scholar"], "title": "Cross-Attention Speculative Decoding", "abstract": "Speculative decoding (SD) is a widely adopted approach for accelerating inference in large language models (LLMs), particularly when the draft and target models are well aligned. However, state-of-the-art SD methods typically rely on tightly coupled, self-attention-based Transformer decoders, often augmented with auxiliary pooling or fusion layers. This coupling makes them increasingly complex and harder to generalize across different models. We present Budget EAGLE (Beagle), the first, to our knowledge, cross-attention-based Transformer decoder SD model that achieves performance on par with leading self-attention SD models (EAGLE-v2) while eliminating the need for pooling or auxiliary components, simplifying the architecture, improving training efficiency, and maintaining stable memory usage during training-time simulation. To enable effective training of this novel architecture, we propose Two-Stage Block-Attention Training, a new method that achieves training stability and convergence efficiency in block-level attention scenarios. Extensive experiments across multiple LLMs and datasets show that Beagle achieves competitive inference speedups and higher training efficiency than EAGLE-v2, offering a strong alternative for architectures in speculative decoding.", "authors": ["Wei Zhong", "Manasa Bharadwaj", "Yixiao Wang", "Yipeng Ji", "Chul Lee"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-30", "url": "https://arxiv.org/abs/2505.24544", "pdf_url": "https://arxiv.org/pdf/2505.24544v4", "arxiv_id": "2505.24544", "doi": "10.48550/arXiv.2505.24544", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1776}
{"id": "2ff89627942428ce4fc33b1f6eb6811df97ed2ea05c84693bb23d5a8f387c873", "sources": ["arxiv", "semantic_scholar"], "title": "CLaSp: In-Context Layer Skip for Self-Speculative Decoding", "abstract": "Speculative decoding (SD) is a promising method for accelerating the decoding process of Large Language Models (LLMs). The efficiency of SD primarily hinges on the consistency between the draft model and the verify model. However, existing drafting approaches typically require additional modules to be trained, which can be challenging to implement and ensure compatibility across various LLMs. In this paper, we propose CLaSp, an in-context layer-skipping strategy for self-speculative decoding. Unlike prior methods, CLaSp does not require additional drafting modules or extra training. Instead, it employs a plug-and-play mechanism by skipping intermediate layers of the verify model to construct a compressed draft model. Specifically, we develop a dynamic programming algorithm that optimizes the layer-skipping process by leveraging the complete hidden states from the last verification stage as an objective. This enables CLaSp to dynamically adjust its layer-skipping strategy after each verification stage, without relying on pre-optimized sets of skipped layers. Experimental results across diverse downstream tasks demonstrate that CLaSp achieves a speedup of 1.3x ~ 1.7x on LLaMA3 series models without altering the original distribution of the generated text.", "authors": ["Longze Chen", "Renke Shan", "Huiming Wang", "Lu Wang", "Ziqiang Liu", "Run Luo", "Jiawei Wang", "Hamid Alinejad-Rokny", "Min Yang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-30", "url": "https://arxiv.org/abs/2505.24196", "pdf_url": "https://arxiv.org/pdf/2505.24196v1", "arxiv_id": "2505.24196", "doi": "10.48550/arXiv.2505.24196", "citation_count": 7, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.2386}
{"id": "9f29098256d9ca7169695ee9ceaaa861a3f9309f7b35d57d5c3c25e8d30d8cd3", "sources": ["arxiv", "semantic_scholar"], "title": "Ghidorah: Fast LLM Inference on Edge with Speculative Decoding and Hetero-Core Parallelism", "abstract": "In-situ LLM inference on end-user devices has gained significant interest due to its privacy benefits and reduced dependency on external infrastructure. However, as the decoding process is memory-bandwidth-bound, the diverse processing units in modern end-user devices cannot be fully exploited, resulting in slow LLM inference. This paper presents Ghidorah, a LLM inference system for end-user devices with the unified memory architecture. The key idea of Ghidorah can be summarized in two steps: 1) leveraging speculative decoding approaches to enhance parallelism, and 2) ingeniously distributing workloads across multiple heterogeneous processing units to maximize computing power utilization. Ghidorah includes the hetero-core model parallelism (HCMP) architecture and the architecture-aware profiling (ARCA) approach. The HCMP architecture guides partitioning by leveraging the unified memory design of end-user devices and adapting to the hybrid computational demands of speculative decoding. The ARCA approach is used to determine the optimal speculative strategy and partitioning strategy, balancing acceptance rate with parallel capability to maximize the speedup. Additionally, we optimize sparse computation on ARM CPUs. Experimental results show that Ghidorah can achieve up to 7.6x speedup in the dominant LLM decoding phase compared to the sequential decoding approach in NVIDIA Jetson NX.", "authors": ["Jinhui Wei", "Ye Huang", "Yuhui Zhou", "Jiazhi Jiang", "Jiangsu Du", "Yutong Lu"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-29", "url": "https://arxiv.org/abs/2505.23219", "pdf_url": "https://arxiv.org/pdf/2505.23219v2", "arxiv_id": "2505.23219", "doi": "10.1109/ICCD65941.2025.00043", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "ICCD", "quality_score": 0.1765}
{"id": "1196216932e701c81a5f2fd544c7a856a00633534f51c6722f91decec81b03a5", "sources": ["arxiv", "semantic_scholar"], "title": "DINGO: Constrained Inference for Diffusion LLMs", "abstract": "Diffusion LLMs have emerged as a promising alternative to conventional autoregressive LLMs, offering significant potential for improved runtime efficiency. However, existing diffusion models lack the ability to provably enforce user-specified formal constraints, such as regular expressions, which makes them unreliable for tasks that require structured outputs, such as fixed-schema JSON generation. Unlike autoregressive models that generate tokens sequentially, diffusion LLMs predict a block of tokens in parallel. This parallelism makes traditional constrained decoding algorithms, which are designed for sequential token prediction, ineffective at preserving the true output distribution. To address this limitation, we propose DINGO, a dynamic programming-based constrained decoding strategy that is both efficient and provably distribution-preserving. DINGO enables sampling of output strings with the highest probability under the model's predicted distribution, while strictly satisfying any user-specified regular expression. On standard symbolic math and JSON generation benchmarks, DINGO achieves up to a 68 percentage point improvement over unconstrained inference", "authors": ["Tarun Suresh", "Debangshu Banerjee", "Shubham Ugare", "Sasa Misailovic", "Gagandeep Singh"], "categories": ["cs.LG", "cs.PL", "cs.SE"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-29", "url": "https://arxiv.org/abs/2505.23061", "pdf_url": "https://arxiv.org/pdf/2505.23061v1", "arxiv_id": "2505.23061", "doi": "10.48550/arXiv.2505.23061", "citation_count": 8, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2386}
{"id": "8da9a8cdfb15682fbc485755ae1b5428f504050c6ced422164b531f5663a7fad", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding Meets Quantization: Compatibility Evaluation and Hierarchical Framework Design", "abstract": "Speculative decoding and quantization effectively accelerate memory-bound inference of large language models. Speculative decoding mitigates the memory bandwidth bottleneck by verifying multiple tokens within a single forward pass, which increases computational effort. Quantization achieves this optimization by compressing weights and activations into lower bit-widths and also reduces computations via low-bit matrix multiplications. To further leverage their strengths, we investigate the integration of these two techniques. Surprisingly, experiments applying the advanced speculative decoding method EAGLE-2 to various quantized models reveal that the memory benefits from 4-bit weight quantization are diminished by the computational load from speculative decoding. Specifically, verifying a tree-style draft incurs significantly more time overhead than a single-token forward pass on 4-bit weight quantized models. This finding led to our new speculative decoding design: a hierarchical framework that employs a small model as an intermediate stage to turn tree-style drafts into sequence drafts, leveraging the memory access benefits of the target quantized model. Experimental results show that our hierarchical approach achieves a 2.78$\\times$ speedup across various tasks for the 4-bit weight Llama-3-70B model on an A100 GPU, outperforming EAGLE-2 by 1.31$\\times$. Code available at https://github.com/AI9Stars/SpecMQuant.", "authors": ["Yudi Zhang", "Weilin Zhao", "Xu Han", "Tiejun Zhao", "Wang Xu", "Hailong Cao", "Conghui Zhu"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-28", "url": "https://arxiv.org/abs/2505.22179", "pdf_url": "https://arxiv.org/pdf/2505.22179v2", "arxiv_id": "2505.22179", "doi": "10.48550/arXiv.2505.22179", "citation_count": 2, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/AI9Stars/SpecMQuant", "venue": "arXiv.org", "quality_score": 0.2709}
{"id": "d0c45ed4e43af45f82f41a0619860bc541db51b4fc50c61709407786c315229b", "sources": ["arxiv", "semantic_scholar"], "title": "Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding", "abstract": "Diffusion-based large language models (Diffusion LLMs) have shown promise for non-autoregressive text generation with parallel decoding capabilities. However, the practical inference speed of open-sourced Diffusion LLMs often lags behind autoregressive models due to the lack of Key-Value (KV) Cache and quality degradation when decoding multiple tokens simultaneously. To bridge this gap, we introduce a novel block-wise approximate KV Cache mechanism tailored for bidirectional diffusion models, enabling cache reuse with negligible performance drop. Additionally, we identify the root cause of generation quality degradation in parallel decoding as the disruption of token dependencies under the conditional independence assumption. To address this, we propose a confidence-aware parallel decoding strategy that selectively decodes tokens exceeding a confidence threshold, mitigating dependency violations and maintaining generation quality. Experimental results on LLaDA and Dream models across multiple LLM benchmarks demonstrate up to \\textbf{27.6$\\times$ throughput} improvement with minimal accuracy loss, closing the performance gap with autoregressive models and paving the way for practical deployment of Diffusion LLMs.", "authors": ["Chengyue Wu", "Hao Zhang", "Shuchen Xue", "Zhijian Liu", "Shizhe Diao", "Ligeng Zhu", "Ping Luo", "Song Han", "Enze Xie"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-28", "url": "https://arxiv.org/abs/2505.22618", "pdf_url": "https://arxiv.org/pdf/2505.22618v3", "arxiv_id": "2505.22618", "doi": "10.48550/arXiv.2505.22618", "citation_count": 311, "influential_citation_count": 82, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.9595}
{"id": "d8697f3996050709469d1ec0dd35c9c83d5e75ef77b2a0603c2c606bf8b3dbb9", "sources": ["arxiv", "semantic_scholar"], "title": "SpecExtend: A Drop-in Enhancement for Speculative Decoding of Long Sequences", "abstract": "Speculative decoding is a widely used technique for accelerating inference in large language models (LLMs), but its performance degrades as input length grows, with significant drops even at moderate lengths. Yet, this early degradation has remained largely underexplored. We introduce SpecExtend, a drop-in enhancement that improves speculative decoding on long sequences without additional training. SpecExtend integrates efficient attention mechanisms such as FlashAttention and Hybrid Tree Attention to accelerate prefill and verification steps. To improve both draft accuracy and speed on long inputs without retraining, we propose Cross-model Retrieval, a novel KV cache eviction strategy that leverages the target model's attention scores to dynamically select relevant context for the smaller draft model. Extensive evaluations show that SpecExtend accelerates speculative decoding by up to 2.84x on 16K-token long document summarization and up to 3.86x on long-form reasoning, while preserving the short-input performance of state-of-the-art frameworks. Our code is available at https://github.com/jycha98/SpecExtend .", "authors": ["Jungyoub Cha", "Hyunjong Kim", "Sungzoon Cho"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-27", "url": "https://arxiv.org/abs/2505.20776", "pdf_url": "https://arxiv.org/pdf/2505.20776v4", "arxiv_id": "2505.20776", "doi": "10.48550/arXiv.2505.20776", "citation_count": 2, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/jycha98/SpecExtend", "venue": "arXiv.org", "quality_score": 0.2692}
{"id": "625fbff4f44ed7c5296ae2f6873b21382ecac6a5adadedcdf50da5e70e39a239", "sources": ["arxiv", "semantic_scholar"], "title": "Fast and Cost-effective Speculative Edge-Cloud Decoding with Early Exits", "abstract": "Large Language Models (LLMs) enable various applications on edge devices such as smartphones, wearables, and embodied robots. However, their deployment often depends on expensive cloud-based APIs, creating high operational costs, which limit access for smaller organizations and raise sustainability concerns. Certain LLMs can be deployed on-device, offering a cost-effective solution with reduced latency and improved privacy. Yet, limited computing resources constrain the size and accuracy of models that can be deployed, necessitating a collaborative design between edge and cloud. We propose a fast and cost-effective speculative edge-cloud decoding framework with a large target model on the server and a small draft model on the device. By introducing early exits in the target model, tokens are generated mid-verification, allowing the client to preemptively draft subsequent tokens before final verification, thus utilizing idle time and enhancing parallelism between edge and cloud. Using an NVIDIA Jetson Nano (client) and an A100 GPU (server) with Vicuna-68M (draft) and Llama2-7B (target) models, our method achieves up to a 35% reduction in latency compared to cloud-based autoregressive decoding, with an additional 11% improvement from preemptive drafting. To demonstrate real-world applicability, we deploy our method on the Unitree Go2 quadruped robot using Vision-Language Model (VLM) based control, achieving a 21% speedup over traditional cloud-based autoregressive decoding. These results demonstrate the potential of our framework for real-time LLM and VLM applications on resource-constrained edge devices.", "authors": ["Yeshwanth Venkatesha", "Souvik Kundu", "Priyadarshini Panda"], "categories": ["cs.RO", "cs.AI", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-27", "url": "https://arxiv.org/abs/2505.21594", "pdf_url": "https://arxiv.org/pdf/2505.21594v1", "arxiv_id": "2505.21594", "doi": "10.48550/arXiv.2505.21594", "citation_count": 11, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.2698}
{"id": "3a0a51d7fa8eb3b28c631ee8cd0731056da881c7a554f0f9ac49196d788974e7", "sources": ["arxiv", "semantic_scholar"], "title": "HAMburger: Accelerating LLM Inference via Token Smashing", "abstract": "The growing demand for efficient Large Language Model (LLM) inference requires a holistic optimization on algorithms, systems, and hardware. However, very few works have fundamentally changed the generation pattern: each token needs one forward pass and one KV cache. This can be sub-optimal because we found that LLMs are extremely capable of self-identifying the exact dose of information that a single KV cache can store, and many tokens can be generated confidently without global context. Based on this insight, we introduce HAMburger, a Hierarchically Auto-regressive Model that redefines resource allocation in LLMs by moving beyond uniform computation and storage per token during inference. Stacking a compositional embedder and a micro-step decoder in between a base LLM, HAMburger smashes multiple tokens into a single KV and generates several tokens per step. Additionally, HAMburger functions as a speculative decoding framework where it can blindly trust self-drafted tokens. As a result, HAMburger shifts the growth of KV cache and forward FLOPs from linear to sub-linear with respect to output length, and adjusts its inference speed based on query perplexity and output structure. Extensive evaluations show that HAMburger reduces the KV cache computation by up to 2$\\times$ and achieves up to 2$\\times$ TPS, while maintaining quality in both short- and long-context tasks. Our method explores an extremely challenging inference regime that requires both computation- and memory-efficiency with a hardware-agnostic design.", "authors": ["Jingyu Liu", "Ce Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-26", "url": "https://arxiv.org/abs/2505.20438", "pdf_url": "https://arxiv.org/pdf/2505.20438v1", "arxiv_id": "2505.20438", "doi": "10.48550/arXiv.2505.20438", "citation_count": 6, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2113}
{"id": "6afdd01eb8661345d865259684968da6504b0bb13a0e7c61598822dad82a788e", "sources": ["arxiv", "semantic_scholar"], "title": "MoESD: Unveil Speculative Decoding's Potential for Accelerating Sparse MoE", "abstract": "Large Language Models (LLMs) have achieved remarkable success across many applications, with Mixture of Experts (MoE) models demonstrating great potential. Compared to traditional dense models, MoEs achieve better performance with less computation. Speculative decoding (SD) is a widely used technique to accelerate LLM inference without accuracy loss, but it has been considered efficient only for dense models. In this work, we first demonstrate that, under medium batch sizes, MoE surprisingly benefits more from SD than dense models. Furthermore, as MoE becomes sparser -- the prevailing trend in MoE designs -- the batch size range where SD acceleration is expected to be effective becomes broader. To quantitatively understand tradeoffs involved in SD, we develop a reliable modeling based on theoretical analyses. While current SD research primarily focuses on improving acceptance rates of algorithms, changes in workload and model architecture can still lead to degraded SD acceleration even with high acceptance rates. To address this limitation, we introduce a new metric 'target efficiency' that characterizes these effects, thus helping researchers identify system bottlenecks and understand SD acceleration more comprehensively. For scenarios like private serving, this work unveils a new perspective to speed up MoE inference, where existing solutions struggle. Experiments on different GPUs show up to 2.29x speedup for Qwen2-57B-A14B at medium batch sizes and validate our theoretical predictions.", "authors": ["Zongle Huang", "Lei Zhu", "Zongyuan Zhan", "Ting Hu", "Weikai Mao", "Xianzhi Yu", "Yongpan Liu", "Tianyu Zhang"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-26", "url": "https://arxiv.org/abs/2505.19645", "pdf_url": "https://arxiv.org/pdf/2505.19645v4", "arxiv_id": "2505.19645", "doi": "10.48550/arXiv.2505.19645", "citation_count": 9, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.25}
{"id": "5b215da05d0a4327df82fb95463fbd3b4684e88b69d41f90d9a66334247cd2ae", "sources": ["arxiv", "semantic_scholar"], "title": "DREAM: Drafting with Refined Target Features and Entropy-Adaptive Cross-Attention Fusion for Multimodal Speculative Decoding", "abstract": "Speculative decoding (SD) has emerged as a powerful method for accelerating autoregressive generation in large language models (LLMs), yet its integration into vision-language models (VLMs) remains underexplored. We introduce DREAM, a novel speculative decoding framework tailored for VLMs that combines three key innovations: (1) a cross-attention-based mechanism to inject intermediate features from the target model into the draft model for improved alignment, (2) adaptive intermediate feature selection based on attention entropy to guide efficient draft model training, and (3) visual token compression to reduce draft model latency. DREAM enables efficient, accurate, and parallel multimodal decoding with significant throughput improvement. Experiments across a diverse set of recent popular VLMs, including LLaVA, Pixtral, SmolVLM and Gemma3, demonstrate up to 3.6x speedup over conventional decoding and significantly outperform prior SD baselines in both inference throughput and speculative draft acceptance length across a broad range of multimodal benchmarks. The code is publicly available at: https://github.com/SAI-Lab-NYU/DREAM.git", "authors": ["Yunhai Hu", "Tianhua Xia", "Zining Liu", "Rahul Raman", "Xingyu Liu", "Bo Bao", "Eric Sather", "Vithursan Thangarasa", "Sai Qian Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-25", "url": "https://arxiv.org/abs/2505.19201", "pdf_url": "https://arxiv.org/pdf/2505.19201v3", "arxiv_id": "2505.19201", "doi": "10.48550/arXiv.2505.19201", "citation_count": 8, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/SAI-Lab-NYU/DREAM.git", "venue": "arXiv.org", "quality_score": 0.2656}
{"id": "febcbb40753457f6a03531857e015cf0b2c86b1cc4081e40c58036c049225ef4", "sources": ["arxiv", "semantic_scholar"], "title": "Think Before You Accept: Semantic Reflective Verification for Faster Speculative Decoding", "abstract": "Large language models (LLMs) suffer from high inference latency due to the auto-regressive decoding process. Speculative decoding accelerates inference by generating multiple draft tokens using a lightweight model and verifying them in parallel. However, existing verification methods rely heavily on distributional consistency while overlooking semantic correctness, thereby limiting the potential speedup of speculative decoding. While some methods employ additional models for relaxed verification of draft tokens, they often fail to generalize effectively to more diverse or open-domain settings. In this work, we propose Reflective Verification, a training-free and semantics-aware approach that achieves a better trade-off between correctness and efficiency. Specifically, we leverage the inherent reflective capacity of LLMs to semantically assess the correctness of draft tokens in parallel during verification. Using prompt-based probing, we obtain both the original and reflective distributions of draft tokens in a single forward pass. The fusion of these distributions enables semantic-level verification of draft tokens that incorporates both consistency and correctness. Experiments across multiple domain benchmarks and model scales demonstrate that our method significantly increases the acceptance length of draft tokens without compromising model performance. Furthermore, we find that the proposed Reflective Verification is orthogonal to existing statistical verification methods, and their combination yields additional 5$\\sim$15\\% improvements in decoding speed.", "authors": ["Yixuan Wang", "Yijun Liu", "Shiyu ji", "Yuzhuang Xu", "Yang Xu", "Qingfu Zhu", "Wanxiang Che"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-24", "url": "https://arxiv.org/abs/2505.18629", "pdf_url": "https://arxiv.org/pdf/2505.18629v1", "arxiv_id": "2505.18629", "doi": "10.48550/arXiv.2505.18629", "citation_count": 5, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1945}
{"id": "56268a8b2bb2defcd8a6f5a472dc49a1b25baf01282358ad82f808106bcf39fa", "sources": ["arxiv", "semantic_scholar"], "title": "A Survey of LLM $\\times$ DATA", "abstract": "The integration of large language model (LLM) and data management (DATA) is rapidly redefining both domains. In this survey, we comprehensively review the bidirectional relationships. On the one hand, DATA4LLM, spanning large-scale data processing, storage, and serving, feeds LLMs with high quality, diversity, and timeliness of data required for stages like pre-training, post-training, retrieval-augmented generation, and agentic workflows: (i) Data processing for LLMs includes scalable acquisition, deduplication, filtering, selection, domain mixing, and synthetic augmentation; (ii) Data Storage for LLMs focuses on efficient data and model formats, distributed and heterogeneous storage hierarchies, KV-cache management, and fault-tolerant checkpointing; (iii) Data serving for LLMs tackles challenges in RAG (e.g., knowledge post-processing), LLM inference (e.g., prompt compression, data provenance), and training strategies (e.g., data packing and shuffling). On the other hand, in LLM4DATA, LLMs are emerging as general-purpose engines for data management. We review recent advances in (i) data manipulation, including automatic data cleaning, integration, discovery; (ii) data analysis, covering reasoning over structured, semi-structured, and unstructured data, and (iii) system optimization (e.g., configuration tuning, query rewriting, anomaly diagnosis), powered by LLM techniques like retrieval-augmented prompting, task-specialized fine-tuning, and multi-agent collaboration.", "authors": ["Xuanhe Zhou", "Junxuan He", "Wei Zhou", "Haodong Chen", "Zirui Tang", "Haoyu Zhao", "Xin Tong", "Guoliang Li", "Youmin Chen", "Jun Zhou", "Zhaojun Sun", "Binyuan Hui", "Shuo Wang", "Conghui He", "Zhiyuan Liu", "Jingren Zhou", "Fan Wu"], "categories": ["cs.DB", "cs.AI", "cs.CL", "cs.IR", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-24", "url": "https://arxiv.org/abs/2505.18458", "pdf_url": "https://arxiv.org/pdf/2505.18458v3", "arxiv_id": "2505.18458", "doi": null, "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/weAIDB/awesome-data-llm", "venue": null, "quality_score": 0.2018}
{"id": "6b408b0f7041c30a9c4b35fed8034860dede9999c4702ca30d3c6a7036adee59", "sources": ["arxiv", "semantic_scholar"], "title": "RoleRAG: Enhancing LLM Role-Playing via Graph Guided Retrieval", "abstract": "Large Language Models (LLMs) have shown promise in character imitation, enabling immersive and engaging conversations. However, they often generate content that is irrelevant or inconsistent with a character's background. We attribute these failures to: (1) the inability to accurately recall character-specific knowledge due to entity ambiguity, and (2) a lack of awareness of the character's cognitive boundaries. To address these issues, we propose RoleRAG, a retrieval-based framework that integrates efficient entity disambiguation for knowledge indexing with a boundary-aware retriever for extracting contextually appropriate information from a structured knowledge graph. Experiments on role-playing benchmarks show that RoleRAG's calibrated retrieval helps both general-purpose and role-specific LLMs better align with character knowledge and reduce hallucinated responses.", "authors": ["Yongjie Wang", "Jonathan Leung", "Zhiqi Shen"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-24", "url": "https://arxiv.org/abs/2505.18541", "pdf_url": "https://arxiv.org/pdf/2505.18541v1", "arxiv_id": "2505.18541", "doi": "10.48550/arXiv.2505.18541", "citation_count": 8, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2386}
{"id": "9cc1c5c242da58e18e22c66688df5aa9cca6e8699efecd57d0411b8bf8b8f012", "sources": ["arxiv", "semantic_scholar"], "title": "An Attack to Break Permutation-Based Private Third-Party Inference Schemes for LLMs", "abstract": "Recent advances in Large Language Models (LLMs) have led to the widespread adoption of third-party inference services, raising critical privacy concerns. Existing methods of performing private third-party inference, such as Secure Multiparty Computation (SMPC), often rely on cryptographic methods. However, these methods are thousands of times slower than standard unencrypted inference, and fail to scale to large modern LLMs. Therefore, recent lines of work have explored the replacement of expensive encrypted nonlinear computations in SMPC with statistical obfuscation methods - in particular, revealing permuted hidden states to the third parties, with accompanying strong claims of the difficulty of reversal into the unpermuted states. In this work, we begin by introducing a novel reconstruction technique that can recover original prompts from hidden states with nearly perfect accuracy across multiple state-of-the-art LLMs. We then show that extensions of our attack are nearly perfectly effective in reversing permuted hidden states of LLMs, demonstrating the insecurity of three recently proposed privacy schemes. We further dissect the shortcomings of prior theoretical `proofs' of permuation security which allow our attack to succeed. Our findings highlight the importance of rigorous security analysis in privacy-preserving LLM inference.", "authors": ["Rahul Thomas", "Louai Zahran", "Erica Choi", "Akilesh Potti", "Micah Goldblum", "Arka Pal"], "categories": ["cs.CR", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-23", "url": "https://arxiv.org/abs/2505.18332", "pdf_url": "https://arxiv.org/pdf/2505.18332v1", "arxiv_id": "2505.18332", "doi": "10.48550/arXiv.2505.18332", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1696}
{"id": "dba740c4fbc963adc8ce65e393c7a45636fa2f85b5eb4c59b17d93c9b454c287", "sources": ["arxiv", "semantic_scholar"], "title": "Reinforcement Speculative Decoding for Fast Ranking", "abstract": "Large Language Models (LLMs) have been widely adopted in ranking systems such as information retrieval (IR) systems and recommender systems (RSs). To alleviate the latency of auto-regressive decoding, some studies explore the single (first) token decoding for ranking approximation, but they suffer from severe degradation in tail positions. Although speculative decoding (SD) methods can be a remedy with verification at different positions, they face challenges in ranking systems due to their left-to-right decoding paradigm. Firstly, ranking systems require strict latency constraints, but verification rounds in SD methods remain agnostic; Secondly, SD methods usually discard listwise ranking knowledge about unaccepted items in previous rounds, hindering future multi-token prediction, especially when candidate tokens are the unaccepted items. In this paper, we propose a Reinforcement Speculative Decoding method for fast ranking inference of LLMs. To meet the ranking systems' latency requirement, we propose an up-to-down decoding paradigm that employs an agent to iteratively modify the ranking sequence under a constrained budget. Specifically, we design a ranking-tailored policy optimization, actively exploring optimal multi-round ranking modification policy verified by LLMs via reinforcement learning (RL). To better approximate the target LLM under the constrained budget, we trigger the agent fully utilizing the listwise ranking knowledge about all items verified by LLMs across different rounds in RL, enhancing the modification policy of the agent. More importantly, we demonstrate the theoretical robustness and advantages of our paradigm and implementation. Experiments on both IR and RS tasks show the effectiveness of our proposed method.", "authors": ["Yingpeng Du", "Tianjun Wei", "Zhu Sun", "Jie Zhang"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-23", "url": "https://arxiv.org/abs/2505.20316", "pdf_url": "https://arxiv.org/pdf/2505.20316v1", "arxiv_id": "2505.20316", "doi": "10.1145/3770854.3780197", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Knowledge Discovery and Data Mining", "quality_score": 0.1747}
{"id": "80997e8b8e921f7183e83e59b57f243090f1f2bf1358c7c2ce515dad577ee0bc", "sources": ["arxiv", "semantic_scholar"], "title": "KNN-SSD: Enabling Dynamic Self-Speculative Decoding via Nearest Neighbor Layer Set Optimization", "abstract": "Speculative Decoding (SD) has emerged as a widely used paradigm to accelerate the inference of large language models (LLMs) without compromising generation quality. It works by efficiently drafting multiple tokens using a compact model and then verifying them in parallel using the target LLM. Notably, Self-Speculative Decoding proposes skipping certain layers to construct the draft model, which eliminates the need for additional parameters or training. Despite its strengths, we observe in this work that drafting with layer skipping exhibits significant sensitivity to domain shifts, leading to a substantial drop in acceleration performance. To enhance the domain generalizability of this paradigm, we introduce KNN-SSD, an algorithm that leverages K-Nearest Neighbor (KNN) search to match different skipped layers with various domain inputs. We evaluated our algorithm in various models and multiple tasks, observing that its application leads to 1.3x-1.6x speedup in LLM inference.", "authors": ["Mingbo Song", "Heming Xia", "Jun Zhang", "Chak Tou Leong", "Qiancheng Xu", "Wenjie Li", "Sujian Li"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-22", "url": "https://arxiv.org/abs/2505.16162", "pdf_url": "https://arxiv.org/pdf/2505.16162v2", "arxiv_id": "2505.16162", "doi": "10.48550/arXiv.2505.16162", "citation_count": 7, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Conference of the European Chapter of the Association for Computational Linguistics", "quality_score": 0.2258}
{"id": "3d88c8e0624bfcf7a85816dbfc7cb8fffbae4a820c8f0cc0ef6dfbe71c1db87f", "sources": ["arxiv", "semantic_scholar"], "title": "BanditSpec: Adaptive Speculative Decoding via Bandit Algorithms", "abstract": "Speculative decoding has emerged as a popular method to accelerate the inference of Large Language Models (LLMs) while retaining their superior text generation performance. Previous methods either adopt a fixed speculative decoding configuration regardless of the prefix tokens, or train draft models in an offline or online manner to align them with the context. This paper proposes a training-free online learning framework to adaptively choose the configuration of the hyperparameters for speculative decoding as text is being generated. We first formulate this hyperparameter selection problem as a Multi-Armed Bandit problem and provide a general speculative decoding framework BanditSpec. Furthermore, two bandit-based hyperparameter selection algorithms, UCBSpec and EXP3Spec, are designed and analyzed in terms of a novel quantity, the stopping time regret. We upper bound this regret under both stochastic and adversarial reward settings. By deriving an information-theoretic impossibility result, it is shown that the regret performance of UCBSpec is optimal up to universal constants. Finally, extensive empirical experiments with LLaMA3 and Qwen2 demonstrate that our algorithms are effective compared to existing methods, and the throughput is close to the oracle best hyperparameter in simulated real-life LLM serving scenarios with diverse input prompts.", "authors": ["Yunlong Hou", "Fengzhuo Zhang", "Cunxiao Du", "Xuan Zhang", "Jiachun Pan", "Tianyu Pang", "Chao Du", "Vincent Y. F. Tan", "Zhuoran Yang"], "categories": ["cs.LG", "cs.AI", "stat.ML"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2025-05-21", "url": "https://arxiv.org/abs/2505.15141", "pdf_url": "https://arxiv.org/pdf/2505.15141v2", "arxiv_id": "2505.15141", "doi": "10.48550/arXiv.2505.15141", "citation_count": 12, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.2785}
{"id": "41fbeadd3e9a1a262543b4bc6ee522f452d040120be0817b0f6e732c7cfae48c", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Autoregressive Speech Synthesis Inference With Speech Speculative Decoding", "abstract": "Modern autoregressive speech synthesis models leveraging language models have demonstrated remarkable performance. However, the sequential nature of next token prediction in these models leads to significant latency, hindering their deployment in scenarios where inference speed is critical. In this work, we propose Speech Speculative Decoding (SSD), a novel framework for autoregressive speech synthesis acceleration. Specifically, our method employs a lightweight draft model to generate candidate token sequences, which are subsequently verified in parallel by the target model using the proposed SSD framework. Experimental results demonstrate that SSD achieves a significant speedup of 1.4x compared with conventional autoregressive decoding, while maintaining high fidelity and naturalness. Subjective evaluations further validate the effectiveness of SSD in preserving the perceptual quality of the target model while accelerating inference.", "authors": ["Zijian Lin", "Yang Zhang", "Yougen Yuan", "Yuming Yan", "Jinjiang Liu", "Zhiyong Wu", "Pengfei Hu", "Qun Yu"], "categories": ["cs.SD", "cs.AI", "eess.AS"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2025-05-21", "url": "https://arxiv.org/abs/2505.15380", "pdf_url": "https://arxiv.org/pdf/2505.15380v2", "arxiv_id": "2505.15380", "doi": "10.48550/arXiv.2505.15380", "citation_count": 6, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Interspeech", "quality_score": 0.2113}
{"id": "349a19a188c37991f76e5642cbccf771803718665627aca56ea602949140ba22", "sources": ["arxiv", "semantic_scholar"], "title": "Semi-Clairvoyant Scheduling of Speculative Decoding Requests to Minimize LLM Inference Latency", "abstract": "Speculative decoding accelerates Large Language Model (LLM) inference by employing a small speculative model (SSM) to generate multiple candidate tokens and verify them using the LLM in parallel. This technique has been widely integrated into LLM inference serving systems. However, inference requests typically exhibit uncertain execution time, which poses a significant challenge of efficiently scheduling requests in these systems. Existing work estimates execution time based solely on predicted output length, which could be inaccurate because execution time depends on both output length and token acceptance rate of verification by the LLM. In this paper, we propose a semi-clairvoyant request scheduling algorithm called Least-Attained/Perceived-Service for Speculative Decoding (LAPS-SD). Given a number of inference requests, LAPS-SD can effectively minimize average inference latency by adaptively scheduling requests according to their features during decoding. When the token acceptance rate is dynamic and execution time is difficult to estimate, LAPS-SD maintains multiple priority queues and allows request execution preemption across different queues. Once the token acceptance rate becomes stable, LAPS-SD can accurately estimate the execution time and schedule requests accordingly. Extensive experiments show that LAPS-SD reduces inference latency by approximately 39\\% compared to state-of-the-art scheduling methods.", "authors": ["Ruixiao Li", "Fahao Chen", "Peng Li"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-20", "url": "https://arxiv.org/abs/2505.17074", "pdf_url": "https://arxiv.org/pdf/2505.17074v1", "arxiv_id": "2505.17074", "doi": "10.48550/arXiv.2505.17074", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "International Joint Conference on Artificial Intelligence", "quality_score": 0.1661}
{"id": "7804e2f5a3739bf3c46ca4cefd0542741de31932d3b973ee9fe8a66393d63819", "sources": ["arxiv", "semantic_scholar"], "title": "STree: Speculative Tree Decoding for Hybrid State-Space Models", "abstract": "Speculative decoding is a technique to leverage hardware concurrency in order to enable multiple steps of token generation in a single forward pass, thus improving the efficiency of large-scale autoregressive (AR) Transformer models. State-space models (SSMs) are already more efficient than AR Transformers, since their state summarizes all past data with no need to cache or re-process tokens in the sliding window context. However, their state can also comprise thousands of tokens; so, speculative decoding has recently been extended to SSMs. Existing approaches, however, do not leverage the tree-based verification methods, since current SSMs lack the means to compute a token tree efficiently. We propose the first scalable algorithm to perform tree-based speculative decoding in state-space models (SSMs) and hybrid architectures of SSMs and Transformer layers. We exploit the structure of accumulated state transition matrices to facilitate tree-based speculative decoding with minimal overhead relative to current SSM implementations. Along with the algorithm, we describe a hardware-aware implementation that improves naive application of AR Transformer tree-based speculative decoding methods to SSMs. Furthermore, we outperform vanilla speculative decoding with SSMs even with a baseline drafting model and tree structure on three different benchmarks, opening up opportunities for further speed up with SSM and hybrid model inference. Code can be found at: https://github.com/wyc1997/stree.", "authors": ["Yangchao Wu", "Zongyue Qin", "Alex Wong", "Stefano Soatto"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-20", "url": "https://arxiv.org/abs/2505.14969", "pdf_url": "https://arxiv.org/pdf/2505.14969v2", "arxiv_id": "2505.14969", "doi": "10.48550/arXiv.2505.14969", "citation_count": 4, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/wyc1997/stree", "venue": "arXiv.org", "quality_score": 0.2568}
{"id": "e6f0c3891a75ddbcfb95e415fa33476df62d6648ec5f3722fe7c605b7ff2d87d", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding Reimagined for Multimodal Large Language Models", "abstract": "This paper introduces Multimodal Speculative Decoding (MSD) to accelerate Multimodal Large Language Models (MLLMs) inference. Speculative decoding has been shown to accelerate Large Language Models (LLMs) without sacrificing accuracy. However, current speculative decoding methods for MLLMs fail to achieve the same speedup as they do for LLMs. To address this, we reimagine speculative decoding specifically for MLLMs. Our analysis of MLLM characteristics reveals two key design principles for MSD: (1) Text and visual tokens have fundamentally different characteristics and need to be processed separately during drafting. (2) Both language modeling ability and visual perception capability are crucial for the draft model. For the first principle, MSD decouples text and visual tokens in the draft model, allowing each to be handled based on its own characteristics. For the second principle, MSD uses a two-stage training strategy: In stage one, the draft model is trained on text-only instruction-tuning datasets to improve its language modeling ability. In stage two, MSD gradually introduces multimodal data to enhance the visual perception capability of the draft model. Experiments show that MSD boosts inference speed by up to $2.29\\times$ for LLaVA-1.5-7B and up to $2.46\\times$ for LLaVA-1.5-13B on multimodal benchmarks, demonstrating its effectiveness. Our code is available at https://github.com/Lyn-Lucy/MSD.", "authors": ["Luxi Lin", "Zhihang Lin", "Zhanpeng Zeng", "Rongrong Ji"], "categories": ["cs.CV", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-20", "url": "https://arxiv.org/abs/2505.14260", "pdf_url": "https://arxiv.org/pdf/2505.14260v1", "arxiv_id": "2505.14260", "doi": "10.48550/arXiv.2505.14260", "citation_count": 6, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/Lyn-Lucy/MSD", "venue": "arXiv.org", "quality_score": 0.301}
{"id": "452cfdf70ebd5ed96dfd486408b4a99b4f8c5cc487ccce6a53b5f6a69b93045c", "sources": ["arxiv", "semantic_scholar"], "title": "HeteroSpec: Leveraging Contextual Heterogeneity for Efficient Speculative Decoding", "abstract": "Autoregressive decoding inherently limits the inference throughput of Large Language Model (LLM) due to its sequential dependency. Speculative decoding mitigates this by verifying multiple predicted tokens in parallel, but its efficiency remains constrained by what we identify as verification heterogeneity -- the uneven difficulty of verifying different speculative candidates. In practice, a small subset of high-confidence predictions accounts for most successful verifications, yet existing methods treat all candidates uniformly, leading to redundant computation. We present HeteroSpec, a heterogeneity-adaptive speculative decoding framework that allocates verification effort in proportion to candidate uncertainty. HeteroSpec estimates verification complexity using a lightweight entropy-based quantifier, partitions candidates via a data-driven stratification policy, and dynamically tunes speculative depth and pruning thresholds through coordinated optimization. Across five benchmarks and four LLMs, HeteroSpec delivers an average 4.24$\\times$ decoding speedup over state-of-the-art methods such as EAGLE-3, while preserving exact output distributions. Crucially, HeteroSpec requires no model retraining and remains compatible with other inference optimizations, making it a practical direction for improving speculative decoding efficiency.", "authors": ["Siran Liu", "Yang Ye", "Qianchao Zhu", "Zane Cao", "Yongchao He"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-19", "url": "https://arxiv.org/abs/2505.13254", "pdf_url": "https://arxiv.org/pdf/2505.13254v2", "arxiv_id": "2505.13254", "doi": "10.48550/arXiv.2505.13254", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.165}
{"id": "c4e3c1a68326dbb905e7c9aa82d85f85a3689de6da87a3d1365623758b352577", "sources": ["arxiv", "semantic_scholar"], "title": "SpecFLASH: A Latent-Guided Semi-autoregressive Speculative Decoding Framework for Efficient Multimodal Generation", "abstract": "Large language models and large multimodal models (LLMs and LMMs) deliver strong generative performance but suffer from slow decoding, a problem that becomes more severe when handling visual inputs, whose sequences typically contain many more tokens with lower information density than text. Speculative decoding accelerates LLM inference by letting a compact draft model propose candidate tokens that are selectively accepted by a larger target model, achieving speed-up without degrading quality. However, existing multimodal speculative decoding approaches largely ignore the structural characteristics of visual representations and usually rely on text-only draft models. In this paper, we introduce SpecFLASH, a speculative decoding framework tailored to LMMs that explicitly exploits multimodal structure when designing the draft model. We first mitigate redundancy in visual token sequences with a lightweight, latent-guided token compression module that compacts visual features while preserving semantics, and then leverage the co-occurrence and local correlations of visual entities via a semi-autoregressive decoding scheme that predicts multiple tokens in a single forward pass. Extensive experiments demonstrate that SpecFLASH consistently surpasses prior speculative decoding baselines, achieving up to $2.68\\times$ speed-up on video captioning and $2.55\\times$ on visual instruction tuning, relative to the original LMM. Our code is available here: https://github.com/ZihuaEvan/FlashSD/.", "authors": ["Zihua Wang", "Ruibo Li", "Haozhe Du", "Joey Tianyi Zhou", "Yu Zhang", "Xu Yang"], "categories": ["cs.CV", "cs.MM"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-19", "url": "https://arxiv.org/abs/2505.12728", "pdf_url": "https://arxiv.org/pdf/2505.12728v3", "arxiv_id": "2505.12728", "doi": null, "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/ZihuaEvan/FlashSD/", "venue": null, "quality_score": 0.195}
{"id": "3b4607797a56c785e0912b27a890e8bf85f3f3e29811ba5de7c9a8196c6caa68", "sources": ["arxiv", "semantic_scholar"], "title": "Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification", "abstract": "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 alignment with training-based methods, e.g., EAGLE, Medusa, involving considerable training costs. In this paper, we present a training-free alignment-augmented speculative decoding algorithm. We propose alignment sampling, which leverages output distribution obtained in the prefilling phase to provide more aligned draft candidates. To further benefit from high-quality but non-aligned draft candidates, we also introduce a simple yet effective flexible verification strategy. Through an adaptive probability threshold, our approach can improve generation accuracy while further improving inference efficiency. Experiments on 8 datasets (including question answering, summarization and code completion tasks) show that our approach increases the average generation score by 3.3 points for the LLaMA3 model. Our method achieves a mean acceptance length up to 2.39 and speed up generation by 2.23.", "authors": ["Jikai Wang", "Zhenxu Tian", "Juntao Li", "Qingrong Xia", "Xinyu Duan", "Zhefeng Wang", "Baoxing Huai", "Min Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-19", "url": "https://arxiv.org/abs/2505.13204", "pdf_url": "https://arxiv.org/pdf/2505.13204v2", "arxiv_id": "2505.13204", "doi": "10.48550/arXiv.2505.13204", "citation_count": 8, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.2386}
{"id": "4eeda6b53d1657b1cf7de184f5e1f06c5b24026fc731a4fa38b9fa0c60ed494a", "sources": ["arxiv", "semantic_scholar"], "title": "Traversal Verification for Speculative Tree Decoding", "abstract": "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. To enhance acceptance rates, existing frameworks typically construct token trees containing multiple candidates in each timestep. However, their reliance on token-level verification mechanisms introduces two critical limitations: First, the probability distribution of a sequence differs from that of individual tokens, leading to suboptimal acceptance length. Second, current verification schemes begin from the root node and proceed layer by layer in a top-down manner. Once a parent node is rejected, all its child nodes should be discarded, resulting in inefficient utilization of speculative candidates. This paper introduces Traversal Verification, a novel speculative decoding algorithm that fundamentally rethinks the verification paradigm through leaf-to-root traversal. Our approach considers the acceptance of the entire token sequence from the current node to the root, and preserves potentially valid subsequences that would be prematurely discarded by existing methods. We theoretically prove that the probability distribution obtained through Traversal Verification is identical to that of the target model, guaranteeing lossless inference while achieving substantial acceleration gains. Experimental results across different large language models and multiple tasks show that our method consistently improves acceptance length and throughput over existing methods.", "authors": ["Yepeng Weng", "Qiao Hu", "Xujie Chen", "Li Liu", "Dianwen Mei", "Huishi Qiu", "Jiang Tian", "Zhongchao Shi"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-18", "url": "https://arxiv.org/abs/2505.12398", "pdf_url": "https://arxiv.org/pdf/2505.12398v2", "arxiv_id": "2505.12398", "doi": "10.48550/arXiv.2505.12398", "citation_count": 6, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2113}
{"id": "84de75f78a7bda16010ed8a9e7b3da7528cb019d3602670e25ee17ef2db45363", "sources": ["arxiv", "semantic_scholar"], "title": "SpecMemo: Speculative Decoding is in Your Pocket", "abstract": "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 speculative decoding on memory-constrained devices, such as mobile GPUs, remains as a significant challenge in real-world scenarios. In this work, we present a device-aware inference engine named SpecMemo that can smartly control memory allocations at finer levels to enable multi-turn chatbots with speculative decoding on such limited memory devices. Our methodology stems from theoretically modeling memory footprint of speculative decoding to determine a lower bound on the required memory budget while retaining speedup. SpecMemo empirically acquires a careful balance between minimizing redundant memory allocations for rejected candidate tokens and maintaining competitive performance gains from speculation. Notably, with SpecMemo's memory management, we maintain 96% of overall throughput from speculative decoding on MT-Bench, with reduced generation-memory by 65% on single Nvidia Titan RTX. Given multiple constrained GPUs, we build on top of previous speculative decoding architectures to facilitate big-model inference by distributing Llama-2-70B-Chat model, on which we provide novel batched speculative decoding to increase usability of multiple small server GPUs. This novel framework demonstrates 2x speedup over distributed and batched vanilla decoding with the base model on eight AMD MI250 GPUs. Moreover, inference throughput increases remarkably 8x with batch size 10. Our work contributes to democratized LLM applications in resource-constrained environments, providing a pathway for faster and cheaper deployment of real-world LLM applications with robust performance.", "authors": ["Selin Yildirim", "Deming Chen"], "categories": ["cs.LG", "cs.AI", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-16", "url": "https://arxiv.org/abs/2506.01986", "pdf_url": "https://arxiv.org/pdf/2506.01986v1", "arxiv_id": "2506.01986", "doi": "10.48550/arXiv.2506.01986", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1616}
{"id": "7d116ea662927547c9a1fdfe569b8c728f1dd63762080583f87a7283d7243c1e", "sources": ["arxiv", "semantic_scholar"], "title": "SpecBranch: Speculative Decoding via Hybrid Drafting and Rollback-Aware Branch Parallelism", "abstract": "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 serialized execution, which causes the mutual waiting bubbles between the draft and target models. To address this challenge, we draw inspiration from branch prediction in modern processors and propose a novel framework \\textbf{SpecBranch} to unlock branch parallelism in SD. Specifically, we first take an in-depth analysis of the potential of branch parallelism in SD, and recognize that the key challenge lies in the trade-offs between parallelization and token rollback. Based on the analysis, we strategically introduce parallel speculative branches to preemptively hedge against likely rejections. Meanwhile, to enhance parallelism, we jointly orchestrate adaptive draft lengths with a hybrid combination of the implicit draft model confidence and explicit reusing of target model features. Extensive experiments across various models and benchmarks show that SpecBranch achieves over \\textbf{1.8}$\\times \\sim$ \\textbf{4.5}$\\times$ speedups against the auto-regressive decoding and reduces rollback tokens by $\\textbf{50}$\\% for poorly aligned models, realizing its applicability for real-world deployments.", "authors": ["Yuhao Shen", "Junyi Shen", "Quan Kong", "Tianyu Liu", "Yao Lu", "Cong Wang"], "categories": ["cs.DC", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-16", "url": "https://arxiv.org/abs/2506.01979", "pdf_url": "https://arxiv.org/pdf/2506.01979v4", "arxiv_id": "2506.01979", "doi": null, "citation_count": 12, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.2785}
{"id": "0a32a41125d14672f832133dfa4090cc77dea1d1b6035f7e47382f9998d02a91", "sources": ["arxiv", "semantic_scholar"], "title": "MASSV: Multimodal Adaptation and Self-Data Distillation for Speculative Decoding of Vision-Language Models", "abstract": "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 could serve as efficient drafters lack the architectural components to process visual inputs, and their token predictions fail to match those of VLM target models that consider visual context. We introduce Multimodal Adaptation and Self-Data Distillation for Speculative Decoding of Vision-Language Models (MASSV), which transforms existing small language models into effective multimodal drafters through a two-phase approach. MASSV first connects the target VLM's vision encoder to the draft model via a lightweight trainable projector, then applies self-distilled visual instruction tuning using responses generated by the target VLM to align token predictions. Comprehensive experiments across the Qwen2.5-VL and Gemma3 model families demonstrate that MASSV increases accepted length by up to 30% and delivers end-to-end inference speedups of up to 1.46x on visually-grounded tasks. MASSV provides a scalable, architecture-compatible method for accelerating both current and future VLMs.", "authors": ["Mugilan Ganesan", "Shane Segal", "Ankur Aggarwal", "Nish Sinnadurai", "Sean Lie", "Vithursan Thangarasa"], "categories": ["cs.LG", "cs.CL", "cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-15", "url": "https://arxiv.org/abs/2505.10526", "pdf_url": "https://arxiv.org/pdf/2505.10526v2", "arxiv_id": "2505.10526", "doi": "10.48550/arXiv.2505.10526", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.1604}
{"id": "5e6c3e7feb57d72657e9ec2437560c424ddd21c135dc1bce2937a2bb410cdbc2", "sources": ["arxiv", "semantic_scholar"], "title": "SpecOffload: Unlocking Latent GPU Capacity for LLM Inference on Resource-Constrained Devices", "abstract": "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 are underutilized, often remaining idle while waiting for data to be loaded; and (2) GPU memory has low impact on performance, as reducing its capacity has minimal effect on overall throughput.In this paper, we propose SpecOffload, a high-throughput inference engine that embeds speculative decoding into offloading. Our key idea is to unlock latent GPU resources for storing and executing a draft model used for speculative decoding, thus accelerating inference at near-zero additional cost. To support this, we carefully orchestrate the interleaved execution of target and draft models in speculative decoding within the offloading pipeline, and propose a planner to manage tensor placement and select optimal parameters. Compared to the best baseline, SpecOffload improves GPU core utilization by 4.49x and boosts inference throughput by 2.54x. Our code is available at https://github.com/MobiSense/SpecOffload-public .", "authors": ["Xiangwen Zhuge", "Xu Shen", "Zeyu Wang", "Fan Dang", "Xuan Ding", "Danyang Li", "Yahui Han", "Tianxiang Hao", "Zheng Yang"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-15", "url": "https://arxiv.org/abs/2505.10259", "pdf_url": "https://arxiv.org/pdf/2505.10259v3", "arxiv_id": "2505.10259", "doi": "10.48550/arXiv.2505.10259", "citation_count": 1, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/MobiSense/SpecOffload-public", "venue": "arXiv.org", "quality_score": 0.2479}
{"id": "3128cc6a9ba2fa0fd63a91c1e8ff9eaa9673b7930f229f6ba2546c7587b6533b", "sources": ["arxiv", "semantic_scholar"], "title": "ELIS: Efficient LLM Iterative Scheduling System with Response Length Predictor", "abstract": "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 lead to the \"head-of-line blocking\" problem. To overcome this limitation, it is necessary to predict LLM inference times and apply a shortest job first scheduling strategy. However, due to the auto-regressive nature of LLMs, predicting the inference latency is challenging. ELIS addresses this challenge by training a response length predictor for LLMs using the BGE model, an encoder-based state-of-the-art model. Additionally, we have devised the ISRTF scheduling strategy, an optimization of shortest remaining time first tailored to existing LLM iteration batching. To evaluate our work in an industrial setting, we simulate streams of requests based on our study of real-world user LLM serving trace records. Furthermore, we implemented ELIS as a cloud-native scheduler system on Kubernetes to evaluate its performance in production environments. Our experimental results demonstrate that ISRTF reduces the average job completion time by up to 19.6%.", "authors": ["Seungbeom Choi", "Jeonghoe Goo", "Eunjoo Jeon", "Mingyu Yang", "Minsung Jang"], "categories": ["cs.DC", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-14", "url": "https://arxiv.org/abs/2505.09142", "pdf_url": "https://arxiv.org/pdf/2505.09142v1", "arxiv_id": "2505.09142", "doi": "10.48550/arXiv.2505.09142", "citation_count": 7, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2258}
{"id": "84dd8a389eaf47c74725765a7b041096988401415ebd69f559639b91f423d5f5", "sources": ["arxiv", "semantic_scholar"], "title": "Automatic Task Detection and Heterogeneous LLM Speculative Decoding", "abstract": "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 model, making it difficult to ensure efficiency across diverse tasks. To address this problem, we propose a speculative decoding algorithm tailored for downstream task optimization. It includes an automatic task partitioning and assigning method, which automatically categorizes downstream tasks into different sub-tasks and assigns them to a set of heterogeneous draft models. Each draft model is aligned with the target model using task-specific data, thereby enhancing the consistency of inference results. In addition, our proposed method incorporates an online lightweight prompt classifier to dynamically route prompts to the appropriate draft model. Experimental results demonstrate that the proposed method improves draft accuracy by 6% to 50% over vanilla speculative decoding, while achieving a speedup of 1.10x to 2.64x in LLM inference.", "authors": ["Danying Ge", "Jianhua Gao", "Qizhi Jiang", "Yifei Feng", "Weixing Ji"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-13", "url": "https://arxiv.org/abs/2505.08600", "pdf_url": "https://arxiv.org/pdf/2505.08600v1", "arxiv_id": "2505.08600", "doi": "10.48550/arXiv.2505.08600", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1581}
{"id": "af0e472eecc7b3b4416d6be55cbbe9eec7bb07a4cf3df5ee65cf7e2bdfb979dc", "sources": ["arxiv", "semantic_scholar"], "title": "SpecRouter: Adaptive Routing for Multi-Level Speculative Decoding in Large Language Models", "abstract": "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 speculative decoding, failing to dynamically adapt to the varying complexities of user requests or fluctuations in system performance. This paper introduces \\systemname{}, a novel framework that reimagines LLM inference as an adaptive routing problem solved through multi-level speculative decoding. \\systemname{} dynamically constructs and optimizes inference \"paths\" (chains of models) based on real-time feedback, addressing the limitations of static approaches. Our contributions are threefold: (1) An \\textbf{adaptive model chain scheduling} mechanism that leverages performance profiling (execution times) and predictive similarity metrics (derived from token distribution divergence) to continuously select the optimal sequence of draft and verifier models, minimizing predicted latency per generated token. (2) A \\textbf{multi-level collaborative verification} framework where intermediate models within the selected chain can validate speculative tokens, reducing the verification burden on the final, most powerful target model. (3) A \\textbf{synchronized state management} system providing efficient, consistent KV cache handling across heterogeneous models in the chain, including precise, low-overhead rollbacks tailored for asynchronous batch processing inherent in multi-level speculation. Preliminary experiments demonstrate the validity of our method.", "authors": ["Hang Wu", "Jianian Zhu", "Yinghui Li", "Haojie Wang", "Biao Hou", "Jidong Zhai"], "categories": ["cs.LG", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-12", "url": "https://arxiv.org/abs/2505.07680", "pdf_url": "https://arxiv.org/pdf/2505.07680v1", "arxiv_id": "2505.07680", "doi": "10.48550/arXiv.2505.07680", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.157}
{"id": "ff5672cdcffa20de10c8d92ecac2d9842be0fad501fc4b43442304f0ec7a87e3", "sources": ["arxiv", "semantic_scholar"], "title": "Scaling Laws for Speculative Decoding", "abstract": "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 establish foundational insights for accelerating reasoning tasks. While speculative decoding methods leveraging parallel draft-verification cycles have emerged as promising acceleration techniques, the scaling laws governing decoding efficiency remain under-explored compared to conventional backbone LLMs developed through Pretraining->SFT->RLHF training paradigms. In this work, we discover Log-linear Scaling Laws (Theorem 1.1, 1.2 and 1.3) governing draft model acceptance rate (or decoding speed) across three dimensions: pretraining token volume, draft model capacity, and decoding batch size. Building on these laws, we achieve Scylla, which coordinates multi-dimensional scaling for popular LLMs (Llama2/3, Qwen2.5). Empirical validation shows Scylla achieves 1.5-2.2 higher acceptance rate than EAGLE2 and 0.3 higher than EAGLE3 at temperature T = 0, with peak performance gains on summarization and QA tasks (Figure 2). Industrial inference engine deployments demonstrate 2X decoding throughput improvements over EAGLE2 (Table 5), validating the transformative potential of systematic scaling for efficient LLM inference. Code will be released later.", "authors": ["Siyuan Yan", "Mo Zhu", "Guo-qing Jiang", "Jianfei Wang", "Jiaxing Chen", "Wentai Zhang", "Xiang Liao", "Xiao Cui", "Chen Zhang", "Zhuoran Song", "Ran Zhu"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-08", "url": "https://arxiv.org/abs/2505.07858", "pdf_url": "https://arxiv.org/pdf/2505.07858v1", "arxiv_id": "2505.07858", "doi": "10.48550/arXiv.2505.07858", "citation_count": 3, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1524}
{"id": "aa29d28a1f46f1a2de01f61cdd67faf40632c4aad43f6f425c9591867abada65", "sources": ["arxiv", "semantic_scholar"], "title": "SOAEsV2-7B/72B: Full-Pipeline Optimization for State-Owned Enterprise LLMs via Continual Pre-Training, Domain-Progressive SFT and Distillation-Enhanced Speculative Decoding", "abstract": "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 domain-specific supervised fine-tuning (SFT) data, which neglects the broader applicability of general language patterns; and 3) inefficient inference acceleration for large models processing long contexts. In this work, we propose SOAEsV2-7B/72B, a specialized LLM series developed via a three-phase framework: 1) continual pre-training integrates domain knowledge while retaining base capabilities; 2) domain-progressive SFT employs curriculum-based learning strategy, transitioning from weakly relevant conversational data to expert-annotated SOAEs datasets to optimize domain-specific tasks; 3) distillation-enhanced speculative decoding accelerates inference via logit distillation between 72B target and 7B draft models, achieving 1.39-1.52$\\times$ speedup without quality loss. Experimental results demonstrate that our domain-specific pre-training phase maintains 99.8% of original general language capabilities while significantly improving domain performance, resulting in a 1.08$\\times$ improvement in Rouge-1 score and a 1.17$\\times$ enhancement in BLEU-4 score. Ablation studies further show that domain-progressive SFT outperforms single-stage training, achieving 1.02$\\times$ improvement in Rouge-1 and 1.06$\\times$ in BLEU-4. Our work introduces a comprehensive, full-pipeline approach for optimizing SOAEs LLMs, bridging the gap between general language capabilities and domain-specific expertise.", "authors": ["Jingyang Deng", "Ran Chen", "Jo-Ku Cheng", "Jinwen Ma"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-07", "url": "https://arxiv.org/abs/2505.04723", "pdf_url": "https://arxiv.org/pdf/2505.04723v1", "arxiv_id": "2505.04723", "doi": "10.48550/arXiv.2505.04723", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1513}
{"id": "20f8beb58b7db52d32910b1513853ab3014089ebbac64d56bafe19b8a3170e5f", "sources": ["arxiv", "semantic_scholar"], "title": "PipeSpec: Breaking Stage Dependencies in Hierarchical LLM Decoding", "abstract": "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 speculative decoding to $k$ models arranged in a hierarchical pipeline, enabling asynchronous execution with lightweight coordination for prediction verification and rollback. Our analytical model characterizes token generation rates across pipeline stages and proves guaranteed throughput improvements over traditional decoding for any non-zero acceptance rate. We further derive closed-form expressions for steady-state verification probabilities that explain the empirical benefits of pipeline depth. Experimental results show that PipeSpec achieves up to 2.54$\\times$ speedup while outperforming state-of-the-art methods. We validate PipeSpec across text summarization and code generation tasks using LLaMA 2 and 3 models, demonstrating that pipeline efficiency increases with model depth, providing a scalable approach to accelerating LLM inference on multi-device systems.", "authors": ["Bradley McDanel", "Sai Qian Zhang", "Yunhai Hu", "Zining Liu"], "categories": ["cs.AI", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-05-02", "url": "https://arxiv.org/abs/2505.01572", "pdf_url": "https://arxiv.org/pdf/2505.01572v1", "arxiv_id": "2505.01572", "doi": "10.48550/arXiv.2505.01572", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.1945}
{"id": "9752dc211fc57f85d7976f78d0d672dca1ab107ba0d1e5e6aa7e933709c15bda", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Sampling via Exponential Races", "abstract": "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 information-theoretic analysis of the speed up that can be achieved by speculative decoding. Leveraging this link, we derive an explicit relation between generation speed-up and the number of tokens $k$ generated by the draft model for large $k$, which serves as an upper bound for all $k$. We also propose a novel speculative decoding method via exponential race ERSD that matches state-of-the-art performance.", "authors": ["Szymon Kobus", "Deniz Gündüz"], "categories": ["cs.CL", "cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2025-04-21", "url": "https://arxiv.org/abs/2504.15475", "pdf_url": "https://arxiv.org/pdf/2504.15475v1", "arxiv_id": "2504.15475", "doi": "10.48550/arXiv.2504.15475", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.1329}
{"id": "8095e57bc85578032b7dc5eb627c54186cb79f0999ad826555f43ab05efbe4b7", "sources": ["arxiv", "semantic_scholar"], "title": "MIST: A Co-Design Framework for Heterogeneous, Multi-Stage LLM Inference", "abstract": "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 costs. Simultaneously, the hardware landscape is rapidly diversifying across AMD GPUs, TPUs, and custom ASICs, while cross-vendor prefill-decode (PD) disaggregated configurations lack unified software stacks for end-to-end evaluation today. To address this gap, we present MIST, a Heterogeneous Multi-stage LLM inference Execution Simulator. MIST models diverse request stages; including RAG, KV retrieval, reasoning, prefill, and decode across complex hardware hierarchies. MIST supports heterogeneous clients executing multiple models concurrently unlike prior frameworks while incorporating advanced batching strategies and multi-level memory hierarchies. By integrating real hardware traces with analytical modeling, MIST captures critical trade-offs such as memory bandwidth contention, inter-cluster communication latency, and batching efficiency in hybrid CPU-accelerator deployments. Through case studies, we explore the impact of reasoning stages on end-to-end latency, optimal batching strategies for hybrid pipelines, and the architectural implications of remote KV cache retrieval. MIST empowers system designers to navigate the evolving landscape of LLM inference, providing actionable insights into optimizing hardware-software co-design for next-generation AI workloads.", "authors": ["Abhimanyu Rajeshkumar Bambhaniya", "Hanjiang Wu", "Suvinay Subramanian", "Sudarshan Srinivasan", "Souvik Kundu", "Amir Yazdanbakhsh", "Midhilesh Elavazhagan", "Madhu Kumar", "Minlan Yu", "Arijit Raychowdhury", "Tushar Krishna"], "categories": ["cs.AR", "cs.AI", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-04-14", "url": "https://arxiv.org/abs/2504.09775", "pdf_url": "https://arxiv.org/pdf/2504.09775v6", "arxiv_id": "2504.09775", "doi": null, "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.1193}
{"id": "1792d10cbdf711c47d677c52532ffdf6e7b1c0f551972561156d287ecedf90b7", "sources": ["arxiv", "semantic_scholar"], "title": "SpecReason: Fast and Accurate Inference-Time Compute via Speculative Reasoning", "abstract": "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 autoregressive nature of decoding. Our key insight in tackling these overheads is that LRM inference, and the reasoning that it embeds, is highly tolerant of approximations: complex tasks are typically broken down into simpler steps, each of which brings utility based on the semantic insight it provides for downstream steps rather than the exact tokens it generates. Accordingly, we introduce SpecReason, a system that automatically accelerates LRM inference by using a lightweight model to (speculatively) carry out simpler intermediate reasoning steps and reserving the costly base model only to assess (and potentially correct) the speculated outputs. Importantly, SpecReason's focus on exploiting the semantic flexibility of thinking tokens in preserving final-answer accuracy is complementary to prior speculation techniques, most notably speculative decoding, which demands token-level equivalence at each step. Across a variety of reasoning benchmarks, SpecReason achieves $1.4-3.0\\times$ speedup over vanilla LRM inference while improving accuracy by $0.4-9.0\\%$. Compared to speculative decoding without SpecReason, their combination yields an additional $8.8-58.0\\%$ latency reduction. We open-source SpecReason at https://github.com/ruipeterpan/specreason.", "authors": ["Rui Pan", "Yinwei Dai", "Zhihao Zhang", "Gabriele Oliaro", "Zhihao Jia", "Ravi Netravali"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-04-10", "url": "https://arxiv.org/abs/2504.07891", "pdf_url": "https://arxiv.org/pdf/2504.07891v2", "arxiv_id": "2504.07891", "doi": "10.48550/arXiv.2504.07891", "citation_count": 57, "influential_citation_count": 12, "has_code": true, "code_url": "https://github.com/ruipeterpan/specreason", "venue": "arXiv.org", "quality_score": 0.557}
{"id": "915bb99440a5c9610ec1a3bea89bcf3481b8232a3aa737c429cab050c6b7dd56", "sources": ["arxiv", "semantic_scholar"], "title": "SPIRe: Boosting LLM Inference Throughput with Speculative Decoding", "abstract": "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 the context is sufficiently long and the draft model's KV cache is sparse. We introduce SPIRe, a draft model that combines static sparse attention, pruned initialization, and feedback memory to increase the modeled throughput of speculative decoding by over 100% compared to speculation with a much smaller draft model and by over 35% compared to the strong baseline of sparse self-speculation. Our approach is particularly effective when context lengths vary significantly across requests.", "authors": ["Sanjit Neelam", "Daniel Heinlein", "Vaclav Cvicek", "Akshay Mishra", "Reiner Pope"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-04-08", "url": "https://arxiv.org/abs/2504.06419", "pdf_url": "https://arxiv.org/pdf/2504.06419v1", "arxiv_id": "2504.06419", "doi": "10.48550/arXiv.2504.06419", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.118}
{"id": "7433730f6f8e81b2e4c9b38b2e2ea86ef83a71281ae262eeab79901d5b6c3346", "sources": ["arxiv", "semantic_scholar"], "title": "DEL: Context-Aware Dynamic Exit Layer for Efficient Self-Speculative Decoding", "abstract": "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 inference compared to auto-regressive decoding. While there are multiple approaches to create a draft model, one promising approach is to use early-exit methods. These methods draft candidate tokens by using a subset of layers of the primary model and applying the remaining layers for verification, allowing a single model to handle both drafting and verification. While this technique reduces memory usage and computational cost, its performance relies on the choice of the exit layer for drafting and the number of tokens drafted (speculation length) in each SD round. Prior works use hyperparameter exploration to statically select these values. However, our evaluations show that these hyperparameter values are task-specific, and even within a task they are dependent on the current sequence context. We introduce DEL (Dynamic Exit Layer), a plug-and-play method that adaptively selects the exit layer and speculation length during inference. DEL dynamically tracks the token acceptance rate if the tokens are drafted at each layer of an LLM and uses that knowledge to heuristically select the optimal exit layer and speculation length. Our experiments across a broad range of models and downstream tasks show that DEL achieves overall speedups of $2.16\\times$$\\sim$$2.62\\times$ over vanilla auto-regressive decoding and improves upon state-of-the-art SD methods, which peak at $2.43\\times$, by up to $0.19\\times$. The code is available at https://github.com/hoenza/DEL.", "authors": ["Hossein Entezari Zarch", "Lei Gao", "Chaoyi Jiang", "Murali Annavaram"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-04-08", "url": "https://arxiv.org/abs/2504.05598", "pdf_url": "https://arxiv.org/pdf/2504.05598v2", "arxiv_id": "2504.05598", "doi": "10.48550/arXiv.2504.05598", "citation_count": 6, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/hoenza/DEL", "venue": "arXiv.org", "quality_score": 0.2386}
{"id": "ecb96502b71d3bbac2182da1478f26fd93d76133bc5925d09aecfd890f732b71", "sources": ["arxiv", "semantic_scholar"], "title": "SpecPipe: Accelerating Pipeline Parallelism-based LLM Inference with Speculative Decoding", "abstract": "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 hardware utilization and narrow speculative window. Inspired by branch prediction in instruction pipelining, we introduce SpecPipe, which fills the pipeline with speculative tokens of a request step-by-step. By maximizing the hardware utilization, SpecPipe decodes one token per pipeline step ideally. Specifically, SpecPipe comprises a dynamic speculative token tree and a pipelined inference framework. The tree dynamically accepts tokens from a speculative token source and outputs the tokens to the inference pipeline. Since the speculative window relaxed in our framework, a high-accuracy draft model is integrated without fine-tuning. The pipeline inference framework follows node-wise computation, pruning propagation, and inter-node communication stages. We implement SpecPipe and a variant SpecPipe-DB with dynamic batching for single- and multi-request inference, respectively. On an 8-stage pipeline, SpecPipe improves time between tokens on diverse single-request workloads by $4.19\\times$-$5.53\\times$ over standard pipeline parallelism and by $2.08\\times$-$2.38\\times$ over prior tree-based speculative decoding methods. For multi-request workloads, SpecPipe-DB achieves $1.64\\times$-$2.08\\times$ higher throughput and $1.61\\times$-$2.06\\times$ lower time between tokens than vLLM.", "authors": ["Haofei Yin", "Mengbai Xiao", "Tinghong Li", "Xiao Zhang", "Dongxiao Yu", "Guanghui Zhang"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-04-05", "url": "https://arxiv.org/abs/2504.04104", "pdf_url": "https://arxiv.org/pdf/2504.04104v2", "arxiv_id": "2504.04104", "doi": null, "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.1747}
{"id": "02f2ea2ebb8810f78d0f4caa8723904d254e6c8c1448c4f66cab6a98dfac0760", "sources": ["arxiv", "semantic_scholar"], "title": "Token-Driven GammaTune: Adaptive Calibration for Enhanced Speculative Decoding", "abstract": "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{GammaTune} and \\textit{GammaTune+}, training-free adaptive algorithms that dynamically adjust speculation length based on token acceptance rates using a heuristic-based switching mechanism. Evaluated on SpecBench across multiple tasks and model pairs, our method outperforms other heuristic-based approaches and fixed-length speculative decoding, achieving an average speedup of 15\\% ($\\pm$5\\%) with \\textit{GammaTune} and 16\\% ($\\pm$3\\%) with \\textit{GammaTune+}, while reducing performance variance. This makes \\textit{GammaTune} a robust and efficient solution for real-world deployment.", "authors": ["Aayush Gautam", "Susav Shrestha", "Narasimha Reddy"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-28", "url": "https://arxiv.org/abs/2504.00030", "pdf_url": "https://arxiv.org/pdf/2504.00030v3", "arxiv_id": "2504.00030", "doi": "10.48550/arXiv.2504.00030", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1193}
{"id": "23e13f6df3b0fa061484c1c935a46ab6730e4b6af458f300a07d991f10293b61", "sources": ["arxiv", "semantic_scholar"], "title": "Collab: Controlled Decoding using Mixture of Agents for LLM Alignment", "abstract": "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 computationally expensive. Controlled Decoding, by contrast, provides a mechanism for aligning a model at inference time without retraining. However, single-agent decoding approaches often struggle to adapt to diverse tasks due to the complexity and variability inherent in these tasks. To strengthen the test-time performance w.r.t the target task, we propose a mixture of agent-based decoding strategies leveraging the existing off-the-shelf aligned LLM policies. Treating each prior policy as an agent in the spirit of mixture of agent collaboration, we develop a decoding method that allows for inference-time alignment through a token-level selection strategy among multiple agents. For each token, the most suitable LLM is dynamically chosen from a pool of models based on a long-term utility metric. This policy-switching mechanism ensures optimal model selection at each step, enabling efficient collaboration and alignment among LLMs during decoding. Theoretical analysis of our proposed algorithm establishes optimal performance with respect to the target task represented via a target reward for the given off-the-shelf models. We conduct comprehensive empirical evaluations with open-source aligned models on diverse tasks and preferences, which demonstrates the merits of this approach over single-agent decoding baselines. Notably, Collab surpasses the current SoTA decoding strategy, achieving an improvement of up to 1.56x in average reward and 71.89% in GPT-4 based win-tie rate.", "authors": ["Souradip Chakraborty", "Sujay Bhatt", "Udari Madhushani Sehwag", "Soumya Suvra Ghosal", "Jiahao Qiu", "Mengdi Wang", "Dinesh Manocha", "Furong Huang", "Alec Koppel", "Sumitra Ganesh"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-27", "url": "https://arxiv.org/abs/2503.21720", "pdf_url": "https://arxiv.org/pdf/2503.21720v1", "arxiv_id": "2503.21720", "doi": "10.48550/arXiv.2503.21720", "citation_count": 19, "influential_citation_count": 1, "has_code": true, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.3253}
{"id": "00471c250e77fe0be58a76d2b261c4723d1f5b7b308d115d15a14208d444eaeb", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerate Parallelizable Reasoning via Parallel Decoding within One Sequence", "abstract": "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 parallelizability of certain tasks to accelerate the reasoning process. Specifically, when multiple parallel reasoning steps exist, we decode multiple tokens per forward pass via a tree-like attention mask within a single sequence, avoiding additional memory usage. Experimental results show that our method achieves up to nearly 100\\% speedup in decoding while basically maintaining the answer quality.", "authors": ["Yijiong Yu"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-26", "url": "https://arxiv.org/abs/2503.20533", "pdf_url": "https://arxiv.org/pdf/2503.20533v4", "arxiv_id": "2503.20533", "doi": "10.48550/arXiv.2503.20533", "citation_count": 4, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/yuyijiong/parallel-decoding-in-one-sequence", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.1747}
{"id": "3875b7afdcb3d338913a88b0842ee39ee285a2dedcc979a1a7700f14ece8b39a", "sources": ["arxiv", "semantic_scholar"], "title": "A Multi-Model Adaptation of Speculative Decoding for Classification", "abstract": "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 decoding. The worker models, tasked with the bulk of the computation, independently predict discrete class labels for a given input. When majority worker models agree on a label, it is accepted as the final label, optimizing efficiency by bypassing the computationally expensive judge model. In cases of disagreement, the judge model intervenes to resolve the label. This approach minimizes redundant computation, leverages the redundancy of multiple workers for confidence, and confines the judge model's role to challenging cases, offering a practical balance of efficiency and accuracy. Our analysis suggests that smaller out of the box instruction/chat finetuned worker models with 3 billion parameters (hereafter, 3B) demonstrate a level of alignment with judge models comparable to that of larger finetuned worker models with 7 billion parameters (hereafter, 7B) across both simple and higher order reasoning tasks. The top performing 3B worker model pair achieve an agreement rate of approximately 80-83% for sentiment and around 50-80% for similar ticket when compared to judge models. Additionally, 3B worker models provide a speedup ranging from 2.8x to 9x relative to the judge models, while 7B worker model combinations achieve a speedup ranging from 1.28x to 0.28x", "authors": ["Somnath Roy", "Padharthi Sreekar", "Srivatsa Narasimha", "Anubhav Anand"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-23", "url": "https://arxiv.org/abs/2503.18076", "pdf_url": "https://arxiv.org/pdf/2503.18076v1", "arxiv_id": "2503.18076", "doi": "10.48550/arXiv.2503.18076", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.0997}
{"id": "8499012a8b1c554a4dca5e69b2c021fc1a6ec1d255aa7fd91ea53a1c2aa8c289", "sources": ["arxiv", "semantic_scholar"], "title": "SPIN: Accelerating Large Language Model Inference with Heterogeneous Speculative Models", "abstract": "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 speculative decoding approaches have three key limitations: handling requests with varying difficulty using homogeneous SSMs, lack of robust support for batch processing, and insufficient holistic optimization for both speculation and verification phases. In this paper, we introduce SPIN, an efficient LLM inference serving system based on speculative decoding, designed to address these challenges through three main innovations. First, SPIN improves token speculation by using multiple heterogeneous SSMs, with a learning-based algorithm for SSM selection that operates without prior knowledge of request difficulty. Second, SPIN employs a request decomposition method to minimize batching overhead during LLM verification. Finally, SPIN orchestrates speculation and verification phases by pipelining their executions on GPUs to achieve further acceleration. Experimental results demonstrate that SPIN significantly outperforms state-of-the-art methods, achieving a performance increase of approximately 2.28X.", "authors": ["Fahao Chen", "Peng Li", "Tom H. Luan", "Zhou Su", "Jing Deng"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-20", "url": "https://arxiv.org/abs/2503.15921", "pdf_url": "https://arxiv.org/pdf/2503.15921v1", "arxiv_id": "2503.15921", "doi": "10.1109/INFOCOM55648.2025.11044522", "citation_count": 14, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE Conference on Computer Communications", "quality_score": 0.294}
{"id": "0c403e5639e006c5d8916f31b1e6f74a4fee97444e3a51081acf4cabb849927c", "sources": ["arxiv", "semantic_scholar"], "title": "SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs", "abstract": "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 on long sequences. Existing KV cache compression methods include eviction, merging, or quantization of the KV cache to reduce its size. However, compression results in irreversible information forgetting, potentially affecting the accuracy of subsequent decoding. In this paper, we propose SpeCache, which takes full advantage of the large and easily expandable CPU memory to offload the complete KV cache, and dynamically fetches KV pairs back in each decoding step based on their importance measured by low-bit KV cache copy in VRAM. To avoid inference latency caused by CPU-GPU communication, SpeCache speculatively predicts the KV pairs that the next token might attend to, allowing us to prefetch them before the next decoding step which enables parallelization of prefetching and computation. Experiments on LongBench and Needle-in-a-Haystack benchmarks verify that SpeCache effectively reduces VRAM usage while avoiding information forgetting for long sequences without re-training, even with a 10x high KV cache compression ratio.", "authors": ["Shibo Jie", "Yehui Tang", "Kai Han", "Zhi-Hong Deng", "Jing Han"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-20", "url": "https://arxiv.org/abs/2503.16163", "pdf_url": "https://arxiv.org/pdf/2503.16163v1", "arxiv_id": "2503.16163", "doi": "10.48550/arXiv.2503.16163", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.1945}
{"id": "138aa166ad0bd48ec771b6ccd2e0111183c5a0d27dc21568bb9146cb9db0f511", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding for Verilog: Speed and Quality, All in One", "abstract": "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 challenges for conventional tokenization and decoding approaches. In this paper, we introduce a novel application of speculative decoding for Verilog code generation, showing that it can improve both inference speed and output quality, effectively achieving speed and quality all in one. Unlike standard LLM tokenization schemes, which often fragment meaningful code structures, our approach aligns decoding stops with syntactically significant tokens, making it easier for models to learn the token distribution. This refinement addresses inherent tokenization issues and enhances the model's ability to capture Verilog's logical constructs more effectively. Our experimental results show that our method achieves up to a 5.05x speedup in Verilog code generation and increases pass@10 functional accuracy on RTLLM by up to 17.19% compared to conventional training strategies. These findings highlight speculative decoding as a promising approach to bridge the quality gap in code generation for specialized programming languages.", "authors": ["Changran Xu", "Yi Liu", "Yunhao Zhou", "Shan Huang", "Ningyi Xu", "Qiang Xu"], "categories": ["cs.LG", "cs.AR", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-18", "url": "https://arxiv.org/abs/2503.14153", "pdf_url": "https://arxiv.org/pdf/2503.14153v1", "arxiv_id": "2503.14153", "doi": "10.1109/DAC63849.2025.11133030", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Design Automation Conference", "quality_score": 0.1193}
{"id": "4577ae26bc97202f5a95529643c096ad80ed4cb0c09395ca7d7c8d4d9071b16a", "sources": ["arxiv", "semantic_scholar"], "title": "ML-SpecQD: Multi-Level Speculative Decoding with Quantized Drafts", "abstract": "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-generated tokens. The efficacy of this method heavily relies on the acceptance ratio of the draft-generated tokens and the relative token throughput of the draft versus the target model. Nevertheless, an efficient SD pipeline requires pre-training and aligning the draft model to the target model, making it impractical for LLM inference in a plug-and-play fashion. In this work, we propose using MXFP4 models as drafts in a plug-and-play fashion since the MXFP4 Weight-Only-Quantization (WOQ) merely direct-casts the BF16 target model weights to MXFP4. In practice, our plug-and-play solution gives speedups up to 2x over the BF16 baseline. Then we pursue an opportunity for further acceleration: the MXFP4 draft token generation itself can be accelerated via speculative decoding by using yet another smaller draft. We call our method ML-SpecQD: Multi-Level Speculative Decoding with Quantized Drafts since it recursively applies speculation for accelerating the draft-token generation. Combining Multi-Level Speculative Decoding with MXFP4 Quantized Drafts we outperform state-of-the-art speculative decoding, yielding speedups up to 2.72x over the BF16 baseline.", "authors": ["Evangelos Georganas", "Dhiraj Kalamkar", "Alexander Kozlov", "Alexander Heinecke"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-17", "url": "https://arxiv.org/abs/2503.13565", "pdf_url": "https://arxiv.org/pdf/2503.13565v1", "arxiv_id": "2503.13565", "doi": "10.48550/arXiv.2503.13565", "citation_count": 7, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2258}
{"id": "c6bb892be526c541d84bfbe0e37fe50302d8fc407564b41896bd22a5398705ec", "sources": ["arxiv", "semantic_scholar"], "title": "Collaborative Speculative Inference for Efficient LLM Inference Serving", "abstract": "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 preserving generation quality. However, existing speculative methods face critical challenges, including inefficient resource utilization and limited draft acceptance, which constrain their scalability and overall effectiveness. To overcome these obstacles, we present CoSine, a novel speculative inference system that decouples sequential speculative decoding from parallel verification, enabling efficient collaboration among multiple nodes. Specifically, CoSine routes inference requests to specialized drafters based on their expertise and incorporates a confidence-based token fusion mechanism to synthesize outputs from cooperating drafters, ensuring high-quality draft generation. Additionally, CoSine dynamically orchestrates the execution of speculative decoding and verification in a pipelined manner, employing batch scheduling to selectively group requests and adaptive speculation control to minimize idle periods. By optimizing parallel workflows through heterogeneous node collaboration, CoSine balances draft generation and verification throughput in real-time, thereby maximizing resource utilization. Experimental results demonstrate that CoSine achieves superior performance compared to state-of-the-art speculative approaches. Notably, with equivalent resource costs, CoSine achieves up to a 23.2% decrease in latency and a 32.5% increase in throughput compared to baseline methods.", "authors": ["Luyao Gao", "Jianchun Liu", "Hongli Xu", "Xichong Zhang", "Yunming Liao", "Liusheng Huang"], "categories": ["cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-13", "url": "https://arxiv.org/abs/2503.10325", "pdf_url": "https://arxiv.org/pdf/2503.10325v2", "arxiv_id": "2503.10325", "doi": "10.48550/arXiv.2503.10325", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1945}
{"id": "b8f12d91ebe3f0c05c86155a77f27bdcb3ddd320bd9279bcf454f80518e68ff0", "sources": ["arxiv", "semantic_scholar"], "title": "Gumiho: A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding", "abstract": "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 sequence and accepts aligned tokens, enabling efficient multi-token generation. However, existing methods assume that all tokens within a sequence are equally important, employing identical head structures and relying on a single-generation paradigm, either serial or parallel. To this end, we theoretically demonstrate that initial tokens in the draft sequence are more important than later ones. Building on this insight, we propose Gumiho, a hybrid model combining serial and parallel heads. Specifically, given the critical importance of early tokens, we employ a sophisticated Transformer architecture for the early draft heads in a serial configuration to improve accuracy. For later tokens, we utilize multiple lightweight MLP heads operating in parallel to enhance efficiency. By allocating more advanced model structures and longer running times to the early heads, Gumiho achieves improved overall performance. The experimental results demonstrate that our method outperforms existing approaches, fully validating its effectiveness.", "authors": ["Jinze Li", "Yixing Xu", "Haiduo Huang", "Xuanwu Yin", "Dong Li", "Edith C. H. Ngai", "Emad Barsoum"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-13", "url": "https://arxiv.org/abs/2503.10135", "pdf_url": "https://arxiv.org/pdf/2503.10135v2", "arxiv_id": "2503.10135", "doi": "10.48550/arXiv.2503.10135", "citation_count": 8, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/AMD-AIG-AIMA/Gumiho", "venue": "International Conference on Machine Learning", "quality_score": 0.2386}
{"id": "f5fe03837757d5ebda52b54eb542048f91be0a069e3c40814d7a878c92c61d94", "sources": ["arxiv", "semantic_scholar"], "title": "Training Domain Draft Models for Speculative Decoding: Best Practices and Insights", "abstract": "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 significantly due to domain shift. In this work, we systematically investigate knowledge distillation techniques for training domain draft models to improve their speculation accuracy. We compare white-box and black-box distillation approaches and explore their effectiveness in various data accessibility scenarios, including historical user queries, curated domain data, and synthetically generated alignment data. Our experiments across Function Calling, Biology, and Chinese domains show that offline distillation consistently outperforms online distillation by 11% to 25%, white-box distillation surpasses black-box distillation by 2% to 10%, and data scaling trends hold across domains. Additionally, we find that synthetic data can effectively align draft models and achieve 80% to 93% of the performance of training on historical user queries. These findings provide practical guidelines for training domain-specific draft models to improve speculative decoding efficiency.", "authors": ["Fenglu Hong", "Ravi Raju", "Jonathan Lingjie Li", "Bo Li", "Urmish Thakker", "Avinash Ravichandran", "Swayambhoo Jain", "Changran Hu"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-10", "url": "https://arxiv.org/abs/2503.07807", "pdf_url": "https://arxiv.org/pdf/2503.07807v2", "arxiv_id": "2503.07807", "doi": "10.48550/arXiv.2503.07807", "citation_count": 7, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2258}
{"id": "fce8f76ecb8379a2d747f62f47434378a61e60ac232ab53372dec2a588995a10", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding for Multi-Sample Inference", "abstract": "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 dynamically analyzing structural patterns across parallel reasoning paths through a probabilistic aggregation mechanism, it identifies consensus token sequences that align with the decoding distribution. Evaluations on mathematical reasoning benchmarks demonstrate a substantial improvement in draft acceptance rates over baselines, while reducing the latency in draft token construction. This work establishes a paradigm shift for efficient multi-sample inference, enabling seamless integration of speculative decoding with sampling-based reasoning techniques.", "authors": ["Yiwei Li", "Jiayi Shi", "Shaoxiong Feng", "Peiwen Yuan", "Xinglin Wang", "Yueqi Zhang", "Ji Zhang", "Chuyi Tan", "Boyuan Pan", "Yao Hu", "Kan Li"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-07", "url": "https://arxiv.org/abs/2503.05330", "pdf_url": "https://arxiv.org/pdf/2503.05330v1", "arxiv_id": "2503.05330", "doi": "10.48550/arXiv.2503.05330", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.1747}
{"id": "071ec866b67bb04e3d9b888ea0938dd3edafeb2fef3f1a3589583ab5cae1e898", "sources": ["arxiv", "semantic_scholar"], "title": "AdaSpec: Adaptive Speculative Decoding for Fast, SLO-Aware Large Language Model Serving", "abstract": "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 accelerate LLM inference. However, existing speculative decoding solutions often fail to adapt to fluctuating workloads and dynamic system environments, resulting in impaired performance and SLO violations. In this paper, we introduce AdaSpec, an efficient LLM inference system that dynamically adjusts speculative strategies according to real-time request loads and system configurations. AdaSpec proposes a theoretical model to analyze and predict the efficiency of speculative strategies across diverse scenarios. Additionally, it implements intelligent drafting and verification algorithms to maximize performance while ensuring high SLO attainment. Experimental results on real-world LLM service traces demonstrate that AdaSpec consistently meets SLOs and achieves substantial performance improvements, delivering up to 66% speedup compared to state-of-the-art speculative inference systems. The source code is publicly available at https://github.com/cerebellumking/AdaSpec", "authors": ["Kaiyu Huang", "Hao Wu", "Zhubo Shi", "Han Zou", "Minchen Yu", "Qingjiang Shi"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-07", "url": "https://arxiv.org/abs/2503.05096", "pdf_url": "https://arxiv.org/pdf/2503.05096v2", "arxiv_id": "2503.05096", "doi": "10.1145/3772052.3772239", "citation_count": 14, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/cerebellumking/AdaSpec", "venue": "ACM Symposium on Cloud Computing", "quality_score": 0.301}
{"id": "a35d039489f2b418130459207e92848aa90a3be33beefaf121caf3ab9a3fe7c7", "sources": ["arxiv", "semantic_scholar"], "title": "RASD: Retrieval-Augmented Speculative Decoding", "abstract": "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 draft model's small size and limited training data, model-based speculative decoding frequently becomes less effective in out-of-domain scenarios. Additionally, the time cost of the drafting phase results in a low upper limit on acceptance length during the verification step, limiting overall efficiency. This paper proposes RASD (Retrieval-Augmented Speculative Decoding), which adopts retrieval methods to enhance model-based speculative decoding. We introduce tree pruning and tree fusion to achieve this. Specifically, we develop a pruning method based on the draft model's probability distribution to construct the optimal retrieval tree. Second, we employ the longest prefix matching algorithm to merge the tree generated by the draft model with the retrieval tree, resulting in a unified tree for verification. Experimental results demonstrate that RASD achieves state-of-the-art inference acceleration across tasks such as DocQA, Summary, Code, and In-Domain QA. Moreover, RASD exhibits strong scalability, seamlessly integrating with various speculative decoding approaches, including both generation-based and retrieval-based methods.", "authors": ["Guofeng Quan", "Wenfeng Feng", "Chuzhan Hao", "Guochao Jiang", "Yuewei Zhang", "Hao Wang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-05", "url": "https://arxiv.org/abs/2503.03434", "pdf_url": "https://arxiv.org/pdf/2503.03434v1", "arxiv_id": "2503.03434", "doi": "10.48550/arXiv.2503.03434", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.1945}
{"id": "f302273702346aa266f44960e65f51a83479eda973110564d5417efef338cd80", "sources": ["arxiv", "semantic_scholar"], "title": "DuoDecoding: Hardware-aware Heterogeneous Speculative Decoding with Dynamic Multi-Sequence Drafting", "abstract": "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 distribution fidelity. Nevertheless, the draft model introduces additional computational overhead, becoming a performance bottleneck and increasing the time to first token (TTFT). Previous approaches to mitigate draft model overhead have primarily relied on heuristics and generally failed to match the quality of the draft language models. To address these challenges, we propose DuoDecoding, a novel approach that strategically deploys the draft and target models on the CPU and GPU respectively, enabling parallel decoding while preserving draft quality. Our method incorporates a hardware-aware optimal draft budget to minimize idle times and employs dynamic multi-sequence drafting to enhance draft quality. Extensive experiments across seven tasks show that DuoDecoding achieves up to 2.61x speedup in generation latency, while reducing TTFT to 83% of that in conventional speculative decoding. The Code is available at https://github.com/KaiLv69/DuoDecoding.", "authors": ["Kai Lv", "Honglin Guo", "Qipeng Guo", "Xipeng Qiu"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-02", "url": "https://arxiv.org/abs/2503.00784", "pdf_url": "https://arxiv.org/pdf/2503.00784v1", "arxiv_id": "2503.00784", "doi": "10.48550/arXiv.2503.00784", "citation_count": 5, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/KaiLv69/DuoDecoding", "venue": "arXiv.org", "quality_score": 0.1945}
{"id": "3fb7815a944b7f3265c6c6d00920b13f8266319f0f85a240b72f7e82e192cea9", "sources": ["arxiv", "semantic_scholar"], "title": "Tutorial Proposal: Speculative Decoding for Efficient LLM Inference", "abstract": "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 decoding in LLMs. At each decoding step, SD efficiently drafts several future tokens and then verifies them in parallel. This approach, unlike traditional autoregressive decoding, facilitates the simultaneous decoding of multiple tokens per step, thereby achieving promising 2x-4x speedups in LLM inference while maintaining original distributions. This tutorial delves into the latest techniques in SD, including draft model architectures and verification strategies. Additionally, it explores the acceleration potential and future research directions in this promising field. We aim for this tutorial to elucidate the current research landscape and offer insights for researchers interested in Speculative Decoding, ultimately contributing to more efficient LLM inference.", "authors": ["Heming Xia", "Cunxiao Du", "Yongqi Li", "Qian Liu", "Wenjie Li"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-03-01", "url": "https://arxiv.org/abs/2503.00491", "pdf_url": "https://arxiv.org/pdf/2503.00491v1", "arxiv_id": "2503.00491", "doi": "10.48550/arXiv.2503.00491", "citation_count": 4, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1747}
{"id": "074ce10412b8df85dd200d973989c39a57492c44232da19b87e15b61a50e8fe6", "sources": ["arxiv", "semantic_scholar"], "title": "Fuzzy Speculative Decoding for a Tunable Accuracy-Runtime Tradeoff", "abstract": "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 exchange for further inference speed gains. To address these limitations, we introduce Fuzzy Speculative Decoding (FSD) - a decoding algorithm that generalizes SD by accepting candidate tokens based on the divergences between the target and draft model distributions. By allowing for controlled divergence from the target model, FSD enables users to flexibly trade generation quality for inference speed. Across several benchmarks, our method is able to achieve significant runtime improvements of over 5 tokens per second faster than SD at only an approximate 2% absolute reduction in benchmark accuracy. In many cases, FSD is even able to match SD benchmark accuracy at over 2 tokens per second faster, demonstrating that distributional equivalence is not necessary to maintain target model performance. Furthermore, FSD can be seamlessly integrated into existing SD extensions; we demonstrate this by applying FSD to EAGLE-2, greatly enhancing this existing extension's efficiency while allowing it to leverage FSD's tunable quality-speed trade-off.", "authors": ["Maximilian Holsman", "Yukun Huang", "Bhuwan Dhingra"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-28", "url": "https://arxiv.org/abs/2502.20704", "pdf_url": "https://arxiv.org/pdf/2502.20704v4", "arxiv_id": "2502.20704", "doi": "10.48550/arXiv.2502.20704", "citation_count": 7, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.2386}
{"id": "be152fba1c0d31be3a4b83bf7df5e47c44487f8098c7f93a08c3b806e2a9deca", "sources": ["arxiv", "semantic_scholar"], "title": "RAPID: Long-Context Inference with Retrieval-Augmented Speculative Decoding", "abstract": "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) traditionally accelerates inference using smaller draft models, its effectiveness diminishes substantially in long-context scenarios due to memory-bound KV cache operations. We introduce Retrieval-Augmented Speculative Decoding (RAPID), which leverages RAG for both accelerating and enhancing generation quality in long-context inference. RAPID introduces the RAG drafter-a draft LLM operating on shortened retrieval contexts-to speculate on the generation of long-context target LLMs. Our approach enables a new paradigm where same-scale or even larger LLMs can serve as RAG drafters while maintaining computational efficiency. To fully leverage the potentially superior capabilities from stronger RAG drafters, we develop an inference-time knowledge transfer that enriches the target distribution by RAG. Extensive experiments on the LLaMA-3.1 and Qwen2.5 backbones demonstrate that RAPID effectively integrates the strengths of both RAG and long-context LLMs, achieving significant performance improvements (e.g., from 39.33 to 42.83 on InfiniteBench for LLaMA-3.1-8B) with more than 2x speedups for long-context inference. Our analyses also reveal the robustness of RAPID across various context lengths and retrieval quality.", "authors": ["Guanzheng Chen", "Qilong Feng", "Jinjie Ni", "Xin Li", "Michael Qizhe Shieh"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-27", "url": "https://arxiv.org/abs/2502.20330", "pdf_url": "https://arxiv.org/pdf/2502.20330v2", "arxiv_id": "2502.20330", "doi": "10.48550/arXiv.2502.20330", "citation_count": 9, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.25}
{"id": "89fa649a8661950248303467c7c21cfb8832cf12741d65607f39dff31a4813f7", "sources": ["arxiv", "semantic_scholar"], "title": "Towards Optimal Multi-draft Speculative Decoding", "abstract": "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 the target LLM verifies them in parallel, ensuring that the final output conforms to the target model distribution. The two main design choices in MDSD are the draft sampling method and the verification algorithm. For a fixed draft sampling method, the optimal acceptance rate is a solution to an optimal transport problem, but the complexity of this problem makes it difficult to solve for the optimal acceptance rate and measure the gap between existing verification algorithms and the theoretical upper bound. This paper discusses the dual of the optimal transport problem, providing a way to efficiently compute the optimal acceptance rate. For the first time, we measure the theoretical upper bound of MDSD efficiency for vocabulary sizes in the thousands and quantify the gap between existing verification algorithms and this bound. We also compare different draft sampling methods based on their optimal acceptance rates. Our results show that the draft sampling method strongly influences the optimal acceptance rate, with sampling without replacement outperforming sampling with replacement. Additionally, existing verification algorithms do not reach the theoretical upper bound for both without replacement and with replacement sampling. Our findings suggest that carefully designed draft sampling methods can potentially improve the optimal acceptance rate and enable the development of verification algorithms that closely match the theoretical upper bound.", "authors": ["Zhengmian Hu", "Tong Zheng", "Vignesh Viswanathan", "Ziyi Chen", "Ryan A. Rossi", "Yihan Wu", "Dinesh Manocha", "Heng Huang"], "categories": ["cs.CL", "cs.DS"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-26", "url": "https://arxiv.org/abs/2502.18779", "pdf_url": "https://arxiv.org/pdf/2502.18779v1", "arxiv_id": "2502.18779", "doi": "10.48550/arXiv.2502.18779", "citation_count": 15, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.301}
{"id": "952dfb175e181df3a1732eab1cae9d6baa147bc74cbb829f9986464cf48caf89", "sources": ["arxiv", "semantic_scholar"], "title": "Harnessing Multiple Large Language Models: A Survey on LLM Ensemble", "abstract": "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 strengths and out-of-the-box usability, has profoundly advanced the field of LLM Ensemble. This paper presents the first systematic review of recent developments in LLM Ensemble. First, we introduce our taxonomy of LLM Ensemble and discuss several related research problems. Then, we provide a more in-depth classification of the methods under the broad categories of \"ensemble-before-inference, ensemble-during-inference, ensemble-after-inference'', and review all relevant methods. Finally, we introduce related benchmarks and applications, summarize existing studies, and suggest several future research directions. A curated list of papers on LLM Ensemble is available at https://github.com/junchenzhi/Awesome-LLM-Ensemble.", "authors": ["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"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-25", "url": "https://arxiv.org/abs/2502.18036", "pdf_url": "https://arxiv.org/pdf/2502.18036v6", "arxiv_id": "2502.18036", "doi": "10.48550/arXiv.2502.18036", "citation_count": 108, "influential_citation_count": 4, "has_code": true, "code_url": "https://github.com/junchenzhi/Awesome-LLM-Ensemble", "venue": "arXiv.org", "quality_score": 0.5094}
{"id": "24eb4df97f0ffef16ad13e5d36c738bdc02c7baeedda2e94dfc3025d191acc8d", "sources": ["arxiv", "semantic_scholar"], "title": "LongSpec: Long-Context Lossless Speculative Decoding with Efficient Drafting and Verification", "abstract": "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 technique compared to lossy alternatives such as quantization and model cascades. However, most state-of-the-art SD methods are trained on short texts (typically fewer than 4k tokens), making them unsuitable for long-context scenarios. Specifically, adapting these methods to long contexts presents three key challenges: (1) the excessive memory demands posed by draft models due to large Key-Value (KV) cache; (2) performance degradation resulting from the mismatch between short-context training and long-context inference; and (3) inefficiencies in tree attention mechanisms when managing long token sequences. This work introduces LongSpec, a framework that addresses these challenges through three core innovations: a memory-efficient draft model with a constant-sized KV cache; novel position indices that mitigate the training-inference mismatch; and an attention aggregation strategy that combines fast prefix computation with standard tree attention to enable efficient decoding. Experimental results confirm the effectiveness of LongSpec, achieving up to a 3.26x speedup over strong Flash Attention baselines across five long-context understanding datasets, as well as a 2.25x reduction in wall-clock time on the AIME24 long reasoning task with the QwQ model, demonstrating significant latency improvements for long-context applications. The code is available at https://github.com/sail-sg/LongSpec.", "authors": ["Penghui Yang", "Cunxiao Du", "Fengzhuo Zhang", "Haonan Wang", "Tianyu Pang", "Chao Du", "Bo An"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-24", "url": "https://arxiv.org/abs/2502.17421", "pdf_url": "https://arxiv.org/pdf/2502.17421v4", "arxiv_id": "2502.17421", "doi": null, "citation_count": 11, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/sail-sg/LongSpec", "venue": null, "quality_score": 0.2698}
{"id": "3c6d0366d983fe0aac9a6270707bf3d2f45a5d0a85d4e204b7bc6c7cae3ec2e2", "sources": ["arxiv", "semantic_scholar"], "title": "Capability Instruction Tuning: A New Paradigm for Dynamic LLM Routing", "abstract": "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-performing LLM for each instruction to achieve better overall performance. We develop a new paradigm, constructing capability instructions with model capability representation, user instruction, and performance inquiry prompts to assess the performance. To learn from capability instructions, we introduce a new end-to-end framework called Model Selection with Aptitude Test (Model-SAT), which generates positive and negative samples based on what different models perform well or struggle with. Model-SAT uses a model capability encoder that extends its model representation to a lightweight LLM. Our experiments show that Model-SAT understands the performance dimensions of candidate models and provides the probabilities of their capability to handle various instructions. Additionally, during deployment, a new model can quickly infer its aptitude test results across 50 tasks, each with 20 shots. Model-SAT performs state-of-the-art model routing without candidate inference and in real-world new model-released scenarios. The code is available at https://github.com/Now-Join-Us/CIT-LLM-Routing", "authors": ["Yi-Kai Zhang", "De-Chuan Zhan", "Han-Jia Ye"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-24", "url": "https://arxiv.org/abs/2502.17282", "pdf_url": "https://arxiv.org/pdf/2502.17282v1", "arxiv_id": "2502.17282", "doi": "10.48550/arXiv.2502.17282", "citation_count": 24, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/Now-Join-Us/CIT-LLM-Routing", "venue": "AAAI Conference on Artificial Intelligence", "quality_score": 0.3495}
{"id": "b5c2fbb02569328ef6a10763215a3feb340618baa30203e05d26dcc177fdfa91", "sources": ["arxiv", "semantic_scholar"], "title": "CORAL: Learning Consistent Representations across Multi-step Training with Lighter Speculative Drafter", "abstract": "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-step training strategy, but the complex inputs of different training steps make it harder for the draft model to converge. To address this, we propose CORAL, a novel framework that improves both accuracy and efficiency in speculative drafting. CORAL introduces Cross-Step Representation Alignment, a method that enhances consistency across multiple training steps, significantly improving speculative drafting performance. Additionally, we identify the LM head as a major bottleneck in the inference speed of the draft model. We introduce a weight-grouping mechanism that selectively activates a subset of LM head parameters during inference, substantially reducing the latency of the draft model. We evaluate CORAL on three LLM families and three benchmark datasets, achieving speedup ratios of 2.50x-4.07x, outperforming state-of-the-art methods such as EAGLE-2 and HASS. Our results demonstrate that CORAL effectively mitigates training-inference misalignment and delivers significant speedup for modern LLMs with large vocabularies.", "authors": ["Yepeng Weng", "Dianwen Mei", "Huishi Qiu", "Xujie Chen", "Li Liu", "Jiang Tian", "Zhongchao Shi"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-24", "url": "https://arxiv.org/abs/2502.16880", "pdf_url": "https://arxiv.org/pdf/2502.16880v3", "arxiv_id": "2502.16880", "doi": "10.48550/arXiv.2502.16880", "citation_count": 9, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.25}
{"id": "a1bd5edafcca4a9bffffa182f00eb7f94731c91888e5fec7648d6a1eff63c632", "sources": ["arxiv", "semantic_scholar"], "title": "TETRIS: Optimal Draft Token Selection for Batch Speculative Decoding", "abstract": "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 accepted when verified in parallel, resulting in fewer rejected tokens and hence less wasted computing resources. Such an effective resource utilization to achieve fast inference in large language models (LLMs) is especially important to service providers with limited inference capacity. Compared to baseline speculative decoding, TETRIS yields a consistently higher acceptance rate and more effective utilization of the limited inference capacity. We show theoretically and empirically that TETRIS outperforms baseline speculative decoding and existing methods that dynamically select draft tokens, leading to a more efficient batch inference in LLMs.", "authors": ["Zhaoxuan Wu", "Zijian Zhou", "Arun Verma", "Alok Prakash", "Daniela Rus", "Bryan Kian Hsiang Low"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-21", "url": "https://arxiv.org/abs/2502.15197", "pdf_url": "https://arxiv.org/pdf/2502.15197v2", "arxiv_id": "2502.15197", "doi": "10.48550/arXiv.2502.15197", "citation_count": 7, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.2258}
{"id": "0d44ff6bbc01fae22dd8f5ed2037e065addd39824e739557cd5d076bf87f62a9", "sources": ["arxiv", "semantic_scholar"], "title": "DReSD: Dense Retrieval for Speculative Decoding", "abstract": "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 tokens from a non-parametric datastore. Sparse retrieval (REST), which operates on the surface form of strings, is currently the dominant paradigm due to its simplicity and scalability. However, its effectiveness is limited due to the usage of short contexts and exact string matching. Instead, we introduce Dense Retrieval for Speculative Decoding (DReSD), a novel framework that uses approximate nearest neighbour search with contextualised token embeddings to retrieve the most semantically relevant token sequences for SD. Extensive experiments show that DReSD achieves (on average) 87% higher acceptance rates, 65% longer accepted tokens and 19% faster generation speeds compared to sparse retrieval (REST).", "authors": ["Milan Gritta", "Huiyin Xue", "Gerasimos Lampouras"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-21", "url": "https://arxiv.org/abs/2502.15572", "pdf_url": "https://arxiv.org/pdf/2502.15572v2", "arxiv_id": "2502.15572", "doi": "10.48550/arXiv.2502.15572", "citation_count": 5, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.1945}
{"id": "4eab278e79a4757612dc66da38cabda2616e826342f82a6b1e67e316181cac2c", "sources": ["arxiv", "semantic_scholar"], "title": "BP-GPT: Auditory Neural Decoding Using fMRI-prompted LLM", "abstract": "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 the LLM in the mapping of fMRI-to-text, leaving space for the exploration of the LLM in auditory decoding. In this paper, we introduce a novel method, the Brain Prompt GPT (BP-GPT). By using the brain representation that is extracted from the fMRI as a prompt, our method can utilize GPT-2 to decode fMRI signals into stimulus text. Further, we introduce the text prompt and align the fMRI prompt to it. By introducing the text prompt, our BP-GPT can extract a more robust brain prompt and promote the decoding of pre-trained LLM. We evaluate our BP-GPT on the open-source auditory semantic decoding dataset and achieve a significant improvement up to 4.61 on METEOR and 2.43 on BERTScore across all the subjects compared to the state-of-the-art method. The experimental results demonstrate that using brain representation as a prompt to further drive LLM for auditory neural decoding is feasible and effective. The code is available at https://github.com/1994cxy/BP-GPT.", "authors": ["Xiaoyu Chen", "Changde Du", "Che Liu", "Yizhe Wang", "Huiguang He"], "categories": ["cs.HC", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-21", "url": "https://arxiv.org/abs/2502.15172", "pdf_url": "https://arxiv.org/pdf/2502.15172v1", "arxiv_id": "2502.15172", "doi": "10.1109/ICASSP49660.2025.10890142", "citation_count": 6, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/1994cxy/BP-GPT", "venue": "IEEE International Conference on Acoustics, Speech, and Signal Processing", "quality_score": 0.2113}
{"id": "5295c397ed4254f2f9bd8107e17a57f01462fd06dc5a8df7127f2c4df53d4811", "sources": ["arxiv", "semantic_scholar"], "title": "C2T: A Classifier-Based Tree Construction Method in Speculative Decoding", "abstract": "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, static tree, and dynamic tree approaches, have limitations in accurately preparing candidate token trees for verification. We propose a novel method named C2T that adopts a lightweight classifier to generate and prune token trees dynamically. Our classifier considers additional feature variables beyond the commonly used joint probability to predict the confidence score for each draft token to determine whether it is the candidate token for verification. This method outperforms state-of-the-art (SOTA) methods such as EAGLE-2 on multiple benchmarks, by reducing the total number of candidate tokens by 25% while maintaining or even improving the acceptance length.", "authors": ["Feiye Huo", "Jianchao Tan", "Kefeng Zhang", "Xunliang Cai", "Shengli Sun"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-19", "url": "https://arxiv.org/abs/2502.13652", "pdf_url": "https://arxiv.org/pdf/2502.13652v1", "arxiv_id": "2502.13652", "doi": "10.48550/arXiv.2502.13652", "citation_count": 7, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2258}
{"id": "6e8d4b4315c6cfdde1b0c0d38a09e5adff35cc7c7bf71e46e9b73b1dc7e7b110", "sources": ["arxiv", "semantic_scholar"], "title": "GRIFFIN: Effective Token Alignment for Faster Speculative Decoding", "abstract": "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 that incorporates a token-alignable training strategy and a token-alignable draft model to mitigate misalignment. The training strategy employs a loss masking mechanism to exclude highly misaligned tokens during training, preventing them from negatively impacting the draft model's optimization. The token-alignable draft model introduces input tokens to correct inconsistencies in generated features. Experiments on LLaMA, Vicuna, Qwen and Mixtral models demonstrate that GRIFFIN achieves an average acceptance length improvement of over 8% and a speedup ratio exceeding 7%, outperforming current speculative decoding state-of-the-art methods. Our code and GRIFFIN's draft models are released publicly in https://github.com/hsj576/GRIFFIN.", "authors": ["Shijing Hu", "Jingyang Li", "Xingyu Xie", "Zhihui Lu", "Kim-Chuan Toh", "Pan Zhou"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-16", "url": "https://arxiv.org/abs/2502.11018", "pdf_url": "https://arxiv.org/pdf/2502.11018v3", "arxiv_id": "2502.11018", "doi": "10.48550/arXiv.2502.11018", "citation_count": 10, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/hsj576/GRIFFIN", "venue": "arXiv.org", "quality_score": 0.2603}
{"id": "1f825cf4c9f844e424a1f12d1259bef23b2aeee8014e67ff86bb29117aa1ec6c", "sources": ["arxiv", "semantic_scholar"], "title": "CopySpec: Accelerating LLMs with Speculative Copy-and-Paste Without Compromising Quality", "abstract": "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 same tokens will follow, enabling seamless copying without compromising output quality and without requiring additional GPU memory. To evaluate the effectiveness of our approach, we conducted experiments using seven LLMs and five datasets: MT-Bench, CNN/DM, GSM8K, HumanEval, and our newly created dataset, MT-Redundant. MT-Redundant, introduced in this paper, transforms the second turn of MT-Bench into a request for variations of the first turn's answer, simulating real-world scenarios where users request modifications to prior responses. Our results demonstrate significant speed-ups: up to 2.35x on CNN/DM, 3.08x on the second turn of select MT-Redundant categories, and 2.66x on the third turn of GSM8K's self-correction tasks. Importantly, we show that CopySpec integrates seamlessly with speculative decoding, yielding an average 49% additional speed-up over speculative decoding for the second turn of MT-Redundant across all eight categories. While LLMs, even with speculative decoding, suffer from slower inference as context size grows, CopySpec leverages larger contexts to accelerate inference, making it a faster complementary solution. Our code and dataset are publicly available at https://github.com/RazvanDu/CopySpec.", "authors": ["Razvan-Gabriel Dumitru", "Minglai Yang", "Vikas Yadav", "Mihai Surdeanu"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-13", "url": "https://arxiv.org/abs/2502.08923", "pdf_url": "https://arxiv.org/pdf/2502.08923v2", "arxiv_id": "2502.08923", "doi": "10.48550/arXiv.2502.08923", "citation_count": 5, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/RazvanDu/CopySpec", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.1945}
{"id": "d206bd2fdaa2950f31fe3b71cf37e549f4e6405f816fb9733e04ef4e457536f0", "sources": ["arxiv", "semantic_scholar"], "title": "Speculate, then Collaborate: Fusing Knowledge of Language Models during Decoding", "abstract": "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 a novel Collaborative Speculative Decoding (CoSD) algorithm that enables efficient LLM knowledge fusion at test time without requiring additional model training. CoSD employs a draft model to generate initial sequences and an easy-to-learn rule or decision tree to decide when to invoke an assistant model to improve these drafts. CoSD not only enhances knowledge fusion but also improves inference efficiency, is transferable across domains and models, and offers greater explainability. Experimental results demonstrate that CoSD improves accuracy by up to 10\\% across benchmarks compared to existing methods, providing a scalable and effective solution for LLM-based applications", "authors": ["Ziyao Wang", "Muneeza Azmat", "Ang Li", "Raya Horesh", "Mikhail Yurochkin"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-11", "url": "https://arxiv.org/abs/2502.08020", "pdf_url": "https://arxiv.org/pdf/2502.08020v2", "arxiv_id": "2502.08020", "doi": "10.48550/arXiv.2502.08020", "citation_count": 8, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.2386}
{"id": "b8317a46b5953116904f8671806482505b9654160e1b47e86ce3348b88a6f007", "sources": ["arxiv", "semantic_scholar"], "title": "Jakiro: Boosting Speculative Decoding with Decoupled Multi-Head via MoE", "abstract": "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 multiple candidates are generated at each step. We identify a key limitation in this approach: the candidates at the same step are derived from the same representation, limiting diversity and reducing overall effectiveness. To address this, we propose Jakiro, leveraging Mixture of Experts (MoE), where independent experts generate diverse predictions, effectively decoupling correlations among candidates. Furthermore, we introduce a hybrid inference strategy, combining autoregressive decoding for initial tokens with parallel decoding for subsequent stages, and enhance the latter with contrastive mechanism in features to improve accuracy. Our method significantly boosts prediction accuracy and achieves higher inference speedups. Extensive experiments across diverse models validate the effectiveness and robustness of our approach, establishing a new SOTA in speculative decoding. Our codes are available at https://github.com/haiduo/Jakiro.", "authors": ["Haiduo Huang", "Fuwei Yang", "Zhenhua Liu", "Yixing Xu", "Jinze Li", "Yang Liu", "Xuanwu Yin", "Dong Li", "Pengju Ren", "Emad Barsoum"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-10", "url": "https://arxiv.org/abs/2502.06282", "pdf_url": "https://arxiv.org/pdf/2502.06282v1", "arxiv_id": "2502.06282", "doi": "10.48550/arXiv.2502.06282", "citation_count": 4, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/haiduo/Jakiro", "venue": "arXiv.org", "quality_score": 0.1747}
{"id": "e463e13b91bcd7bae9f0b54e5d4544de938051f27ff8e0ad8bc6675480e8151a", "sources": ["arxiv", "semantic_scholar"], "title": "LANTERN++: Enhancing Relaxed Speculative Decoding with Static Tree Drafting for Visual Auto-regressive Models", "abstract": "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 with dynamic tree drafting was proposed to mitigate this ambiguity, demonstrating promising results in accelerating visual AR models. However, we observe that token selection ambiguity still negatively affects dynamic tree drafting, resulting in shallow draft trees and limited acceleration. To overcome this issue, we introduce LANTERN++, a refined framework that integrates static tree drafting with a tailored relaxed acceptance condition, allowing drafts to be selected independently of low-confidence predictions. This enables the acceptance of deeper sequences, improving decoding efficiency while preserving image quality. Extensive experiments on state-of-the-art visual AR models demonstrate that LANTERN++ significantly accelerates inference, achieving up to $\\mathbf{\\times 2.56}$ speedup over standard AR decoding while maintaining high image quality. The code is publicly available at https://github.com/jadohu/LANTERN.", "authors": ["Sihwan Park", "Doohyuk Jang", "Sungyub Kim", "Souvik Kundu", "Eunho Yang"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-10", "url": "https://arxiv.org/abs/2502.06352", "pdf_url": "https://arxiv.org/pdf/2502.06352v2", "arxiv_id": "2502.06352", "doi": "10.48550/arXiv.2502.06352", "citation_count": 11, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/jadohu/LANTERN", "venue": "arXiv.org", "quality_score": 0.301}
{"id": "368cc9d45aaa33fd610dceed97284283c25b3e056fbe0b321f57087891d5b94d", "sources": ["arxiv", "semantic_scholar"], "title": "Acceleration Multiple Heads Decoding for LLM via Dynamic Tree Attention", "abstract": "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 on multiple head decoding, specifically in the context of MEDUSA. We propose a simple and low complexity strategy to generate candidates and construct the dynamic tree structure. Preliminary experiments show that the proposed method improves the decoding efficiency of multiple head decoding for LLMs while maintaining the generation quality. This result demonstrates the potential for improvement of multiple head decoding in candidate generation.", "authors": ["Zhendong Zhang"], "categories": ["cs.CV", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-09", "url": "https://arxiv.org/abs/2502.05947", "pdf_url": "https://arxiv.org/pdf/2502.05947v1", "arxiv_id": "2502.05947", "doi": "10.48550/arXiv.2502.05947", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.0516}
{"id": "ae671a898c0c36ef5f1d48b51eb9b634636be5417f5db2525bb4f1635257ca2a", "sources": ["arxiv", "semantic_scholar"], "title": "Lossless Acceleration of Large Language Models with Hierarchical Drafting based on Temporal Locality in Speculative Decoding", "abstract": "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 multiple tokens in a single forward pass. However, current drafting strategies usually require significant fine-tuning or have inconsistent performance across tasks. To address these challenges, we propose Hierarchy Drafting (HD), a novel lossless drafting approach that organizes various token sources into multiple databases in a hierarchical framework based on temporal locality. In the drafting step, HD sequentially accesses multiple databases to obtain draft tokens from the highest to the lowest locality, ensuring consistent acceleration across diverse tasks and minimizing drafting latency. Our experiments on Spec-Bench using LLMs with 7B and 13B parameters demonstrate that HD outperforms existing database drafting methods, achieving robust inference speedups across model sizes, tasks, and temperatures.", "authors": ["Sukmin Cho", "Sangjin Choi", "Taeho Hwang", "Jeongyeon Seo", "Soyeong Jeong", "Huije Lee", "Hoyun Song", "Jong C. Park", "Youngjin Kwon"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-08", "url": "https://arxiv.org/abs/2502.05609", "pdf_url": "https://arxiv.org/pdf/2502.05609v1", "arxiv_id": "2502.05609", "doi": "10.48550/arXiv.2502.05609", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.1945}
{"id": "c5e41008fc1f7af6788ed83e5d935319495ac0bb0073831f53148782bbb3f444", "sources": ["arxiv", "semantic_scholar"], "title": "Speeding up Speculative Decoding via Sequential Approximate Verification", "abstract": "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 parallel calls to the target LLM for verification prevent SD from achieving even lower latencies. We propose SPRINTER, which utilizes a low-complexity verifier trained to predict if tokens generated from a draft LLM would be accepted by the target LLM. By performing sequential approximate verification, SPRINTER does not require verification by the target LLM and is only invoked when a token is deemed unacceptable. This reduces the number of calls to the larger LLM, achieving further speedups and lower computation cost. We present a theoretical analysis of SPRINTER, examining the statistical properties of the generated tokens, as well as the expected reduction in latency as a function of the verifier. We evaluate SPRINTER on several datasets and model pairs, demonstrating that approximate verification can still maintain high quality generation while further reducing latency.", "authors": ["Meiyu Zhong", "Noel Teku", "Ravi Tandon"], "categories": ["cs.LG", "cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2025-02-06", "url": "https://arxiv.org/abs/2502.04557", "pdf_url": "https://arxiv.org/pdf/2502.04557v3", "arxiv_id": "2502.04557", "doi": null, "citation_count": 10, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.2603}
{"id": "34945b27884ef4ef7fd735328da26b2598beb14bc81ec907b7f7ee3bc2eb7e15", "sources": ["arxiv", "semantic_scholar"], "title": "QuantSpec: Self-Speculative Decoding with Hierarchical Quantized KV Cache", "abstract": "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 loaded for each decoding step. While speculative decoding is a widely accepted technique to accelerate autoregressive decoding, existing methods often struggle to achieve significant speedups due to inefficient KV cache optimization strategies and result in low acceptance rates. To address these challenges, we propose a novel self-speculative decoding framework, QuantSpec, where the draft model shares the architecture of the target model but employs a hierarchical 4-bit quantized KV cache and 4-bit quantized weights for acceleration. QuantSpec maintains high acceptance rates ($>$90%) and reliably provides consistent end-to-end speedups upto $\\sim2.5\\times$, outperforming other self-speculative decoding methods that use sparse KV cache for long-context LLM inference. QuantSpec also reduces the memory requirements by $\\sim 1.3\\times$ compared to these alternatives.", "authors": ["Rishabh Tiwari", "Haocheng Xi", "Aditya Tomar", "Coleman Hooper", "Sehoon Kim", "Maxwell Horton", "Mahyar Najibi", "Michael W. Mahoney", "Kurt Keutzer", "Amir Gholami"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-05", "url": "https://arxiv.org/abs/2502.10424", "pdf_url": "https://arxiv.org/pdf/2502.10424v1", "arxiv_id": "2502.10424", "doi": "10.48550/arXiv.2502.10424", "citation_count": 18, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.3197}
{"id": "cdf165dd9931532c8527c1cab3f803fef55d8156e8835fd9dd97ead7701960fc", "sources": ["arxiv", "semantic_scholar"], "title": "EasySpec: Layer-Parallel Speculative Decoding for Efficient Multi-GPU Utilization", "abstract": "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), while the optimal TP size of the draft model is typically smaller than that of the base model, leading to GPU idling during the drafting stage. We observe that such inefficiency stems from the sequential execution of layers, which is seemingly natural but actually unnecessary. Therefore, we propose EasySpec, a layer-parallel speculation strategy that optimizes the efficiency of multi-GPU utilization. EasySpec breaks the inter-layer data dependencies in the draft model, enabling multiple layers to run simultaneously across multiple devices as 'fuzzy' speculation. After each drafting-and-verification iteration, the draft model's key-value cache is calibrated in a single forward pass, preventing long-term fuzzy-error accumulation at minimal additional latency. EasySpec is a training-free and plug-in method. We evaluated EasySpec on several mainstream open-source LLMs, using smaller versions of models from the same series as drafters. The results demonstrate that EasySpec can achieve a peak speedup of 4.17x compared to vanilla decoding, while preserving the original distributions of the base LLMs. Specifically, the drafting stage can be accelerated by up to 1.62x with a maximum speculation accuracy drop of only 7%. The code is available at https://github.com/Yize-Wu/EasySpec.", "authors": ["Yize Wu", "Ke Gao", "Ling Li", "Yanjun Wu"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-04", "url": "https://arxiv.org/abs/2502.02493", "pdf_url": "https://arxiv.org/pdf/2502.02493v2", "arxiv_id": "2502.02493", "doi": "10.48550/arXiv.2502.02493", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Yize-Wu/EasySpec", "venue": "arXiv.org", "quality_score": 0.0753}
{"id": "bce356f4f90edfc82727db7e21d64032fbfdc3d7dc82ee7c19b2e7d27ebdf109", "sources": ["arxiv", "semantic_scholar"], "title": "Fast Large Language Model Collaborative Decoding via Speculation", "abstract": "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 decoding without compromising performance. Inspired by Speculative Decoding--where a small proposal model generates tokens sequentially, and a larger target model verifies them in parallel, our approach builds on two key insights: (1) the verification distribution can be the combined distribution of both the proposal and target models, and (2) alternating each model as the proposer and verifier can further enhance efficiency. We generalize this method to collaboration among n models and theoretically prove that CoS is never slower than standard collaborative decoding, typically achieving faster speed. Extensive experiments demonstrate CoS is 1.11x-2.23x faster than standard collaborative decoding without compromising generation quality. Our code is available at https://github.com/Kamichanw/CoS/.", "authors": ["Jiale Fu", "Yuchu Jiang", "Junkai Chen", "Jiaming Fan", "Xin Geng", "Xu Yang"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-02-01", "url": "https://arxiv.org/abs/2502.01662", "pdf_url": "https://arxiv.org/pdf/2502.01662v2", "arxiv_id": "2502.01662", "doi": null, "citation_count": 10, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/Kamichanw/CoS/", "venue": "International Conference on Machine Learning", "quality_score": 0.2603}
{"id": "fd660812c5e37daa5a00032cac9b1e01b6689591a9ba340f221157a890b61d41", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating LLM Inference with Lossless Speculative Decoding Algorithms for Heterogeneous Vocabularies", "abstract": "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 same vocabulary, thus limiting the pool of possible drafters, often necessitating the training of a drafter from scratch. We present three new SD methods that remove this shared-vocabulary constraint. All three methods preserve the target distribution (i.e., they are lossless) and work with off-the-shelf models without requiring additional training or modifications. Empirically, on summarization, programming, and long-context tasks, our algorithms demonstrate significant speedups of up to 2.8x over standard autoregressive decoding. By enabling any off-the-shelf model to serve as a drafter and requiring no retraining, this work substantially broadens the applicability of the SD framework in practice.", "authors": ["Nadav Timor", "Jonathan Mamou", "Daniel Korat", "Moshe Berchansky", "Gaurav Jain", "Oren Pereg", "Moshe Wasserblat", "David Harel"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-01-31", "url": "https://arxiv.org/abs/2502.05202", "pdf_url": "https://arxiv.org/pdf/2502.05202v3", "arxiv_id": "2502.05202", "doi": "10.48550/arXiv.2502.05202", "citation_count": 19, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.3253}
{"id": "08d779b15d93e309f2d6cc1c82db177f07921093606b6c76aec4e15081bf2309", "sources": ["arxiv", "semantic_scholar"], "title": "Reward-Guided Speculative Decoding for Efficient LLM Reasoning", "abstract": "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 contrast to existing speculative decoding methods that enforce strict unbiasedness. RSD employs a process reward model to evaluate intermediate decoding steps and dynamically decide whether to invoke the target model, optimizing the trade-off between computational cost and output quality. We theoretically demonstrate that a threshold-based mixture strategy achieves an optimal balance between resource utilization and performance. Extensive evaluations on challenging reasoning benchmarks, including Olympiad-level tasks, show that RSD delivers significant efficiency gains against decoding with the target model only (up to 4.4x fewer FLOPs), while achieving significant better accuracy than parallel decoding method on average (up to +3.5). These results highlight RSD as a robust and cost-effective approach for deploying LLMs in resource-intensive scenarios. The code is available at https://github.com/BaohaoLiao/RSD.", "authors": ["Baohao Liao", "Yuhui Xu", "Hanze Dong", "Junnan Li", "Christof Monz", "Silvio Savarese", "Doyen Sahoo", "Caiming Xiong"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2025-01-31", "url": "https://arxiv.org/abs/2501.19324", "pdf_url": "https://arxiv.org/pdf/2501.19324v3", "arxiv_id": "2501.19324", "doi": "10.48550/arXiv.2501.19324", "citation_count": 104, "influential_citation_count": 11, "has_code": true, "code_url": "https://github.com/BaohaoLiao/RSD", "venue": "International Conference on Machine Learning", "quality_score": 0.5396}
{"id": "16026c717e4c81cffc61c1aa5397b09f339e633bf13be978b11ea7361c0c5935", "sources": ["arxiv", "semantic_scholar"], "title": "Judge Decoding: Faster Speculative Sampling Requires Going Beyond Model Alignment", "abstract": "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 then verified in parallel based on their likelihood under the target model. While this approach guarantees to reproduce the target output, it incurs a substantial penalty: many high-quality draft tokens are rejected, even when they represent objectively valid continuations. Indeed, we show that even powerful draft models such as GPT-4o, as well as human text cannot achieve high acceptance rates under the standard verification scheme. This severely limits the speedup potential of current speculative decoding methods, as an early rejection becomes overwhelmingly likely when solely relying on alignment of draft and target. We thus ask the following question: Can we adapt verification to recognize correct, but non-aligned replies? To this end, we draw inspiration from the LLM-as-a-judge framework, which demonstrated that LLMs are able to rate answers in a versatile way. We carefully design a dataset to elicit the same capability in the target model by training a compact module on top of the embeddings to produce ``judgements\" of the current continuation. We showcase our strategy on the Llama-3.1 family, where our 8b/405B-Judge achieves a speedup of 9x over Llama-405B, while maintaining its quality on a large range of benchmarks. These benefits remain present even in optimized inference frameworks, where our method reaches up to 141 tokens/s for 8B/70B-Judge and 129 tokens/s for 8B/405B on 2 and 8 H100s respectively.", "authors": ["Gregor Bachmann", "Sotiris Anagnostidis", "Albert Pumarola", "Markos Georgopoulos", "Artsiom Sanakoyeu", "Yuming Du", "Edgar Schönfeld", "Ali Thabet", "Jonas Kohler"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2025-01-31", "url": "https://arxiv.org/abs/2501.19309", "pdf_url": "https://arxiv.org/pdf/2501.19309v1", "arxiv_id": "2501.19309", "doi": "10.48550/arXiv.2501.19309", "citation_count": 53, "influential_citation_count": 11, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.5396}
{"id": "e9771fbdedaec9c4b0c9315a5c9380800e1deb8888e3c395a65ff585128aaae0", "sources": ["arxiv", "semantic_scholar"], "title": "Probing LLM World Models: Enhancing Guesstimation with Wisdom of Crowds Decoding", "abstract": "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 fit in a cup) to abstract predictions (e.g., the 2024 U.S. presidential election). Inspired by the social science concept of Wisdom of Crowds (WOC)- where the median of multiple estimates improves accuracy-we propose WOC decoding for LLMs. We replicate WOC effects in human participants and find that LLMs exhibit similar benefits: median aggregation across sampled responses consistently improves accuracy over greedy decoding, self-consistency decoding, and mean decoding. This suggests that LLMs encode a world model that supports approximate reasoning. Our results position guesstimation as a useful probe of LLM world knowledge and highlight WOC decoding as a strategy for enhancing LLM guesstimation performance on real-world tasks.", "authors": ["Yun-Shiuan Chuang", "Sameer Narendran", "Nikunj Harlalka", "Alexander Cheung", "Sizhe Gao", "Siddharth Suresh", "Junjie Hu", "Timothy T. Rogers"], "categories": ["cs.AI", "cs.HC"], "fields_of_study": ["Computer Science"], "published_date": "2025-01-28", "url": "https://arxiv.org/abs/2501.17310", "pdf_url": "https://arxiv.org/pdf/2501.17310v4", "arxiv_id": "2501.17310", "doi": "10.48550/arXiv.2501.17310", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.0753}
{"id": "1a68a08a344317440f59905cacc3416d121fa2e1f55fdef9124ee2e249ec03a4", "sources": ["arxiv", "semantic_scholar"], "title": "AdaServe: Accelerating Multi-SLO LLM Serving with SLO-Customized Speculative Decoding", "abstract": "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 these heterogeneous SLOs concurrently. We present AdaServe, the first LLM serving system designed to support efficient multi-SLO serving through SLO-customized speculative decoding. AdaServe formulates multi-SLO serving as a constrained optimization problem and introduces a hardware-aware algorithm that constructs a speculation tree tailored to each request's latency target. It features a speculate-select-verify pipeline that enables fine-grained control over decoding speed while maximizing system throughput. AdaServe further adapts to workload variation by dynamically adjusting speculation parameters. Evaluations across diverse workloads show that AdaServe reduces SLO violations by up to 4.3$\\times$ and improves goodput by up to 1.9$\\times$ compared to the best performing baselines, highlighting its effectiveness in multi-SLO serving.", "authors": ["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"], "categories": ["cs.CL", "cs.AI", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2025-01-21", "url": "https://arxiv.org/abs/2501.12162", "pdf_url": "https://arxiv.org/pdf/2501.12162v2", "arxiv_id": "2501.12162", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.019}
{"id": "7d8ad91c4e6b98ddaf46fdc740b4a11992dbcd603685e2efb9a6f5287bc35116", "sources": ["arxiv", "semantic_scholar"], "title": "HADES: Hardware Accelerated Decoding for Efficient Speculation in Large Language Models", "abstract": "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 (HADES), a novel approach to enhance the performance and energy efficiency of LLMs. We address the design of an LLM accelerator with hardware-level speculative decoding support, a concept not previously explored in existing literature. Our work demonstrates how speculative decoding can significantly improve the efficiency of LLM operations, paving the way for more advanced and practical applications of these models.", "authors": ["Ze Yang", "Yihong Jin", "Xinhe Xu"], "categories": ["cs.CL", "cs.AI", "cs.AR"], "fields_of_study": ["Computer Science"], "published_date": "2024-12-27", "url": "https://arxiv.org/abs/2412.19925", "pdf_url": "https://arxiv.org/pdf/2412.19925v2", "arxiv_id": "2412.19925", "doi": "10.1109/ICCEA65460.2025.11103323", "citation_count": 17, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "International Conference Civil Engineering and Architecture", "quality_score": 0.3138}
{"id": "ec419d860fedf52cfc59d1ace891358321c8f606121f74a22921619e54d752b8", "sources": ["arxiv", "semantic_scholar"], "title": "Dovetail: A CPU/GPU Heterogeneous Speculative Decoding for LLM inference", "abstract": "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 and stronger CPUs. Although techniques such as parameter offloading and partial offloading can alleviate GPU memory pressure to some extent, their effectiveness is limited due to communication latency and suboptimal hardware resource utilization. To address this issue, we propose Dovetail, a lossless inference acceleration method that leverages the complementary characteristics of heterogeneous devices and the advantages of speculative decoding. Dovetail deploys a draft model on the GPU to perform preliminary predictions, while a target model running on the CPU validates these outputs. By reducing the granularity of data transfer, Dovetail significantly minimizes communication overhead. To further improve efficiency, we optimize the draft model specifically for heterogeneous hardware environments by reducing the number of draft tokens to lower parallel verification latency, increasing model depth to enhance predictive capabilities, and introducing a Dynamic Gating Fusion (DGF) mechanism to improve the integration of feature and embedding information. We conduct comprehensive evaluations of Dovetail across various consumer-grade GPUs, covering multiple tasks and mainstream models. Experimental results on 13B models demonstrate that Dovetail achieves inference speedups ranging from 1.79x to 10.1x across different devices, while maintaining consistency and stability in the distribution of generated texts.", "authors": ["Libo Zhang", "Zhaoning Zhang", "Baizhou Xu", "Rui Li", "Zhiliang Tian", "Songzhu Mei", "Dongsheng Li"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-12-25", "url": "https://arxiv.org/abs/2412.18934", "pdf_url": "https://arxiv.org/pdf/2412.18934v2", "arxiv_id": "2412.18934", "doi": "10.48550/arXiv.2412.18934", "citation_count": 18, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.3197}
{"id": "e41a6cc89acea55efca9236afb2a0e711941c20257be26f3917dd1699486306b", "sources": ["arxiv", "semantic_scholar"], "title": "AdaEAGLE: Optimizing Speculative Decoding via Explicit Modeling of Adaptive Draft Structures", "abstract": "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 adaptive draft structures are limited by their performance, modeling approaches, and applicability. In this paper, we introduce AdaEAGLE, the first SD framework that explicitly models adaptive draft structures. AdaEAGLE leverages the Lightweight Draft Length Predictor (LDLP) module to explicitly predict the optimal number of draft tokens during inference to guide the draft model. It achieves comparable speedup results without manual thresholds and allows for deeper, more specialized optimizations. Moreover, together with threshold-based strategies, AdaEAGLE achieves a $1.62\\times$ speedup over the vanilla AR decoding and outperforms fixed-length SotA baseline while maintaining output quality.", "authors": ["Situo Zhang", "Hankun Wang", "Da Ma", "Zichen Zhu", "Lu Chen", "Kunyao Lan", "Kai Yu"], "categories": ["cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-12-25", "url": "https://arxiv.org/abs/2412.18910", "pdf_url": "https://arxiv.org/pdf/2412.18910v1", "arxiv_id": "2412.18910", "doi": "10.48550/arXiv.2412.18910", "citation_count": 11, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2698}
{"id": "3f51517c30d349c5df2d4080244480584389de5be71577e8b31691a88a807915", "sources": ["arxiv", "semantic_scholar"], "title": "Falcon: Faster and Parallel Inference of Large Language Models through Enhanced Semi-Autoregressive Drafting and Custom-Designed Decoding Tree", "abstract": "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 augment both the drafter's parallelism and output quality. Falcon incorporates the Coupled Sequential Glancing Distillation technique, which fortifies inter-token dependencies within the same block, leading to increased speculation accuracy. We offer a comprehensive theoretical analysis to illuminate the underlying mechanisms. Additionally, we introduce a Custom-Designed Decoding Tree, which permits the drafter to generate multiple tokens in a single forward pass and accommodates multiple forward passes as needed, thereby boosting the number of drafted tokens and significantly improving the overall acceptance rate. Comprehensive evaluations on benchmark datasets such as MT-Bench, HumanEval, and GSM8K demonstrate Falcon's superior acceleration capabilities. The framework achieves a lossless speedup ratio ranging from 2.91x to 3.51x when tested on the Vicuna and LLaMA2-Chat model series. These results outstrip existing speculative decoding methods for LLMs, including Eagle, Medusa, Lookahead, SPS, and PLD, while maintaining a compact drafter architecture equivalent to merely two Transformer layers.", "authors": ["Xiangxiang Gao", "Weisheng Xie", "Yiwei Xiang", "Feng Ji"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-12-17", "url": "https://arxiv.org/abs/2412.12639", "pdf_url": "https://arxiv.org/pdf/2412.12639v3", "arxiv_id": "2412.12639", "doi": "10.48550/arXiv.2412.12639", "citation_count": 24, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": "AAAI Conference on Artificial Intelligence", "quality_score": 0.3495}
{"id": "868ba97a8771ccf32395279cee99ea19ddc23f32f04ccec37bdeec3db00b72f5", "sources": ["arxiv", "semantic_scholar"], "title": "Constrained Decoding with Speculative Lookaheads", "abstract": "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 strategies such as greedy decoding are extremely efficient, but achieve very low constraint satisfaction. We propose constrained decoding with speculative lookaheads (CDSL), a technique that significantly improves upon the inference efficiency of CDLH without experiencing the drastic performance reduction seen with greedy decoding. CDSL is motivated by the recently proposed idea of speculative decoding that uses a much smaller draft LLM for generation and a larger target LLM for verification. In CDSL, the draft model is used to generate lookaheads which is verified by a combination of target LLM and task-specific reward functions. This process accelerates decoding by reducing the computational burden while maintaining strong performance. We evaluate CDSL in two constraint decoding tasks with three LLM families and achieve 2.2x to 12.15x speedup over CDLH without significant performance reduction.", "authors": ["Nishanth Nakshatri", "Shamik Roy", "Rajarshi Das", "Suthee Chaidaroon", "Leonid Boytsov", "Rashmi Gangadharaiah"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-12-09", "url": "https://arxiv.org/abs/2412.10418", "pdf_url": "https://arxiv.org/pdf/2412.10418v2", "arxiv_id": "2412.10418", "doi": "10.48550/arXiv.2412.10418", "citation_count": 8, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.2386}
{"id": "099276887181b59b379f2937f19e48d648120e383f9ba00fd89f52aeb4b1194c", "sources": ["arxiv", "semantic_scholar"], "title": "PLD+: Accelerating LLM inference by leveraging Language Model Artifacts", "abstract": "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 and fine-tuning, which limits its out-of-the-box usability. To address these challenges, we present PLD+, a suite of novel algorithms developed to accelerate the inference process of LLMs, particularly for input-guided tasks. These tasks, which include code editing, text editing, summarization, etc., often feature outputs with substantial overlap with their inputs-an attribute PLD+ is designed to exploit. PLD+ also leverages the artifacts (attention and hidden states) generated during inference to accelerate inference speed. We test our approach on five input-guided tasks and through extensive experiments we find that PLD+ outperforms all tuning-free approaches. In the greedy setting, it even outperforms the state-of-the-art tuning-dependent approach EAGLE on four of the tasks. (by a margin of upto 2.31 in terms of avg. speedup). Our approach is tuning free, does not require any additional compute and can easily be used for accelerating inference of any LLM.", "authors": ["Shwetha Somasundaram", "Anirudh Phukan", "Apoorv Saxena"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-12-02", "url": "https://arxiv.org/abs/2412.01447", "pdf_url": "https://arxiv.org/pdf/2412.01447v1", "arxiv_id": "2412.01447", "doi": "10.48550/arXiv.2412.01447", "citation_count": 13, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.2865}
{"id": "e39f7f27fca82d58bcd40d064439cde5481be1150c513714f9a42a8781beff56", "sources": ["arxiv", "semantic_scholar"], "title": "Draft Model Knows When to Stop: Self-Verification Speculative Decoding for Long-Form Generation", "abstract": "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 hardly satisfies such a requirement. Moreover, such discrepancy is further exacerbated in scenarios involving complex reasoning and long-form generation, particularly under test-time scaling for reasoning-specialized models. Through both theoretical and empirical estimation, we establish that the discrepancy between the draft and target models can be approximated by the draft model's prediction entropy: a high entropy indicates a low acceptance rate of draft tokens, and vice versa. Based on this insight, we propose SVIP: Self-Verification Length Policy for Long-Context Speculative Decoding, which is a training-free dynamic length policy for speculative decoding systems that adaptively determines the lengths of draft sequences by referring to the draft entropy. Experimental results on mainstream SD benchmarks as well as reasoning-heavy benchmarks demonstrate the superior performance of SVIP, achieving up to 17% speedup on MT-Bench at 8K context compared with fixed draft lengths, and 22% speedup for QwQ in long-form reasoning.", "authors": ["Ziyin Zhang", "Jiahao Xu", "Tian Liang", "Xingyu Chen", "Zhiwei He", "Rui Wang", "Zhaopeng Tu"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-27", "url": "https://arxiv.org/abs/2411.18462", "pdf_url": "https://arxiv.org/pdf/2411.18462v2", "arxiv_id": "2411.18462", "doi": "10.18653/v1/2025.emnlp-main.844", "citation_count": 9, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.25}
{"id": "4e5018ee695bce5c2c8fb3d32c5d822a16c798090e593c4817e3f0888b6f0751", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding with CTC-based Draft Model for LLM Inference Acceleration", "abstract": "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 LLM verifies the draft for acceptance or rejection. In this framework, the final inference speed is decided by the decoding speed of the draft model and the acceptance rate of the draft provided by the draft model. Currently the widely used draft models usually generate draft tokens for the next several positions in a non-autoregressive way without considering the correlations between draft tokens. Therefore, it has a high decoding speed but an unsatisfactory acceptance rate. In this paper, we focus on how to improve the performance of the draft model and aim to accelerate inference via a high acceptance rate. To this end, we propose a CTC-based draft model which strengthens the correlations between draft tokens during the draft phase, thereby generating higher-quality draft candidate sequences. Experiment results show that compared to strong baselines, the proposed method can achieve a higher acceptance rate and hence a faster inference speed.", "authors": ["Zhuofan Wen", "Shangtong Gui", "Yang Feng"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-25", "url": "https://arxiv.org/abs/2412.00061", "pdf_url": "https://arxiv.org/pdf/2412.00061v1", "arxiv_id": "2412.00061", "doi": "10.48550/arXiv.2412.00061", "citation_count": 17, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Neural Information Processing Systems", "quality_score": 0.3138}
{"id": "84cb1691e5cd343e4452297706935345d82e1de5146b800f02374f6f5ba11ebf", "sources": ["arxiv", "semantic_scholar"], "title": "A Survey on LLM-as-a-Judge", "abstract": "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 are employed as evaluators for complex tasks. With their ability to process diverse data types and provide scalable, cost-effective, and consistent assessments, LLMs present a compelling alternative to traditional expert-driven evaluations. However, ensuring the reliability of LLM-as-a-Judge systems remains a significant challenge that requires careful design and standardization. This paper provides a comprehensive survey of LLM-as-a-Judge, addressing the core question: How can reliable LLM-as-a-Judge systems be built? We explore strategies to enhance reliability, including improving consistency, mitigating biases, and adapting to diverse assessment scenarios. Additionally, we propose methodologies for evaluating the reliability of LLM-as-a-Judge systems, supported by a novel benchmark designed for this purpose. To advance the development and real-world deployment of LLM-as-a-Judge systems, we also discussed practical applications, challenges, and future directions. This survey serves as a foundational reference for researchers and practitioners in this rapidly evolving field.", "authors": ["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"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-23", "url": "https://arxiv.org/abs/2411.15594", "pdf_url": "https://arxiv.org/pdf/2411.15594v6", "arxiv_id": "2411.15594", "doi": "10.48550/arXiv.2411.15594", "citation_count": 1447, "influential_citation_count": 120, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 1.0}
{"id": "9a4115a62b6d3a37d0ae39dde6a6c75a7d15ca2c2e69f9a891dcbe0ee5e28770", "sources": ["arxiv", "semantic_scholar"], "title": "Closer Look at Efficient Inference Methods: A Survey of Speculative Decoding", "abstract": "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 introducing a two-stage framework: drafting and verification. A smaller, efficient model generates a preliminary draft, which is then refined by a larger, more sophisticated model. This paper provides a comprehensive survey of speculative decoding methods, categorizing them into draft-centric and model-centric approaches. We discuss key ideas associated with each method, highlighting their potential for scaling LLM inference. This survey aims to guide future research in optimizing speculative decoding and its integration into real-world LLM applications.", "authors": ["Hyun Ryu", "Eric Kim"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-20", "url": "https://arxiv.org/abs/2411.13157", "pdf_url": "https://arxiv.org/pdf/2411.13157v2", "arxiv_id": "2411.13157", "doi": "10.48550/arXiv.2411.13157", "citation_count": 7, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2258}
{"id": "03004d57fd097f502c45b6259e86a65a231638b602dfc21282405a7422117ec0", "sources": ["arxiv", "semantic_scholar"], "title": "Continuous Speculative Decoding for Autoregressive Image Generation", "abstract": "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 absence of an analogous theory for continuous distributions precludes its use in accelerating continuous AR models. To fill this gap, this work presents continuous speculative decoding, and addresses challenges from: 1) low acceptance rate, caused by inconsistent output distribution between target and draft models, and 2) modified distribution without analytic expression, caused by complex integral. To address challenge 1), we propose denoising trajectory alignment and token pre-filling strategies. To address challenge 2), we introduce acceptance-rejection sampling algorithm with an appropriate upper bound, thereby avoiding explicitly calculating the integral. Furthermore, our denoising trajectory alignment is also reused in acceptance-rejection sampling, effectively avoiding repetitive diffusion model inference. Extensive experiments demonstrate that our proposed continuous speculative decoding achieves over $2\\times$ speedup on off-the-shelf models, while maintaining the original generation quality. Codes is available at: https://github.com/MarkXCloud/CSpD", "authors": ["Zili Wang", "Robert Zhang", "Kun Ding", "Qi Yang", "Fei Li", "Shiming Xiang"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-18", "url": "https://arxiv.org/abs/2411.11925", "pdf_url": "https://arxiv.org/pdf/2411.11925v2", "arxiv_id": "2411.11925", "doi": "10.48550/arXiv.2411.11925", "citation_count": 17, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/MarkXCloud/CSpD", "venue": "arXiv.org", "quality_score": 0.3138}
{"id": "03fd5745fca7988259118fd262f85a6300cdd2469df9e9929383c9de76df5fdf", "sources": ["arxiv", "semantic_scholar"], "title": "SAM Decoding: Speculative Decoding via Suffix Automaton", "abstract": "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 certain domains. This paper presents a novel retrieval-based speculative decoding method that adapts suffix automaton (SAM) for efficient and accurate draft generation by utilizing common text corpus and dynamic text sequence. Unlike existing $n$-gram matching methods, SAM-Decoding finds the exact longest suffix match, achieving an average time complexity of O(1) per generation step of SAM update and suffix retrieval. It can also integrate with existing methods, adaptively selecting a draft generation strategy based on match length to generalize to broader domains. Extensive experiments on Spec-Bench show that our method is $18\\%+$ faster than other retrieval-based SD methods. Additionally, when combined with advanced EAGLE-2, it provides an additional speedup of $3.28\\%$ -- $11.13\\%$ across various-sized LLM backbones. Our code is available at our \\href{https://github.com/hyx1999/SAM-Decoding}{repository}.", "authors": ["Yuxuan Hu", "Ke Wang", "Xiaokang Zhang", "Fanjin Zhang", "Cuiping Li", "Hong Chen", "Jing Zhang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-16", "url": "https://arxiv.org/abs/2411.10666", "pdf_url": "https://arxiv.org/pdf/2411.10666v3", "arxiv_id": "2411.10666", "doi": "10.48550/arXiv.2411.10666", "citation_count": 24, "influential_citation_count": 4, "has_code": true, "code_url": "https://github.com/hyx1999/SAM-Decoding}{repository}", "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.3495}
{"id": "cce3697d6a635c2423d92c52ce5bb28bae823675a2dd2919cc409f2203cee5b4", "sources": ["arxiv", "semantic_scholar"], "title": "SSSD: Simply-Scalable Speculative Decoding", "abstract": "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 components, increasing deployment and maintenance complexity. This added complexity reduces flexibility, particularly when serving workloads shift to tasks, domains, or languages that are not well represented in the draft model's training data. We introduce Simply-Scalable Speculative Decoding (SSSD), a training-free method that combines lightweight n-gram matching with hardware-aware speculation. Relative to standard autoregressive decoding, SSSD reduces latency by up to 2.9x. It achieves performance on par with leading training-based approaches across a broad range of benchmarks, while requiring substantially lower adoption effort--no data preparation, training or tuning are needed--and exhibiting superior robustness under language and domain shift, as well as in long-context settings.", "authors": ["Michele Marzollo", "Jiawei Zhuang", "Niklas Roemer", "Niklas Zwingenberger", "Lorenz K. Müller", "Lukas Cavigelli"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-08", "url": "https://arxiv.org/abs/2411.05894", "pdf_url": "https://arxiv.org/pdf/2411.05894v3", "arxiv_id": "2411.05894", "doi": "10.48550/arXiv.2411.05894", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1193}
{"id": "3181ff1c3c23fbd3c9cef004c0c3999c575448479bd8b3dad0e4e1dc4601ef63", "sources": ["arxiv", "semantic_scholar"], "title": "SpecHub: Provable Acceleration to Multi-Draft Speculative Decoding", "abstract": "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 LLM verifies in parallel. However, current heuristic approaches, such as Recursive Rejection Sampling (RRS), suffer from low acceptance rates in subsequent drafts, limiting the advantages of using multiple drafts. Meanwhile, Optimal Transport with Membership Cost (OTM) can theoretically improve acceptance rates, but its computational cost is too high for real-time use. We present SpecHub, a novel, efficient sampling-verification method for MDSD that improves acceptance rates with only linear computational overhead. By simplifying the OTM problem into a compact Linear Programming model, SpecHub significantly reduces computational complexity. It further accelerates sampling by leveraging a sparse joint distribution, focusing computation on high-probability token sequences. In extensive experiments, Spechub consistently generates 0.05-0.27 and 0.02-0.16 more tokens per step than RRS and RRS without replacement. We attach our code at \\url{https://github.com/MasterGodzilla/Speculative_decoding_OT}.", "authors": ["Ryan Sun", "Tianyi Zhou", "Xun Chen", "Lichao Sun"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-08", "url": "https://arxiv.org/abs/2411.05289", "pdf_url": "https://arxiv.org/pdf/2411.05289v1", "arxiv_id": "2411.05289", "doi": "10.18653/v1/2024.emnlp-main.1148", "citation_count": 10, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/MasterGodzilla/Speculative_decoding_OT}", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.2603}
{"id": "a8369e2dda58650b4b7c97d55e1a9edee70952783152b14d1ddca3ab717599a6", "sources": ["arxiv", "semantic_scholar"], "title": "SuffixDecoding: Extreme Speculative Decoding for Emerging AI Applications", "abstract": "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, agentic frameworks typically submit repetitive inference requests, such as multi-agent pipelines performing similar subtasks or self-refinement loops iteratively enhancing outputs. These workloads result in long and highly predictable sequences, which current speculative decoding methods do not effectively exploit. To address this gap, we introduce \\emph{SuffixDecoding}, a novel method that utilizes efficient suffix trees to cache long token sequences from prompts and previous outputs. By adaptively speculating more tokens when acceptance likelihood is high and fewer when it is low, SuffixDecoding effectively exploits opportunities for longer speculations while conserving computation when those opportunities are limited. Evaluations on agentic benchmarks, including SWE-Bench and Text-to-SQL, demonstrate that SuffixDecoding achieves speedups of up to 5.3$\\times$, outperforming state-of-the-art methods -- 2.8$\\times$ faster than model-based approaches like EAGLE-2/3 and 1.9$\\times$ faster than model-free approaches such as Token Recycling. SuffixDecoding is open-sourced at https://github.com/snowflakedb/ArcticInference", "authors": ["Gabriele Oliaro", "Zhihao Jia", "Daniel Campos", "Aurick Qiao"], "categories": ["cs.CL", "cs.AI", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-07", "url": "https://arxiv.org/abs/2411.04975", "pdf_url": "https://arxiv.org/pdf/2411.04975v3", "arxiv_id": "2411.04975", "doi": null, "citation_count": 17, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/snowflakedb/ArcticInference", "venue": null, "quality_score": 0.3138}
{"id": "5dcabae329dec8df1448551fa0449aee3a85fbe2a7d0b8052411e16e36ee3dd5", "sources": ["arxiv", "semantic_scholar"], "title": "When Speculation Spills Secrets: Side Channels via Speculative Decoding In LLMs", "abstract": "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 monitoring per-iteration token counts or packet sizes. In evaluations using research prototypes and production-grade vLLM serving frameworks, we show that an adversary monitoring these patterns can fingerprint user queries (from a set of 50 prompts) with over 75% accuracy across four speculative-decoding schemes at temperature 0.3: REST (100%), LADE (91.6%), BiLD (95.2%), and EAGLE (77.6%). Even at temperature 1.0, accuracy remains far above the 2% random baseline - REST (99.6%), LADE (61.2%), BiLD (63.6%), and EAGLE (24%). We also show the capability of the attacker to leak confidential datastore contents used for prediction at rates exceeding 25 tokens/sec. To defend against these, we propose and evaluate a suite of mitigations, including packet padding and iteration-wise token aggregation.", "authors": ["Jiankun Wei", "Abdulrahman Abdulrazzag", "Tianchen Zhang", "Adel Muursepp", "Gururaj Saileshwar"], "categories": ["cs.CL", "cs.AI", "cs.CR", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-11-01", "url": "https://arxiv.org/abs/2411.01076", "pdf_url": "https://arxiv.org/pdf/2411.01076v4", "arxiv_id": "2411.01076", "doi": null, "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.1747}
{"id": "c832770bf60a9e05bc3e1036528826a28be5210da647c61f8c42de8218fd3ae0", "sources": ["arxiv", "semantic_scholar"], "title": "A Theoretical Perspective for Speculative Decoding Algorithm", "abstract": "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 effectiveness, the theoretical understanding of Speculative Decoding is falling behind. This paper tackles this gap by conceptualizing the decoding problem via markov chain abstraction and studying the key properties, \\emph{output quality and inference acceleration}, from a theoretical perspective. Our analysis covers the theoretical limits of speculative decoding, batch algorithms, and output quality-inference acceleration tradeoffs. Our results reveal the fundamental connections between different components of LLMs via total variation distances and show how they jointly affect the efficiency of decoding algorithms.", "authors": ["Ming Yin", "Minshuo Chen", "Kaixuan Huang", "Mengdi Wang"], "categories": ["cs.LG", "cs.AI", "cs.CL", "stat.ML"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2024-10-30", "url": "https://arxiv.org/abs/2411.00841", "pdf_url": "https://arxiv.org/pdf/2411.00841v1", "arxiv_id": "2411.00841", "doi": "10.48550/arXiv.2411.00841", "citation_count": 27, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": "Neural Information Processing Systems", "quality_score": 0.3618}
{"id": "2efc7e9d844e87ea5ca0bfda6514a36b1445a67848e8900e0a7a573946edbfe1", "sources": ["arxiv", "semantic_scholar"], "title": "Fast and High-Quality Auto-Regressive Speech Synthesis via Speculative Decoding", "abstract": "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 to accelerate auto-regressive TTS through speculative decoding. Our results show that VADUSA not only significantly improves inference speed but also enhances performance by incorporating draft heads to predict future speech content auto-regressively. Furthermore, the inclusion of a tolerance mechanism during sampling accelerates inference without compromising quality. Our approach demonstrates strong generalization across large datasets and various types of speech tokens.", "authors": ["Bohan Li", "Hankun Wang", "Situo Zhang", "Yiwei Guo", "Kai Yu"], "categories": ["eess.AS", "cs.AI", "cs.SD"], "fields_of_study": ["Engineering", "Computer Science"], "published_date": "2024-10-29", "url": "https://arxiv.org/abs/2410.21951", "pdf_url": "https://arxiv.org/pdf/2410.21951v2", "arxiv_id": "2410.21951", "doi": "10.1109/ICASSP49660.2025.10888194", "citation_count": 22, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE International Conference on Acoustics, Speech, and Signal Processing", "quality_score": 0.3404}
{"id": "6b9c34c19b56104c44ca32d9f768aa298195c84c7434e25ba1aaafec7c180d70", "sources": ["arxiv", "semantic_scholar"], "title": "Fast Best-of-N Decoding via Speculative Rejection", "abstract": "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 a phase called post-training. While predominant, these post-training methods add substantial complexity before LLMs can be deployed. Inference-time alignment methods avoid the complex post-training step and instead bias the generation towards responses that are aligned with human preferences. The best-known inference-time alignment method, called Best-of-N, is as effective as the state-of-the-art post-training procedures. Unfortunately, Best-of-N requires vastly more resources at inference time than standard decoding strategies, which makes it computationally not viable. In this work, we introduce Speculative Rejection, a computationally-viable inference-time alignment algorithm. It generates high-scoring responses according to a given reward model, like Best-of-N does, while being between 16 to 32 times more computationally efficient.", "authors": ["Hanshi Sun", "Momin Haider", "Ruiqi Zhang", "Huitao Yang", "Jiahao Qiu", "Ming Yin", "Mengdi Wang", "Peter Bartlett", "Andrea Zanette"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-26", "url": "https://arxiv.org/abs/2410.20290", "pdf_url": "https://arxiv.org/pdf/2410.20290v2", "arxiv_id": "2410.20290", "doi": "10.48550/arXiv.2410.20290", "citation_count": 131, "influential_citation_count": 15, "has_code": false, "code_url": null, "venue": "Neural Information Processing Systems", "quality_score": 0.6021}
{"id": "61a1b36da45fdb72b3ad3ee3ec0fa977dff1a17f5e592b7a532c366666f7c642", "sources": ["arxiv", "semantic_scholar"], "title": "AdaEDL: Early Draft Stopping for Speculative Decoding of Large Language Models via an Entropy-based Lower Bound on Token Acceptance Probability", "abstract": "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 model to propose draft tokens which are then verified in parallel. The number of draft tokens produced in each drafting round is referred to as the draft length and is often a static hyperparameter chosen based on the acceptance rate statistics of the draft tokens. However, setting a static draft length can negatively impact performance, especially in scenarios where drafting is expensive and there is a high variance in the number of tokens accepted. Adaptive Entropy-based Draft Length (AdaEDL) is a simple, training and parameter-free criteria which allows for early stopping of the token drafting process by approximating a lower bound on the expected acceptance probability of the drafted token based on the currently observed entropy of the drafted logits. We show that AdaEDL consistently outperforms static draft-length speculative decoding by 10%-57% as well as other training-free draft-stopping techniques by upto 10% in a variety of settings and datasets. At the same time, we show that AdaEDL is more robust than these techniques and preserves performance in high-sampling-temperature scenarios. Since it is training-free, in contrast to techniques that rely on the training of dataset-specific draft-stopping predictors, AdaEDL can seamlessly be integrated into a variety of pre-existing LLM systems.", "authors": ["Sudhanshu Agrawal", "Wonseok Jeon", "Mingu Lee"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-24", "url": "https://arxiv.org/abs/2410.18351", "pdf_url": "https://arxiv.org/pdf/2410.18351v1", "arxiv_id": "2410.18351", "doi": "10.48550/arXiv.2410.18351", "citation_count": 17, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3138}
{"id": "22c84d0293840fef668552d3801312ef6db48915c69e8e77f0c7fa9fb0d1dbd9", "sources": ["arxiv", "semantic_scholar"], "title": "POD-Attention: Unlocking Full Prefill-Decode Overlap for Faster LLM Inference", "abstract": "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 inefficient for attention computation because existing attention kernels specialize execution independently for the prefill and decode phases. In this paper, we present POD-Attention - the first GPU kernel that efficiently computes attention for hybrid batches. POD-Attention aims to maximize the utilization of both compute and memory bandwidth by carefully allocating the GPU's resources such that prefill and decode operations happen concurrently on the same multiprocessor. POD-Attention speeds up attention computation by up to $59\\%$ (mean $28\\%$), enabling higher throughput and lower latency LLM inference compared to the use of independently optimized prefill and decode attention kernels.", "authors": ["Aditya K Kamath", "Ramya Prabhu", "Jayashree Mohan", "Simon Peter", "Ramachandran Ramjee", "Ashish Panwar"], "categories": ["cs.LG", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-23", "url": "https://arxiv.org/abs/2410.18038", "pdf_url": "https://arxiv.org/pdf/2410.18038v2", "arxiv_id": "2410.18038", "doi": "10.1145/3676641.3715996", "citation_count": 76, "influential_citation_count": 7, "has_code": false, "code_url": null, "venue": "International Conference on Architectural Support for Programming Languages and Operating Systems", "quality_score": 0.4716}
{"id": "1c093c880586c630d3eec1fa16739838534ec398c013a14f92c03fdadfeebb17", "sources": ["arxiv", "semantic_scholar"], "title": "AMUSD: Asynchronous Multi-Device Speculative Decoding for LLM Acceleration", "abstract": "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 larger (original) verify model, resulting in overall speedup compared to using the larger model by itself in an autoregressive fashion. In this work, we introduce AMUSD (Asynchronous Multi-device Speculative Decoding), a system that further accelerates generation by decoupling the draft and verify phases into a continuous, asynchronous approach. Unlike conventional speculative decoding, where only one model (draft or verify) performs token generation at a time, AMUSD enables both models to perform predictions independently on separate devices (e.g., GPUs). We evaluate our approach over multiple datasets and show that AMUSD achieves an average 29% improvement over speculative decoding and up to 1.96$\\times$ speedup over conventional autoregressive decoding, while achieving identical output quality. Our system is open-source and available at https://github.com/BradMcDanel/AMUSD/.", "authors": ["Bradley McDanel"], "categories": ["cs.CL", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-22", "url": "https://arxiv.org/abs/2410.17375", "pdf_url": "https://arxiv.org/pdf/2410.17375v1", "arxiv_id": "2410.17375", "doi": "10.1109/ISCAS56072.2025.11043575", "citation_count": 13, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/BradMcDanel/AMUSD/", "venue": "International Symposium on Circuits and Systems", "quality_score": 0.301}
{"id": "91445b29ab819d595956f9e9fb5a20fda991cb16c9770b1296bdfee695d6a795", "sources": ["arxiv", "semantic_scholar"], "title": "Progressive Mixed-Precision Decoding for Efficient LLM Inference", "abstract": "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, utilizing low precisions (i.e.~2/3-bit) to substantially alleviate the memory-boundedness of LLM decoding, still suffers from prohibitive performance drop. In this work, we argue that existing approaches fail to explore the diversity in computational patterns, redundancy, and sensitivity to approximations of the different phases of LLM inference, resorting to a uniform quantization policy throughout. Instead, we propose a novel phase-aware method that selectively allocates precision during different phases of LLM inference, achieving both strong context extraction during prefill and efficient memory bandwidth utilization during decoding. To further address the memory-boundedness of the decoding phase, we introduce Progressive Mixed-Precision Decoding (PMPD), a technique that enables the gradual lowering of precision deeper in the generated sequence, together with a spectrum of precision-switching schedulers that dynamically drive the precision-lowering decisions in either task-adaptive or prompt-adaptive manner. Extensive evaluation across diverse language tasks shows that when targeting Nvidia GPUs, PMPD achieves 1.4$-$12.2$\\times$ speedup in matrix-vector multiplications over fp16 models, while when targeting an LLM-optimized NPU, our approach delivers a throughput gain of 3.8$-$8.0$\\times$ over fp16 models and up to 1.54$\\times$ over uniform quantization approaches while preserving the output quality.", "authors": ["Hao Mark Chen", "Fuwen Tan", "Alexandros Kouris", "Royson Lee", "Hongxiang Fan", "Stylianos I. Venieris"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-17", "url": "https://arxiv.org/abs/2410.13461", "pdf_url": "https://arxiv.org/pdf/2410.13461v2", "arxiv_id": "2410.13461", "doi": "10.48550/arXiv.2410.13461", "citation_count": 14, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.294}
{"id": "f02aa6149c1fda7bee6cf7225cbe8e955277d0d744ec66648ed5b88853f8fd1c", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Codec-based Speech Synthesis with Multi-Token Prediction and Speculative Decoding", "abstract": "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 per inference step of the AR module using multiple prediction heads, resulting in a linear reduction in synthesis time as the number of heads increases. Furthermore, we introduce a novel speculative decoding technique that utilises a Viterbi-based algorithm to select the optimal sequence of generated tokens at each decoding step. In our experiments, we demonstrate that the time required to predict each token is reduced by a factor of 4 to 5 compared to baseline models, with minimal quality trade-off or even improvement in terms of speech intelligibility. Audio samples are available at: multpletokensprediction.github.io/multipletokensprediction.github.io/.", "authors": ["Tan Dat Nguyen", "Ji-Hoon Kim", "Jeongsoo Choi", "Shukjae Choi", "Jinseok Park", "Younglo Lee", "Joon Son Chung"], "categories": ["cs.SD", "cs.AI", "eess.AS"], "fields_of_study": ["Computer Science", "Engineering"], "published_date": "2024-10-17", "url": "https://arxiv.org/abs/2410.13839", "pdf_url": "https://arxiv.org/pdf/2410.13839v1", "arxiv_id": "2410.13839", "doi": "10.1109/ICASSP49660.2025.10887855", "citation_count": 14, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "IEEE International Conference on Acoustics, Speech, and Signal Processing", "quality_score": 0.294}
{"id": "17965272fb0d48fc110b7c1084ccad857bacd8754c3f53f85e5a9f4c56bed98b", "sources": ["arxiv", "semantic_scholar"], "title": "Cerberus: Efficient Inference with Adaptive Parallel Decoding and Sequential Knowledge Enhancement", "abstract": "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 heads fail to balance prediction accuracy and the parallelism of execution, and (2) parallel decoding is not a universal solution, as it can bring unnecessary overheads at some challenging decoding steps. To address these issues, we propose Cerberus, an adaptive parallel decoding framework introduces the gating mechanism to enable the LLMs to adaptively choose appropriate decoding approaches at each decoding step, along with introducing a new paradigm of decoding heads that introduce the sequential knowledge while maintaining execution parallelism. The experiment results demonstrate that the Cerberus can achieve up to 2.12x speed up compared to auto-regressive decoding, and outperforms one of the leading parallel decoding frameworks, Medusa, with a 10% - 30% increase in acceleration and superior generation quality.", "authors": ["Yuxuan Liu", "Wenyuan Li", "Laizhong Cui", "Hailiang Yang"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-17", "url": "https://arxiv.org/abs/2410.13344", "pdf_url": "https://arxiv.org/pdf/2410.13344v1", "arxiv_id": "2410.13344", "doi": "10.48550/arXiv.2410.13344", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.0753}
{"id": "ab684e7d67d4a3784a2dbf32fedc845f9ad7e01e8df98fd80be66a3c208ee77b", "sources": ["arxiv", "semantic_scholar"], "title": "QSpec: Speculative Decoding with Complementary Quantization Schemes", "abstract": "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 quantization paradigm that decouples efficiency from quality by integrating two complementary schemes via speculative decoding: low-precision joint quantization for fast drafting and high-precision weight-only quantization for accurate verification. QSpec reuses both weights and KV cache across stages, enabling near-zero-cost switching without retraining or auxiliary models. Compared to high-precision baselines, QSpec achieves up to 1.64x speedup without quality degradation, and outperforms state-of-the-art speculative decoding methods by up to 1.55x in batched settings. Furthermore, QSpec supports plug-and-play deployment and generalizes well across model scales, quantization methods, and workloads. These properties make QSpec a practical and scalable solution for high-fidelity quantized LLM serving under memory-constrained scenarios. Our code is available at https://github.com/hku-netexplo-lab/QSpec.", "authors": ["Juntao Zhao", "Wenhao Lu", "Sheng Wang", "Lingpeng Kong", "Chuan Wu"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-15", "url": "https://arxiv.org/abs/2410.11305", "pdf_url": "https://arxiv.org/pdf/2410.11305v3", "arxiv_id": "2410.11305", "doi": "10.48550/arXiv.2410.11305", "citation_count": 14, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/hku-netexplo-lab/QSpec", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.301}
{"id": "a93b5f3e1e4bd6a3848a063174fa5d114516887d7df869f91610601233788675", "sources": ["arxiv", "semantic_scholar"], "title": "DySpec: Faster Speculative Decoding with Dynamic Token Tree Structure", "abstract": "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 fails to generalize to diverse query distributions. In this paper, we propose DySpec, a faster speculative decoding algorithm with a novel dynamic token tree structure. We begin by bridging the draft distribution and acceptance rate from intuitive and empirical clues, and successfully show that the two variables are strongly correlated. Based on this, we employ a greedy strategy to dynamically expand the token tree at run time. Theoretically, we show that our method can achieve optimal results under mild assumptions. Empirically, DySpec yields a higher acceptance rate and speedup than fixed trees. DySpec can drastically improve the throughput and reduce the latency of token generation across various data distribution and model sizes, which significantly outperforms strong competitors, including Specinfer and Sequoia. Under low temperature setting, DySpec can improve the throughput up to 9.1$\\times$ and reduce the latency up to 9.4$\\times$ on Llama2-70B. Under high temperature setting, DySpec can also improve the throughput up to 6.21$\\times$, despite the increasing difficulty of speculating more than one token per step for draft model.", "authors": ["Yunfan Xiong", "Ruoyu Zhang", "Yanzeng Li", "Tianhao Wu", "Lei Zou"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-15", "url": "https://arxiv.org/abs/2410.11744", "pdf_url": "https://arxiv.org/pdf/2410.11744v1", "arxiv_id": "2410.11744", "doi": "10.1007/s11280-025-01344-0", "citation_count": 19, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "World wide web (Bussum)", "quality_score": 0.3253}
{"id": "4b5c1b6f8c51771d1edff1a9435c1d958151e5c62a17aebbdc0e586f96dda66c", "sources": ["arxiv", "semantic_scholar"], "title": "Temperature-Centric Investigation of Speculative Decoding with Knowledge Distillation", "abstract": "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 speculative decoding, the influence of generation configurations on the decoding process remains poorly understood, especially concerning decoding temperatures. This paper delves into the effects of decoding temperatures on speculative decoding's efficacy. Beginning with knowledge distillation (KD), we first highlight the challenge of decoding at higher temperatures, and demonstrate KD in a consistent temperature setting could be a remedy. We also investigate the effects of out-of-domain testing sets with out-of-range temperatures. Building upon these findings, we take an initial step to further the speedup for speculative decoding, particularly in a high-temperature generation setting. Our work offers new insights into how generation configurations drastically affect the performance of speculative decoding, and underscores the need for developing methods that focus on diverse decoding configurations. Code is publically available at https://github.com/ozyyshr/TempSpec.", "authors": ["Siru Ouyang", "Shuohang Wang", "Minhao Jiang", "Ming Zhong", "Donghan Yu", "Jiawei Han", "Yelong Shen"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-14", "url": "https://arxiv.org/abs/2410.10141", "pdf_url": "https://arxiv.org/pdf/2410.10141v1", "arxiv_id": "2410.10141", "doi": "10.48550/arXiv.2410.10141", "citation_count": 8, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/ozyyshr/TempSpec", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.2386}
{"id": "54100a3ee502ecb7f548f7cc5df9ea300ed1e391e306f486483f7be776f8e0fa", "sources": ["arxiv", "semantic_scholar"], "title": "Local and Global Decoding in Text Generation", "abstract": "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 investigate the effect of this distortion by introducing globally-normalised versions of these decoding methods. Additionally, we propose an independent Metropolis-Hastings algorithm to approximate sampling from globally-normalised distributions without explicitly computing them. Our empirical analysis compares the performance of local and global normalisation across two decoding algorithms (top-$k$ and top-$π$) with various hyperparameters, using Pythia language models. Results show that, in most configurations, global decoding performs worse than the local decoding version of the same algorithms -- despite preserving the distribution's integrity. Our results suggest that distortion is an important feature of local decoding algorithms.", "authors": ["Daniel Gareev", "Thomas Hofmann", "Ezhilmathi Krishnasamy", "Tiago Pimentel"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-14", "url": "https://arxiv.org/abs/2410.10810", "pdf_url": "https://arxiv.org/pdf/2410.10810v1", "arxiv_id": "2410.10810", "doi": "10.48550/arXiv.2410.10810", "citation_count": 3, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/lowlypalace/global-decoding", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.1505}
{"id": "aeefa6380c28437d1b69e53bc3c92f98d3d5f6ea0e7a3be725a2448026dfd541", "sources": ["arxiv", "semantic_scholar"], "title": "Nudging: Inference-time Alignment of LLMs via Guided Decoding", "abstract": "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 at inference time using a small aligned model. NUDGING is motivated by recent findings that alignment primarily alters the model's behavior on a small subset of stylistic tokens (e.g., discourse markers). We find that base models are significantly more uncertain when generating these tokens. Building on this insight, NUDGING employs a small aligned model to generate nudging tokens to guide the base model's output during decoding when the base model's uncertainty is high, with only a minor additional inference overhead. We evaluate NUDGING across 3 model families on a diverse range of open-instruction tasks. Without any training, nudging a large base model with a 7x-14x smaller aligned model achieves zero-shot performance comparable to, and sometimes surpassing, that of large aligned models. By operating at the token level, NUDGING enables off-the-shelf collaboration between model families. For instance, nudging Gemma-2-27b with Llama-27b-chat outperforms Llama-2-70b-chat on various tasks. Overall, our work offers a modular and cost-efficient solution to LLM alignment. Our code and demo are available at: https://fywalter.github.io/nudging/ .", "authors": ["Yu Fei", "Yasaman Razeghi", "Sameer Singh"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-11", "url": "https://arxiv.org/abs/2410.09300", "pdf_url": "https://arxiv.org/pdf/2410.09300v4", "arxiv_id": "2410.09300", "doi": "10.18653/v1/2025.acl-long.623", "citation_count": 23, "influential_citation_count": 7, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.4515}
{"id": "faf30e0935df27e700a08ef92f72991789e531a3d14b7b2b7135627f520a9ae8", "sources": ["arxiv", "semantic_scholar"], "title": "SWIFT: On-the-Fly Self-Speculative Decoding for LLM Inference Acceleration", "abstract": "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 existing approaches necessitate either additional parameters or extensive training to construct effective draft models, thereby restricting their applicability across different LLMs and tasks. To address this limitation, we explore a novel plug-and-play SD solution with layer-skipping, which skips intermediate layers of the target LLM as the compact draft model. Our analysis reveals that LLMs exhibit great potential for self-acceleration through layer sparsity and the task-specific nature of this sparsity. Building on these insights, we introduce SWIFT, an on-the-fly self-speculative decoding algorithm that adaptively selects intermediate layers of LLMs to skip during inference. SWIFT does not require auxiliary models or additional training, making it a plug-and-play solution for accelerating LLM inference across diverse input data streams. Our extensive experiments across a wide range of models and downstream tasks demonstrate that SWIFT can achieve over a 1.3x-1.6x speedup while preserving the original distribution of the generated text. We release our code in https://github.com/hemingkx/SWIFT.", "authors": ["Heming Xia", "Yongqi Li", "Jun Zhang", "Cunxiao Du", "Wenjie Li"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-09", "url": "https://arxiv.org/abs/2410.06916", "pdf_url": "https://arxiv.org/pdf/2410.06916v2", "arxiv_id": "2410.06916", "doi": "10.48550/arXiv.2410.06916", "citation_count": 63, "influential_citation_count": 7, "has_code": true, "code_url": "https://github.com/hemingkx/SWIFT", "venue": "International Conference on Learning Representations", "quality_score": 0.4515}
{"id": "5d2f3d3194aba121f17d6303b217f950bea95cee26872108081b1f39ffc1255d", "sources": ["arxiv", "semantic_scholar"], "title": "Root Defence Strategies: Ensuring Safety of LLM at the Decoding Level", "abstract": "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 responses from the prefill-level lacks utilization of the model's decoding outputs, leading to relatively lower effectiveness and robustness. 2) Rejecting potentially harmful responses based on a single evaluation can significantly impair the model's helpfulness.This paper examines the LLMs' capability to recognize harmful outputs, revealing and quantifying their proficiency in assessing the danger of previous tokens. Motivated by pilot experiment results, we design a robust defense mechanism at the decoding level. Our novel decoder-oriented, step-by-step defense architecture corrects harmful queries directly rather than rejecting them outright. We introduce speculative decoding to enhance usability and facilitate deployment to boost secure decoding speed. Extensive experiments demonstrate that our approach improves model security without compromising reasoning speed. Notably, our method leverages the model's ability to discern hazardous information, maintaining its helpfulness compared to existing methods.", "authors": ["Xinyi Zeng", "Yuying Shang", "Jiawei Chen", "Jingyuan Zhang", "Yu Tian"], "categories": ["cs.CL", "cs.CR"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-09", "url": "https://arxiv.org/abs/2410.06809", "pdf_url": "https://arxiv.org/pdf/2410.06809v3", "arxiv_id": "2410.06809", "doi": "10.48550/arXiv.2410.06809", "citation_count": 13, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2865}
{"id": "5323256de97396efd50cccfab12e97b271dc6ecb8e46fc91ff97c4441decc897", "sources": ["arxiv", "semantic_scholar"], "title": "ParallelSpec: Parallel Drafter for Efficient Speculative Decoding", "abstract": "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 dependency in language modeling, which we consider a huge computational burden in speculative decoding. We present ParallelSpec, an alternative to auto-regressive drafting strategies in state-of-the-art speculative decoding approaches. In contrast to auto-regressive drafting in the speculative stage, we train a parallel drafter to serve as an efficient speculative model. ParallelSpec learns to efficiently predict multiple future tokens in parallel using a single model, and it can be integrated into any speculative decoding framework that requires aligning the output distributions of the drafter and the target model with minimal training cost. Experimental results show that ParallelSpec accelerates baseline methods in latency up to 62% on text generation benchmarks from different domains, and it achieves 2.84X overall speedup on the Llama-2-13B model using third-party evaluation criteria.", "authors": ["Zilin Xiao", "Hongming Zhang", "Tao Ge", "Siru Ouyang", "Vicente Ordonez", "Dong Yu"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-08", "url": "https://arxiv.org/abs/2410.05589", "pdf_url": "https://arxiv.org/pdf/2410.05589v1", "arxiv_id": "2410.05589", "doi": "10.48550/arXiv.2410.05589", "citation_count": 21, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3356}
{"id": "0e153fc8151a5930e39e2f20ea16a4e9e93dcc12cd84fb140d4edcb9a0228ad2", "sources": ["arxiv", "semantic_scholar"], "title": "PAD: Personalized Alignment of LLMs at Decoding-Time", "abstract": "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 novel framework designed to align LLM outputs with diverse personalized preferences during the inference phase, eliminating the need for additional training. By introducing a unique personalized reward modeling strategy, this framework decouples the text generation process from personalized preferences, facilitating the generation of generalizable token-level personalized rewards. The PAD algorithm leverages these rewards to guide the decoding process, dynamically tailoring the base model's predictions to personalized preferences. Extensive experimental results demonstrate that PAD not only outperforms existing training-based alignment methods in terms of aligning with diverse preferences but also shows significant generalizability to preferences unseen during training and scalability across different base models. This work advances the capability of LLMs to meet user needs in real-time applications, presenting a substantial step forward in personalized LLM alignment.", "authors": ["Ruizhe Chen", "Xiaotian Zhang", "Meng Luo", "Wenhao Chai", "Zuozhu Liu"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-05", "url": "https://arxiv.org/abs/2410.04070", "pdf_url": "https://arxiv.org/pdf/2410.04070v7", "arxiv_id": "2410.04070", "doi": null, "citation_count": 55, "influential_citation_count": 6, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.437}
{"id": "75d6d2345b8da07774dfd290af494015f86565b95bc3828a29668718ac3d9319", "sources": ["arxiv", "semantic_scholar"], "title": "LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding", "abstract": "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 diffusion-based methods that operate more efficiently. While speculative decoding has proven effective for accelerating LLMs by generating multiple tokens in a single forward, its application in visual AR models remains largely unexplored. In this work, we identify a challenge in this setting, which we term \\textit{token selection ambiguity}, wherein visual AR models frequently assign uniformly low probabilities to tokens, hampering the performance of speculative decoding. To overcome this challenge, we propose a relaxed acceptance condition referred to as LANTERN that leverages the interchangeability of tokens in latent space. This relaxation restores the effectiveness of speculative decoding in visual AR models by enabling more flexible use of candidate tokens that would otherwise be prematurely rejected. Furthermore, by incorporating a total variation distance bound, we ensure that these speed gains are achieved without significantly compromising image quality or semantic coherence. Experimental results demonstrate the efficacy of our method in providing a substantial speed-up over speculative decoding. In specific, compared to a naïve application of the state-of-the-art speculative decoding, LANTERN increases speed-ups by $\\mathbf{1.75}\\times$ and $\\mathbf{1.82}\\times$, as compared to greedy decoding and random sampling, respectively, when applied to LlamaGen, a contemporary visual AR model. The code is publicly available at https://github.com/jadohu/LANTERN.", "authors": ["Doohyuk Jang", "Sihwan Park", "June Yong Yang", "Yeonsung Jung", "Jihun Yun", "Souvik Kundu", "Sung-Yub Kim", "Eunho Yang"], "categories": ["cs.CV", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-04", "url": "https://arxiv.org/abs/2410.03355", "pdf_url": "https://arxiv.org/pdf/2410.03355v3", "arxiv_id": "2410.03355", "doi": "10.48550/arXiv.2410.03355", "citation_count": 48, "influential_citation_count": 10, "has_code": true, "code_url": "https://github.com/jadohu/LANTERN", "venue": "International Conference on Learning Representations", "quality_score": 0.5207}
{"id": "aa0047df85dde3f7bcf7a7a607ec1c0e1ed0f3543f58cc0322c86cd4e4f31d1c", "sources": ["arxiv", "semantic_scholar"], "title": "Mixture of Attentions For Speculative Decoding", "abstract": "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. Small models that utilise activations from the LLM currently achieve the fastest decoding speeds. However, we identify several limitations of SD models including the lack of on-policyness during training and partial observability. To address these shortcomings, we propose a more grounded architecture for small models by introducing a Mixture of Attentions for SD. Our novel architecture can be applied in two scenarios: a conventional single device deployment and a novel client-server deployment where the small model is hosted on a consumer device and the LLM on a server. In a single-device scenario, we demonstrate state-of-the-art speedups improving EAGLE-2 by 9.5% and its acceptance length by 25%. In a client-server setting, our experiments demonstrate: 1) state-of-the-art latencies with minimal calls to the server for different network conditions, and 2) in the event of a complete disconnection, our approach can maintain higher accuracy compared to other SD methods and demonstrates advantages over API calls to LLMs, which would otherwise be unable to continue the generation process.", "authors": ["Matthieu Zimmer", "Milan Gritta", "Gerasimos Lampouras", "Haitham Bou Ammar", "Jun Wang"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-04", "url": "https://arxiv.org/abs/2410.03804", "pdf_url": "https://arxiv.org/pdf/2410.03804v2", "arxiv_id": "2410.03804", "doi": "10.48550/arXiv.2410.03804", "citation_count": 17, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.3138}
{"id": "7c7483264deeca5cbb6cdf8ceb8ebf881691fbc8d81f8866467895f5f8a54e7c", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding", "abstract": "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, has been used to accelerate the auto-regressive generation and can be executed without training. However, the Jacobi decoding relies on a deterministic criterion to determine the convergence of iterations. Thus, it works for greedy decoding but is incompatible with sampling-based decoding which is crucial for visual quality and diversity in the current auto-regressive text-to-image generation. In this paper, we propose a training-free probabilistic parallel decoding algorithm, Speculative Jacobi Decoding (SJD), to accelerate auto-regressive text-to-image generation. By introducing a probabilistic convergence criterion, our SJD accelerates the inference of auto-regressive text-to-image generation while maintaining the randomness in sampling-based token decoding and allowing the model to generate diverse images. Specifically, SJD facilitates the model to predict multiple tokens at each step and accepts tokens based on the probabilistic criterion, enabling the model to generate images with fewer steps than the conventional next-token-prediction paradigm. We also investigate the token initialization strategies that leverage the spatial locality of visual data to further improve the acceleration ratio under specific scenarios. We conduct experiments for our proposed SJD on multiple auto-regressive text-to-image generation models, showing the effectiveness of model acceleration without sacrificing the visual quality. The code of our work is available here: https://github.com/tyshiwo1/Accelerating-T2I-AR-with-SJD/.", "authors": ["Yao Teng", "Han Shi", "Xian Liu", "Xuefei Ning", "Guohao Dai", "Yu Wang", "Zhenguo Li", "Xihui Liu"], "categories": ["cs.CV"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-02", "url": "https://arxiv.org/abs/2410.01699", "pdf_url": "https://arxiv.org/pdf/2410.01699v2", "arxiv_id": "2410.01699", "doi": "10.48550/arXiv.2410.01699", "citation_count": 61, "influential_citation_count": 11, "has_code": true, "code_url": "https://github.com/tyshiwo1/Accelerating-T2I-AR-with-SJD/", "venue": "International Conference on Learning Representations", "quality_score": 0.5396}
{"id": "d7f2a64117ff70b7c5f75112b9810d531001e70ae10f8ba868bff7617a8becb0", "sources": ["arxiv", "semantic_scholar"], "title": "Draft on the Fly: Adaptive Self-Speculative Decoding using Cosine Similarity", "abstract": "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 input context. We show empirically that our light-weight algorithm is competitive with the current SOTA for self-speculative decoding, while being a truly plug-and-play method.", "authors": ["Michael R. Metel", "Peng Lu", "Boxing Chen", "Mehdi Rezagholizadeh", "Ivan Kobyzev"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-10-01", "url": "https://arxiv.org/abs/2410.01028", "pdf_url": "https://arxiv.org/pdf/2410.01028v1", "arxiv_id": "2410.01028", "doi": "10.48550/arXiv.2410.01028", "citation_count": 12, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.2785}
{"id": "57666f2b5c6effd4ffad19c2ac5a8eb9baa6c8cdbb45bfefc8a4794f643d1fa0", "sources": ["arxiv", "semantic_scholar"], "title": "Dynamic-Width Speculative Beam Decoding for Efficient LLM Inference", "abstract": "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 are then validated simultaneously by the larger model, achieving a speed-up of 1-2x. Although speculative decoding matches the same distribution as multinomial sampling, multinomial sampling itself is prone to suboptimal outputs, whereas beam sampling is widely recognized for producing higher-quality results by maintaining multiple candidate sequences at each step. This paper explores the novel integration of speculative decoding with beam sampling. However, there are four key challenges: (1) how to generate multiple sequences from the larger model's distribution given drafts sequences from the small model; (2) how to dynamically optimize the number of beams to balance efficiency and accuracy; (3) how to efficiently verify the multiple drafts in parallel; and (4) how to address the extra memory costs inherent in beam sampling. To address these challenges, we propose dynamic-width speculative beam decoding (DSBD). Specifically, we first introduce a novel draft and verification scheme that generates multiple sequences following the large model's distribution based on beam sampling trajectories from the small model. Then, we introduce an adaptive mechanism to dynamically tune the number of beams based on the context, optimizing efficiency and effectiveness. Besides, we extend tree-based parallel verification to handle multiple trees simultaneously, accelerating the verification process. Finally, we illustrate a simple modification to our algorithm to mitigate the memory overhead of beam sampling...", "authors": ["Zongyue Qin", "Zifan He", "Neha Prakriya", "Jason Cong", "Yizhou Sun"], "categories": ["cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-09-25", "url": "https://arxiv.org/abs/2409.16560", "pdf_url": "https://arxiv.org/pdf/2409.16560v2", "arxiv_id": "2409.16560", "doi": "10.48550/arXiv.2409.16560", "citation_count": 7, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2386}
{"id": "897fefe7dea369540b92914d82dd69bfd413b5ff41923eabe887805f357e141b", "sources": ["arxiv", "semantic_scholar"], "title": "Improving Multi-candidate Speculative Decoding", "abstract": "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 candidate tokens from the draft model at each step and verifying them in parallel, thus increasing the chances of accepting a token and reducing generation time. Existing MCSD methods rely on the draft model to initialize the multi-candidate sequences and use static length and tree attention structure for draft generation. However, such an approach suffers from the draft and target model's output distribution differences, especially in a dynamic generation context. In this work, we introduce a new version of MCSD that includes a target model initialized multi-candidate generation, a dynamic sliced topology-aware causal mask for dynamic length adjustment, and decision models to optimize early stopping. We experimented with our method on Llama 2-7B and its variants and observed a maximum 27.5% speedup compared to our MCSD baseline across three benchmarks with Llama 2-7B as the target model and JackFram 68M as the draft model. Additionally, we evaluate the effects of using the target model initialized multi-candidate process with different draft models on output quality.", "authors": ["Xiaofan Lu", "Yixiao Zeng", "Feiyang Ma", "Zixu Yu", "Marco Levorato"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-09-16", "url": "https://arxiv.org/abs/2409.10644", "pdf_url": "https://arxiv.org/pdf/2409.10644v3", "arxiv_id": "2409.10644", "doi": "10.48550/arXiv.2409.10644", "citation_count": 6, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.2113}
{"id": "2b80eaed4e9a91740e4a9f421b6dbb92f9ba3dc97b89197ea2dba920a341dc19", "sources": ["arxiv", "semantic_scholar"], "title": "Dynamic Depth Decoding: Faster Speculative Decoding for LLMs", "abstract": "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 optimises EAGLE-2's tree drafting method using a dynamic depth. This extends the average speedup that EAGLE-2 achieves over EAGLE by $44\\%$, giving DDD an average speedup of $3.16$x.", "authors": ["Oscar Brown", "Zhengjie Wang", "Andrea Do", "Nikhil Mathew", "Cheng Yu"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-30", "url": "https://arxiv.org/abs/2409.00142", "pdf_url": "https://arxiv.org/pdf/2409.00142v1", "arxiv_id": "2409.00142", "doi": "10.48550/arXiv.2409.00142", "citation_count": 17, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3138}
{"id": "2da62c5a97f735d04c021fc624526aa374355314904ac96594a7e5c49cfb9afd", "sources": ["arxiv", "semantic_scholar"], "title": "Boosting Lossless Speculative Decoding via Feature Sampling and Partial Alignment Distillation", "abstract": "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 autoregression within the draft model to facilitate more straightforward predictions and enhanced knowledge distillation. In this paper, we reassess these approaches and propose FSPAD (Feature Sampling and Partial Alignment Distillation for Lossless Speculative Decoding), which introduces two straightforward and effective components within the existing framework to boost lossless speculative decoding. Firstly, FSPAD utilizes token embeddings to sample features of the target LLM in high-dimensional space before feeding them into the draft model, due to the inherent uncertainty of the features preventing the draft model from obtaining the specific token output by the target LLM. Secondly, FSPAD introduces partial alignment distillation to weaken the draft model's connection between features and logits, aiming to reduce the conflict between feature alignment and logit confidence during training. Our experiments include both greedy and non-greedy decoding on the largest and smallest models from the Vicuna and LLaMA3-Instruct series, as well as tasks in multi-turn conversation, translation, summarization, question answering, mathematical reasoning, and retrieval-augmented generation. The results show that FSPAD outperforms the state-of-the-art method across all the aforementioned tasks and target LLMs.", "authors": ["Lujun Gui", "Bin Xiao", "Lei Su", "Weipeng Chen"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-28", "url": "https://arxiv.org/abs/2408.15562", "pdf_url": "https://arxiv.org/pdf/2408.15562v1", "arxiv_id": "2408.15562", "doi": "10.48550/arXiv.2408.15562", "citation_count": 9, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.25}
{"id": "c84e0ec48439b9449dc2da31ce84c2a16274440f66ea6ce54f4f0c72dec07993", "sources": ["arxiv", "semantic_scholar"], "title": "Learning Harmonized Representations for Speculative Sampling", "abstract": "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 context between training and decoding. We also observe another discrepancy between the training and decoding objectives in existing speculative sampling methods. In this work, we propose a solution named HArmonized Speculative Sampling (HASS) that learns harmonized representations to address these issues. HASS accelerates the decoding stage without adding inference overhead through harmonized objective distillation and harmonized context alignment. Experiments on four LLaMA models demonstrate that HASS achieves 2.81x-4.05x wall-clock time speedup ratio averaging across three datasets, surpassing EAGLE-2 by 8%-20%. The code is available at https://github.com/HArmonizedSS/HASS.", "authors": ["Lefan Zhang", "Xiaodan Wang", "Yanhua Huang", "Ruiwen Xu"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-28", "url": "https://arxiv.org/abs/2408.15766", "pdf_url": "https://arxiv.org/pdf/2408.15766v3", "arxiv_id": "2408.15766", "doi": null, "citation_count": 67, "influential_citation_count": 5, "has_code": true, "code_url": "https://github.com/HArmonizedSS/HASS", "venue": "International Conference on Learning Representations", "quality_score": 0.4581}
{"id": "ec9d4b656d8c4e193d8b5a0c921a6ead190da563d3d3da5211f26d9aa727fa16", "sources": ["arxiv", "semantic_scholar"], "title": "Can Unconfident LLM Annotations Be Used for Confident Conclusions?", "abstract": "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. Still, guidelines for collecting and using LLM annotations, without compromising the validity of downstream conclusions, remain limited. We introduce Confidence-Driven Inference: a method that combines LLM annotations and LLM confidence indicators to strategically select which human annotations should be collected, with the goal of producing accurate statistical estimates and provably valid confidence intervals while reducing the number of human annotations needed. Our approach comes with safeguards against LLM annotations of poor quality, guaranteeing that the conclusions will be both valid and no less accurate than if we only relied on human annotations. We demonstrate the effectiveness of Confidence-Driven Inference over baselines in statistical estimation tasks across three CSS settings--text politeness, stance, and bias--reducing the needed number of human annotations by over 25% in each. Although we use CSS settings for demonstration, Confidence-Driven Inference can be used to estimate most standard quantities across a broad range of NLP problems.", "authors": ["Kristina Gligorić", "Tijana Zrnic", "Cinoo Lee", "Emmanuel J. Candès", "Dan Jurafsky"], "categories": ["cs.CL", "cs.AI", "cs.HC"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-27", "url": "https://arxiv.org/abs/2408.15204", "pdf_url": "https://arxiv.org/pdf/2408.15204v2", "arxiv_id": "2408.15204", "doi": "10.48550/arXiv.2408.15204", "citation_count": 51, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.429}
{"id": "a29b63ac083c472e85fc6558020bb27a52751eb03841f0e1f70d97e041904732", "sources": ["arxiv", "semantic_scholar"], "title": "The Mamba in the Llama: Distilling and Accelerating Hybrid Models", "abstract": "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 is feasible to distill large Transformers into linear RNNs by reusing the linear projection weights from attention layers with academic GPU resources. The resulting hybrid model, which incorporates a quarter of the attention layers, achieves performance comparable to the original Transformer in chat benchmarks and outperforms open-source hybrid Mamba models trained from scratch with trillions of tokens in both chat benchmarks and general benchmarks. Moreover, we introduce a hardware-aware speculative decoding algorithm that accelerates the inference speed of Mamba and hybrid models. Overall we show how, with limited computation resources, we can remove many of the original attention layers and generate from the resulting model more efficiently. Our top-performing model, distilled from Llama3-8B-Instruct, achieves a 29.61 length-controlled win rate on AlpacaEval 2 against GPT-4 and 7.35 on MT-Bench, surpassing the best 8B scale instruction-tuned linear RNN model. We also find that the distilled model has natural length extrapolation, showing almost perfect accuracy in the needle-in-a-haystack test at 20x the distillation length. Code and pre-trained checkpoints are open-sourced at https://github.com/jxiw/MambaInLlama and https://github.com/itsdaniele/speculative_mamba.", "authors": ["Junxiong Wang", "Daniele Paliotta", "Avner May", "Alexander M. Rush", "Tri Dao"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-27", "url": "https://arxiv.org/abs/2408.15237", "pdf_url": "https://arxiv.org/pdf/2408.15237v4", "arxiv_id": "2408.15237", "doi": "10.48550/arXiv.2408.15237", "citation_count": 124, "influential_citation_count": 21, "has_code": true, "code_url": "https://github.com/jxiw/MambaInLlama", "venue": "Neural Information Processing Systems", "quality_score": 0.6712}
{"id": "fe26000d46f6e01a99726870640909867c4242645cb720f7d388c008f34369b8", "sources": ["arxiv", "semantic_scholar"], "title": "Systematic Evaluation of LLM-as-a-Judge in LLM Alignment Tasks: Explainable Metrics and Diverse Prompt Templates", "abstract": "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 judges and their alignment with human preferences. However, the employed evaluation metrics often lack adequate explainability and fail to address LLM internal inconsistency. Additionally, existing studies inadequately explore the impact of various prompt templates when applying LLM-as-a-Judge methods, leading to potentially inconsistent comparisons between different alignment algorithms. In this work, we systematically evaluate LLM-as-a-Judge on alignment tasks by defining more theoretically interpretable evaluation metrics and explicitly mitigating LLM internal inconsistency from reliability metrics. We develop an open-source framework to evaluate, compare, and visualize the reliability and alignment of LLM judges, which facilitates practitioners to choose LLM judges for alignment tasks. In the experiments, we examine effects of diverse prompt templates on LLM-judge reliability and also demonstrate our developed framework by comparing various LLM judges on two common alignment datasets (i.e., TL;DR Summarization and HH-RLHF-Helpfulness). Our results indicate a significant impact of prompt templates on LLM judge performance, as well as a mediocre alignment level between the tested LLM judges and human evaluators.", "authors": ["Hui Wei", "Shenghua He", "Tian Xia", "Fei Liu", "Andy Wong", "Jingyang Lin", "Mei Han"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-23", "url": "https://arxiv.org/abs/2408.13006", "pdf_url": "https://arxiv.org/pdf/2408.13006v2", "arxiv_id": "2408.13006", "doi": "10.48550/arXiv.2408.13006", "citation_count": 81, "influential_citation_count": 6, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4785}
{"id": "5f75adcab36fdafaeffd2cadfdbad069f1caaf11b8e2d172026d1b921a2f6425", "sources": ["arxiv", "semantic_scholar"], "title": "MagicDec: Breaking the Latency-Throughput Tradeoff for Long Context Generation with Speculative Decoding", "abstract": "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 losslessly, but the conventional wisdom suggests that its efficacy is limited to small batch sizes. In MagicDec, we show that surprisingly SD can achieve speedup even for a high throughput inference regime for moderate to long sequences. More interestingly, an intelligent drafting strategy can achieve better speedup with increasing batch size based on our rigorous analysis. MagicDec first identifies the bottleneck shifts with increasing batch size and sequence length, and uses these insights to deploy SD more effectively for high throughput inference. We leverage draft model with sparse KV cache to address the KV bottleneck, which scales with both sequence length and batch size. Additionally, we propose a theoretical model to select the optimal drafting strategy for maximum speedup. Our work highlights the broad applicability of speculative decoding in long-context serving, as it can enhance throughput and reduce latency without compromising accuracy. For moderate to long sequences, we demonstrate up to 2.51x speedup for Llama3.1-8B when serving batch sizes ranging from 32 to 256 on various types of hardware and tasks.", "authors": ["Ranajoy Sadhukhan", "Jian Chen", "Zhuoming Chen", "Vashisth Tiwari", "Ruihang Lai", "Jinyuan Shi", "Ian En-Hsu Yen", "Avner May", "Tianqi Chen", "Beidi Chen"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-20", "url": "https://arxiv.org/abs/2408.11049", "pdf_url": "https://arxiv.org/pdf/2408.11049v5", "arxiv_id": "2408.11049", "doi": "10.48550/arXiv.2408.11049", "citation_count": 83, "influential_citation_count": 10, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.5207}
{"id": "064a394a74613620e565b507fd44c2eec7f1cf2e039d89799b1d5825dd9dbe87", "sources": ["arxiv", "semantic_scholar"], "title": "Strategist: Self-improvement of LLM Decision Making via Bi-Level Tree Search", "abstract": "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 action spaces. We introduce STRATEGIST, a novel approach that integrates the strengths of both methods. Our approach leverages LLMs to search and update high-level strategies (as text), which are then refined and executed by low-level Monte Carlo Tree Search (MCTS). STRATEGIST is a generalizable framework to optimize the strategy through population-based self-play simulations without the need for any training data. We demonstrate the effectiveness of STRATEGIST in learning optimal strategies for competitive, multi-turn games with partial information, including Game of Pure Strategy (GOPS) and multi-agent, hidden-identity discussion games like The Resistance: Avalon. Our results show that agents equipped with STRATEGIST outperform those trained with traditional RL methods, other LLM-based skill acquisition techniques, pre-existing LLM agents across both game environments and achieves comparable performance against human players.", "authors": ["Jonathan Light", "Min Cai", "Weiqin Chen", "Guanzhi Wang", "Xiusi Chen", "Wei Cheng", "Yisong Yue", "Ziniu Hu"], "categories": ["cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-20", "url": "https://arxiv.org/abs/2408.10635", "pdf_url": "https://arxiv.org/pdf/2408.10635v3", "arxiv_id": "2408.10635", "doi": null, "citation_count": 14, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.294}
{"id": "ed971c5dd045a30c17672f24c34a6829477feb4f08b770fbaf3144b2ae1256ab", "sources": ["arxiv", "semantic_scholar"], "title": "KOALA: Enhancing Speculative Decoding for LLM via Multi-Layer Draft Heads with Adversarial Learning", "abstract": "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 approach to the draft head. By transforming the conventional single-layer draft head into a multi-layer architecture and incorporating adversarial learning into the traditional supervised training, KOALA significantly improves the accuracy of the draft head in predicting subsequent tokens, thus more closely mirroring the functionality of LLMs. Although this improvement comes at the cost of slightly increased drafting overhead, KOALA substantially unlocks the draft head's potential, greatly enhancing speculative decoding. We conducted comprehensive evaluations of KOALA, including both autoregressive and non-autoregressive draft heads across various tasks, demonstrating a latency speedup ratio improvement of 0.24x-0.41x, which is 10.57%-14.09% faster than the original draft heads.", "authors": ["Kaiqi Zhang", "Jing Zhao", "Rui Chen"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-15", "url": "https://arxiv.org/abs/2408.08146", "pdf_url": "https://arxiv.org/pdf/2408.08146v1", "arxiv_id": "2408.08146", "doi": "10.1109/CSCWD64889.2025.11033265", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "International Conference on Computer Supported Cooperative Work in Design", "quality_score": 0.1945}
{"id": "1f7a4cd0af7d7b9ce1dcec72f540aeaef944b909c1210f7c005ab9a2853612f3", "sources": ["arxiv", "semantic_scholar"], "title": "PEARL: Parallel Speculative Decoding with Adaptive Draft Length", "abstract": "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 model gets stuck when the draft model is guessing tokens, and vice versa. This problem is directly incurred by the asynchronous execution of the draft model and the target model and is exacerbated due to the fixed draft length in speculative decoding. To address these challenges, we propose a conceptually simple, flexible, and general framework to boost speculative decoding, namely Parallel spEculative decoding with Adaptive dRaft Length (PEARL). Specifically, PEARL proposes pre-verify to verify the first draft token in advance during the drafting phase, and post-verify to generate more draft tokens during the verification phase. PEARL parallels the drafting phase and the verification phase via applying the two strategies, and achieves adaptive draft length for different scenarios, which effectively alleviates the mutual waiting problem. Experiments on various text generation benchmarks demonstrate the effectiveness of our PEARL, leading to a superior speed up performance up to 4.43$\\times$ and 1.50$\\times$, compared to auto-regressive decoding and vanilla speculative decoding, respectively. Our code is available at https://github.com/smart-lty/ParallelSpeculativeDecoding.", "authors": ["Tianyu Liu", "Yun Li", "Qitan Lv", "Kai Liu", "Jianchen Zhu", "Winston Hu", "Xiao Sun"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-13", "url": "https://arxiv.org/abs/2408.11850", "pdf_url": "https://arxiv.org/pdf/2408.11850v3", "arxiv_id": "2408.11850", "doi": "10.48550/arXiv.2408.11850", "citation_count": 74, "influential_citation_count": 7, "has_code": true, "code_url": "https://github.com/smart-lty/ParallelSpeculativeDecoding", "venue": "International Conference on Learning Representations", "quality_score": 0.4688}
{"id": "18c86cd7db58bc39263e38afc6c13f2693fc2e798c8050328a6fdc25ecf378df", "sources": ["arxiv", "semantic_scholar"], "title": "Efficiency Unleashed: Inference Acceleration for LLM-based Recommender Systems with Speculative Decoding", "abstract": "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 scenarios with numerous users and items, even offline knowledge generation with LLMs demands significant time and computational resources. This inefficiency arises from the autoregressive nature of LLMs. A promising solution is speculative decoding, a Draft-Then-Verify approach that increases the number of tokens generated per decoding step. In this work, we first identify recommendation knowledge generation as a highly fitting use case for retrieval-based speculative decoding. Then, we discern its two characteristics: (1) the vast number of items and users in RSs leads to retrieval inefficiency, and (2) RSs exhibit high diversity tolerance for LLM-generated text. Building on these insights, we introduce Lossless Acceleration via Speculative Decoding for LLM-based Recommender Systems (LASER), which features a Customized Retrieval Pool to enhance retrieval efficiency and Relaxed Verification to improve the acceptance rate of draft tokens. LASER achieves a 3-5x speedup on public datasets and saves about 67\\% of computational resources during the online A/B test on a large-scale advertising scenario with lossless downstream recommendation performance. Our code is available at https://github.com/YunjiaXi/LASER", "authors": ["Yunjia Xi", "Hangyu Wang", "Bo Chen", "Jianghao Lin", "Menghui Zhu", "Weiwen Liu", "Ruiming Tang", "Zhewei Wei", "Weinan Zhang", "Yong Yu"], "categories": ["cs.IR"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-11", "url": "https://arxiv.org/abs/2408.05676", "pdf_url": "https://arxiv.org/pdf/2408.05676v2", "arxiv_id": "2408.05676", "doi": "10.1145/3726302.3729961", "citation_count": 14, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/YunjiaXi/LASER", "venue": "Annual International ACM SIGIR Conference on Research and Development in Information Retrieval", "quality_score": 0.294}
{"id": "7b64842ab54d8e6b89a05d0da83d12278a66987bc33f0344ae34e79d33a612be", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Diffusion Decoding: Accelerating Language Generation through Diffusion", "abstract": "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 reliance on incremental token generation in existing draft models. To overcome this limitation, this paper proposes an adaptation of speculative decoding which uses discrete diffusion models to generate draft sequences. This allows parallelization of both the drafting and verification steps, providing significant speedups to the inference process. Our proposed approach, $\\textit{Speculative Diffusion Decoding (SpecDiff)}$, is validated on standard language generation benchmarks and empirically demonstrated to provide up to 7.2x speedups over standard generation processes and up to 1.75x speedups over existing speculative decoding approaches.", "authors": ["Jacob K Christopher", "Brian R Bartoldson", "Tal Ben-Nun", "Michael Cardei", "Bhavya Kailkhura", "Ferdinando Fioretto"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-10", "url": "https://arxiv.org/abs/2408.05636", "pdf_url": "https://arxiv.org/pdf/2408.05636v4", "arxiv_id": "2408.05636", "doi": "10.48550/arXiv.2408.05636", "citation_count": 47, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.4203}
{"id": "a90f81ad13efb3e6dc609d5b3bd739aca0301530705a00c1ff9fc9f6f82da3f0", "sources": ["arxiv", "semantic_scholar"], "title": "CREST: Effectively Compacting a Datastore For Retrieval-Based Speculative Decoding", "abstract": "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 is to only store a subset of the smallest and most common n-grams in the datastore with the hope of achieving comparable performance with less storage space. We found that storing a subset of n-grams both reduces storage space and improves performance. CREST matches REST's accepted token length with 10.6-13.5x less storage space and achieves a 16.5-17.1% higher acceptance length than REST using the same storage space on the HumanEval and MT Bench benchmarks.", "authors": ["Sophia Ho", "Jinsol Park", "Patrick Wang"], "categories": ["cs.CL", "cs.AI", "cs.DB"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-08", "url": "https://arxiv.org/abs/2408.04678", "pdf_url": "https://arxiv.org/pdf/2408.04678v1", "arxiv_id": "2408.04678", "doi": "10.48550/arXiv.2408.04678", "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.0}
{"id": "ddb871574bb02d70447dd8b089e6c55acd743758688826735cdca4e2ef78326f", "sources": ["arxiv", "semantic_scholar"], "title": "Clover-2: Accurate Inference for Regressive Lightweight Speculative Decoding", "abstract": "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 text generation tasks. This approach utilizes a lightweight regressive draft model, like a Recurrent Neural Network (RNN) or a single transformer decoder layer, leveraging sequential information to iteratively predict potential tokens. Specifically, RNN draft models are computationally economical but tend to deliver lower accuracy, while attention decoder layer models exhibit the opposite traits. This paper presents Clover-2, an advanced iteration of Clover, an RNN-based draft model designed to achieve comparable accuracy to that of attention decoder layer models while maintaining minimal computational overhead. Clover-2 enhances the model architecture and incorporates knowledge distillation to increase Clover's accuracy and improve overall efficiency. We conducted experiments using the open-source Vicuna 7B and LLaMA3-Instruct 8B models. The results demonstrate that Clover-2 surpasses existing methods across various model architectures, showcasing its efficacy and robustness.", "authors": ["Bin Xiao", "Lujun Gui", "Lei Su", "Weipeng Chen"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-08-01", "url": "https://arxiv.org/abs/2408.00264", "pdf_url": "https://arxiv.org/pdf/2408.00264v1", "arxiv_id": "2408.00264", "doi": "10.48550/arXiv.2408.00264", "citation_count": 5, "influential_citation_count": 0, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1945}
{"id": "428718b4c19974153cddfdd4f2beab5591a5f3bd22a531cdb196e0658f9d287a", "sources": ["arxiv", "semantic_scholar"], "title": "Graph-Structured Speculative Decoding", "abstract": "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 efficiency of the draft model. In our research, we focus on enhancing the proportion of draft tokens that are accepted to the final output by generating multiple hypotheses instead of just one. This allows the LLM more options to choose from and select the longest sequence that meets its standards. Our analysis reveals that hypotheses produced by the draft model share many common token sequences, suggesting a potential for optimizing computation. Leveraging this observation, we introduce an innovative approach utilizing a directed acyclic graph (DAG) to manage the drafted hypotheses. This structure enables us to efficiently predict and merge recurring token sequences, vastly reducing the computational demands of the draft model. We term this approach Graph-structured Speculative Decoding (GSD). We apply GSD across a range of LLMs, including a 70-billion parameter LLaMA-2 model, and observe a remarkable speedup of 1.73$\\times$ to 1.96$\\times$, significantly surpassing standard speculative decoding.", "authors": ["Zhuocheng Gong", "Jiahao Liu", "Ziyue Wang", "Pengfei Wu", "Jingang Wang", "Xunliang Cai", "Dongyan Zhao", "Rui Yan"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-07-23", "url": "https://arxiv.org/abs/2407.16207", "pdf_url": "https://arxiv.org/pdf/2407.16207v1", "arxiv_id": "2407.16207", "doi": "10.48550/arXiv.2407.16207", "citation_count": 7, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.2258}
{"id": "dadbb611d1d138a6f11d071145604220e650be30d6a7897cbe4583908502ed7c", "sources": ["arxiv", "semantic_scholar"], "title": "PipeInfer: Accelerating LLM Inference using Asynchronous Pipelined Speculation", "abstract": "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 inference run, requiring high speculation acceptance rates to improve performance. Combined with a variable rate of acceptance across tasks, speculative inference techniques can result in reduced performance. Additionally, pipeline-parallel designs require many user requests to maintain maximum utilization. As a remedy, we propose PipeInfer, a pipelined speculative acceleration technique to reduce inter-token latency and improve system utilization for single-request scenarios while also improving tolerance to low speculation acceptance rates and low-bandwidth interconnects. PipeInfer exhibits up to a 2.15$\\times$ improvement in generation speed over standard speculative inference. PipeInfer achieves its improvement through Continuous Asynchronous Speculation and Early Inference Cancellation, the former improving latency and generation speed by running single-token inference simultaneously with several speculative runs, while the latter improves speed and latency by skipping the computation of invalidated runs, even in the middle of inference.", "authors": ["Branden Butler", "Sixing Yu", "Arya Mazaheri", "Ali Jannesari"], "categories": ["cs.CL", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-07-16", "url": "https://arxiv.org/abs/2407.11798", "pdf_url": "https://arxiv.org/pdf/2407.11798v2", "arxiv_id": "2407.11798", "doi": "10.1109/SC41406.2024.00046", "citation_count": 27, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "International Conference for High Performance Computing, Networking, Storage and Analysis", "quality_score": 0.3618}
{"id": "306d6d9f9fded19d1a3ca7281bec07eeac0aff5e30c21e5aa6d31c11b7dd5e77", "sources": ["arxiv", "semantic_scholar"], "title": "Optimized Multi-Token Joint Decoding with Auxiliary Model for LLM Inference", "abstract": "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 multiple tokens per step, each token is still generated by its single-token distribution, thereby enhancing speed without improving effectiveness. In contrast, our work simultaneously enhances inference speed and improves the output effectiveness. We consider multi-token joint decoding (MTJD), which generates multiple tokens from their joint distribution at each iteration, theoretically reducing perplexity and enhancing task performance. However, MTJD suffers from the high cost of sampling from the joint distribution of multiple tokens. Inspired by speculative decoding, we introduce multi-token assisted decoding (MTAD), a novel framework designed to accelerate MTJD. MTAD leverages a smaller auxiliary model to approximate the joint distribution of a larger model, incorporating a verification mechanism that not only ensures the accuracy of this approximation, but also improves the decoding efficiency over conventional speculative decoding. Theoretically, we demonstrate that MTAD closely approximates exact MTJD with bounded error. Empirical evaluations using Llama-2 and OPT models ranging from 13B to 70B parameters across various tasks reveal that MTAD reduces perplexity by 21.2% and improves downstream performance compared to standard single-token sampling. Furthermore, MTAD achieves a 1.42x speed-up and consumes 1.54x less energy than conventional speculative decoding methods. These results highlight MTAD's ability to make multi-token joint decoding both effective and efficient, promoting more sustainable and high-performance deployment of LLMs.", "authors": ["Zongyue Qin", "Ziniu Hu", "Zifan He", "Neha Prakriya", "Jason Cong", "Yizhou Sun"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-07-12", "url": "https://arxiv.org/abs/2407.09722", "pdf_url": "https://arxiv.org/pdf/2407.09722v4", "arxiv_id": "2407.09722", "doi": null, "citation_count": 12, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.2785}
{"id": "1f783f2b3fd029ca1701073fd121b350dbc6477cc27e408a97abe79a291431ed", "sources": ["arxiv", "semantic_scholar"], "title": "FBI-LLM: Scaling Up Fully Binarized LLMs from Scratch via Autoregressive Distillation", "abstract": "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-based LLMs. It achieves this by employing an autoregressive distillation (AD) loss with maintaining equivalent model dimensions (130M, 1.3B, 7B) and training data volume as regular LLM pretraining, while delivering competitive results in terms of perplexity and task-specific effectiveness. Intriguingly, by analyzing the training trajectory, we find that the pretrained weight is not necessary for training binarized LLMs from scratch. This research encourages a new computational framework and may facilitate the future design of specialized hardware tailored for fully 1-bit LLMs. We make all models, code, and training dataset fully accessible and transparent to support further research (Code: https://github.com/LiqunMa/FBI-LLM. Model: https://huggingface.co/LiqunMa/).", "authors": ["Liqun Ma", "Mingjie Sun", "Zhiqiang Shen"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-07-09", "url": "https://arxiv.org/abs/2407.07093", "pdf_url": "https://arxiv.org/pdf/2407.07093v1", "arxiv_id": "2407.07093", "doi": "10.48550/arXiv.2407.07093", "citation_count": 15, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/LiqunMa/FBI-LLM", "venue": "arXiv.org", "quality_score": 0.301}
{"id": "abe057f735ac280b7723f4fa0f7986ffa76606d2d551e291ba01f34cbdbb7427", "sources": ["arxiv", "semantic_scholar"], "title": "S2D: Sorted Speculative Decoding For More Efficient Deployment of Nested Large Language Models", "abstract": "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 promising approaches to speed up the LLM decoding process by verifying multiple tokens in parallel and using an auxiliary smaller draft model to generate the possible tokens. In SD, usually, one draft model is used to serve a specific target model; however, in practice, LLMs are diverse, and we might need to deal with many target models or more than one target model simultaneously. In this scenario, it is not clear which draft model should be used for which target model, and searching among different draft models or training customized draft models can further increase deployment costs. In this paper, we first introduce a novel multi-target scenario for the deployment of draft models for faster inference. Then, we present a novel, more efficient sorted speculative decoding mechanism that outperforms regular baselines in multi-target settings. We evaluated our method on Spec-Bench in different settings, including base models such as Vicuna 7B, 13B, and LLama Chat 70B. Our results suggest that our draft models perform better than baselines for multiple target models at the same time.", "authors": ["Parsa Kavehzadeh", "Mohammadreza Pourreza", "Mojtaba Valipour", "Tinashu Zhu", "Haoli Bai", "Ali Ghodsi", "Boxing Chen", "Mehdi Rezagholizadeh"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-07-02", "url": "https://arxiv.org/abs/2407.01955", "pdf_url": "https://arxiv.org/pdf/2407.01955v1", "arxiv_id": "2407.01955", "doi": "10.48550/arXiv.2407.01955", "citation_count": 1, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.0753}
{"id": "db3aa7aa458b66fd5617181280bdabef432e81fe16d346e8fe5d3503e195b44f", "sources": ["arxiv", "semantic_scholar"], "title": "SEED: Accelerating Reasoning Tree Construction via Scheduled Speculative Decoding", "abstract": "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, such methods introduce significant inference latency due to the systematic exploration and evaluation of multiple thought paths. This paper introduces SeeD, a novel and efficient inference framework to optimize runtime speed and GPU memory management concurrently. By employing a scheduled speculative execution, SeeD efficiently handles multiple iterations for the thought generation and the state evaluation, leveraging a rounds-scheduled strategy to manage draft model dispatching. Extensive experimental evaluations on three reasoning datasets demonstrate superior speedup performance of SeeD, providing a viable path for batched inference in training-free speculative decoding.", "authors": ["Zhenglin Wang", "Jialong Wu", "Yilong Lai", "Congzhi Zhang", "Deyu Zhou"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-26", "url": "https://arxiv.org/abs/2406.18200", "pdf_url": "https://arxiv.org/pdf/2406.18200v2", "arxiv_id": "2406.18200", "doi": "10.48550/arXiv.2406.18200", "citation_count": 13, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "International Conference on Computational Linguistics", "quality_score": 0.2865}
{"id": "ce3dad317ad49b0f63949f786033f1df3dc130c3c458aae6a317a60d2cdf07de", "sources": ["arxiv", "semantic_scholar"], "title": "OPT-Tree: Speculative Decoding with Adaptive Draft Tree Structure", "abstract": "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\" mechanism to allow multiple tokens to be generated in one step, realizing lossless acceleration. Existing methods mainly adopt fixed heuristic draft structures, which fail to adapt to different situations to maximize the acceptance length during verification. To alleviate this dilemma, we proposed OPT-Tree, an algorithm to construct adaptive and scalable draft trees. It searches the optimal tree structure that maximizes the mathematical expectation of the acceptance length in each decoding step. Experimental results reveal that OPT-Tree outperforms the existing draft structures and achieves a speed-up ratio of up to 3.2 compared with autoregressive decoding. If the draft model is powerful enough and the node budget is sufficient, it can generate more than ten tokens in a single step. Our code is available at https://github.com/Jikai0Wang/OPT-Tree.", "authors": ["Jikai Wang", "Yi Su", "Juntao Li", "Qingrong Xia", "Zi Ye", "Xinyu Duan", "Zhefeng Wang", "Min Zhang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-25", "url": "https://arxiv.org/abs/2406.17276", "pdf_url": "https://arxiv.org/pdf/2406.17276v4", "arxiv_id": "2406.17276", "doi": "10.48550/arXiv.2406.17276", "citation_count": 56, "influential_citation_count": 6, "has_code": true, "code_url": "https://github.com/Jikai0Wang/OPT-Tree", "venue": "Transactions of the Association for Computational Linguistics", "quality_score": 0.439}
{"id": "ef379ac790a45ee522ae1f4ab05fc54855c7cefd14099389190ee49b7bb256b3", "sources": ["arxiv", "semantic_scholar"], "title": "Towards Fast Multilingual LLM Inference: Speculative Decoding and Specialized Drafters", "abstract": "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 of an assistant model in speculative decoding, which is leveraged to draft and-then its future tokens are verified by the target LLM. We show that language-specific draft models, optimized through a targeted pretrain-and-finetune strategy, substantially brings a speedup in inference time compared to the previous methods. We validate these models across various languages in inference time, out-of-domain speedup, and GPT-4o evaluation.", "authors": ["Euiin Yi", "Taehyeon Kim", "Hongseok Jeung", "Du-Seong Chang", "Se-Young Yun"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-24", "url": "https://arxiv.org/abs/2406.16758", "pdf_url": "https://arxiv.org/pdf/2406.16758v2", "arxiv_id": "2406.16758", "doi": "10.48550/arXiv.2406.16758", "citation_count": 13, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.2865}
{"id": "8034281d98b6ffbc0501915288535e43ee0f5b8602f82c1b8241714cf633bec2", "sources": ["arxiv", "semantic_scholar"], "title": "TurboSpec: Closed-loop Speculation Control System for Optimizing LLM Serving Goodput", "abstract": "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 from speculative decoding as a solution to this problem. Unfortunately, benefits from intra-request parallelism are often fragile, as speculative decoding causes overhead, and speculated tokens may miss. We observe that speculative decoding may degrade LLM serving performance if added naively without tuning to the incoming requests and the speculation method. To alleviate the need for expert tuning and make speculative decoding more robust, we present TurboSpec, a speculation control system that automatically profiles the execution environment and utilizes a feedback-based algorithm to dynamically adjust the amount of intra-request parallelism in LLM serving. TurboSpec predicts \"goodput\" - the amount of successfully generated tokens - to evaluate and adjust intra-request parallelism amount to that with the highest goodput in runtime. We implement TurboSpec on a real-world LLM serving system vLLM and demonstrate its effectiveness across diverse workloads and hardware configurations, providing consistent performance improvements across all test scenarios.", "authors": ["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"], "categories": ["cs.AI", "cs.PF"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-20", "url": "https://arxiv.org/abs/2406.14066", "pdf_url": "https://arxiv.org/pdf/2406.14066v3", "arxiv_id": "2406.14066", "doi": null, "citation_count": 33, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3829}
{"id": "22e0561929a8c2460264287c414e6f1894c51b738d452f52cd8381d1413867c9", "sources": ["arxiv", "semantic_scholar"], "title": "Amphista: Bi-directional Multi-head Decoding for Accelerating LLM Inference", "abstract": "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 limitation, we introduce Amphista, an enhanced speculative decoding framework that builds upon Medusa. Specifically, Amphista models an Auto-embedding Block capable of parallel inference, incorporating bi-directional attention to enable interaction between different drafting heads. Additionally, Amphista integrates Staged Adaptation Layers, which ensure a seamless transition of semantic information from the target model's autoregressive inference to the drafting heads' non-autoregressive inference, effectively achieving paradigm shift and feature fusion. Experimental results on Vicuna models using MT-Bench and Spec-Bench demonstrate that Amphista achieves substantial acceleration while maintaining generation quality. On MT-Bench, Amphista delivers up to 2.75$\\times$ speedup over vanilla autoregressive decoding and 1.40$\\times$ over Medusa on Vicuna 33B in wall-clock time.", "authors": ["Zeping Li", "Xinlong Yang", "Ziheng Gao", "Ji Liu", "Guanchen Li", "Zhuang Liu", "Dong Li", "Jinzhang Peng", "Lu Tian", "Emad Barsoum"], "categories": ["cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-19", "url": "https://arxiv.org/abs/2406.13170", "pdf_url": "https://arxiv.org/pdf/2406.13170v2", "arxiv_id": "2406.13170", "doi": "10.18653/v1/2025.naacl-long.450", "citation_count": 2, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.1193}
{"id": "785c8f41d6f86de01e8a77ecd1765a09b127639d7e6e945e404c5c79645ec71c", "sources": ["arxiv", "semantic_scholar"], "title": "Open-LLM-Leaderboard: From Multi-choice to Open-style Questions for LLMs Evaluation, Benchmark, and Arena", "abstract": "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 biases of priori unbalanced probabilities, influencing the prediction of answers based on these IDs. Previous research has introduced methods to reduce this ''selection bias'' by simply permutating options on a few test samples and applying to new ones. Another problem of MCQ is the lottery ticket choice by ''random guessing''. The LLM does not learn particular knowledge, but the option is guessed correctly. This situation is especially serious for those small-scale LLMs. To address them, a more thorough approach involves shifting from MCQ to open-style questions, which can fundamentally eliminate selection bias and random guessing issues. However, transitioning causes its own set of challenges in (1) identifying suitable open-style questions and (2) validating the correctness of LLM open-style responses against human-annotated ground-truths. This work aims to tackle these significant difficulties, and establish a new LLM evaluation benchmark through entirely open-style questions. Consequently, we introduce the Open-LLM-Leaderboard to track various LLMs' performance and reflect true capability of them, such as GPT-4o/4/3.5, Claude 3, Gemini, etc. Our code and dataset are available at https://github.com/VILA-Lab/Open-LLM-Leaderboard.", "authors": ["Aidar Myrzakhan", "Sondos Mahmoud Bsharat", "Zhiqiang Shen"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-11", "url": "https://arxiv.org/abs/2406.07545", "pdf_url": "https://arxiv.org/pdf/2406.07545v1", "arxiv_id": "2406.07545", "doi": "10.48550/arXiv.2406.07545", "citation_count": 86, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/VILA-Lab/Open-LLM-Leaderboard", "venue": "arXiv.org", "quality_score": 0.4849}
{"id": "5eb4eb7cd49f02bd42c0bf866f40f445d115c2fcad865fd5d686690df1124f9f", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding via Early-exiting for Faster LLM Inference with Thompson Sampling Control Mechanism", "abstract": "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. Specifically, EESD utilizes a segment of the LLM to generate draft tokens, incorporating Early-exiting structures after the first N layers. To enhance the quality of draft tokens, a self-distillation method is integrated. This early-exiting design not only reduces deployment and training costs but also significantly accelerates the token generation speed. Moreover, we introduce a novel sampling mechanism that leverages Thompson Sampling to regulate the generation processes, automatically determining the quantity of draft tokens in each round. The original LLM is then employed to validate these draft tokens through a single forward pass, and thus guarantees that the final output text maintains a distribution consistent with vanilla auto-regressive decoding. The experimental results on both 13B and 70B models demonstrate that our approach decodes tokens at a markedly accelerated rate compared to prior methods, showing the effectiveness of our approach.", "authors": ["Jiahao Liu", "Qifan Wang", "Jingang Wang", "Xunliang Cai"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-06", "url": "https://arxiv.org/abs/2406.03853", "pdf_url": "https://arxiv.org/pdf/2406.03853v1", "arxiv_id": "2406.03853", "doi": "10.48550/arXiv.2406.03853", "citation_count": 37, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.3949}
{"id": "ef5db1ddb7159d4aba37bcce453e447b78f8d5c9213b6a510357586cb1688d9d", "sources": ["arxiv", "semantic_scholar"], "title": "SpecExec: Massively Parallel Speculative Decoding for Interactive LLM Inference on Consumer Devices", "abstract": "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: how fast can we run LLMs on consumer machines? Consumer GPUs can no longer fit the largest available models (50B+ parameters) and must offload them to RAM or SSD. When running with offloaded parameters, the inference engine can process batches of hundreds or thousands of tokens at the same time as just one token, making it a natural fit for speculative decoding. We propose SpecExec (Speculative Execution), a simple parallel decoding method that can generate up to 20 tokens per target model iteration for popular LLM families. It utilizes the high spikiness of the token probabilities distribution in modern LLMs and a high degree of alignment between model output probabilities. SpecExec takes the most probable tokens continuation from the draft model to build a \"cache\" tree for the target model, which then gets validated in a single pass. Using SpecExec, we demonstrate inference of 50B+ parameter LLMs on consumer GPUs with RAM offloading at 4-6 tokens per second with 4-bit quantization or 2-3 tokens per second with 16-bit weights.", "authors": ["Ruslan Svirschevski", "Avner May", "Zhuoming Chen", "Beidi Chen", "Zhihao Jia", "Max Ryabinin"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-04", "url": "https://arxiv.org/abs/2406.02532", "pdf_url": "https://arxiv.org/pdf/2406.02532v3", "arxiv_id": "2406.02532", "doi": "10.48550/arXiv.2406.02532", "citation_count": 56, "influential_citation_count": 5, "has_code": false, "code_url": null, "venue": "Neural Information Processing Systems", "quality_score": 0.439}
{"id": "7abac9e41e114e05fbc1fbeb490ba5b3f3bf45535c9f583bbac3f5c19a5bfb5d", "sources": ["arxiv", "semantic_scholar"], "title": "Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller", "abstract": "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 LLM's self-evaluation of the suffix with respect to its latent representations. The gradients are used to directly control the auto-regressive generation process towards desired behaviors, which eliminates human supervision, achieves precise and transparent control, and offers on-the-fly adaptability. To further enhance efficiency, we introduce SelfControl_{Prefix}, a compact module that encapsulates the learned representations from gradients into a SelfControl_{Prefix}, facilitating efficient inference-time control with no latency compared to the original model and allowing control for multiple behaviors simultaneously. Our experiments demonstrate SelfControl's efficacy across multiple domains, where it improves over SOTA for 8.3% in detoxification, 3.1% in truthfulness enhancement, 4%~10% in controlling on emotion tones, and 48.2% in privacy protection, i.e., completely remove privacy leakage issue. Additionally, we demonstrate that SelfControl can be used for data synthesis and to improve reasoning abilities.", "authors": ["Min Cai", "Yuchen Zhang", "Shichang Zhang", "Fan Yin", "Dan Zhang", "Difan Zou", "Yisong Yue", "Ziniu Hu"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-04", "url": "https://arxiv.org/abs/2406.02721", "pdf_url": "https://arxiv.org/pdf/2406.02721v3", "arxiv_id": "2406.02721", "doi": "10.48550/arXiv.2406.02721", "citation_count": 5, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.1945}
{"id": "5d510621df1af61993e486569d632a3169160ed273a382ffb3269724f1bcce10", "sources": ["arxiv", "semantic_scholar"], "title": "Demystifying AI Platform Design for Distributed Inference of Next-Generation LLM models", "abstract": "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. With constant innovation in LLM serving optimizations and model architecture evolving at breakneck speed, the hardware requirements to meet Service Level Objectives (SLOs) remain an open research question. To answer the question, we present an analytical tool, GenZ, to efficiently navigate the relationship between diverse LLM model architectures(Dense, GQA, MoE, Mamba), LLM serving optimizations(Chunking, Speculative decoding, quanitization), and AI platform design parameters. Our tool estimates LLM inference performance metrics for the given scenario. We have validated against real hardware platforms running various different LLM models, achieving a max geomean error of 5.82.We use GenZ to identify compute, memory capacity, memory bandwidth, network latency, and network bandwidth requirements across diverse LLM inference use cases. We also study diverse architectural choices in use today (inspired by LLM serving platforms from several vendors) to help inform computer architects designing next-generation AI hardware accelerators and platforms. The trends and insights derived from GenZ can guide AI engineers deploying LLMs as well as computer architects designing next-generation hardware accelerators and platforms. Ultimately, this work sheds light on the platform design considerations for unlocking the full potential of large language models across a spectrum of applications. The source code is available at https://github.com/abhibambhaniya/GenZ-LLM-Analyzer . Users can also be tried it on at https://genz-llm-analyzer.streamlit.app/ without any setup on your web browser.", "authors": ["Abhimanyu Bambhaniya", "Ritik Raj", "Geonhwa Jeong", "Souvik Kundu", "Sudarshan Srinivasan", "Suvinay Subramanian", "Midhilesh Elavazhagan", "Madhu Kumar", "Tushar Krishna"], "categories": ["cs.AR", "cs.AI", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-06-03", "url": "https://arxiv.org/abs/2406.01698", "pdf_url": "https://arxiv.org/pdf/2406.01698v3", "arxiv_id": "2406.01698", "doi": null, "citation_count": 12, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/abhibambhaniya/GenZ-LLM-Analyzer", "venue": null, "quality_score": 0.2785}
{"id": "e73f3105161668b763b4acd4585d5612ad49ff1af88863abc2bb8070eb9fb246", "sources": ["arxiv", "semantic_scholar"], "title": "Snowflake: A Distributed Streaming Decoder", "abstract": "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, local implementation. In designing Snowflake, we propose a new method for general stream decoding that eliminates the processing overhead due to window overlap in existing windowing methods.", "authors": ["Tim Chan"], "categories": ["quant-ph"], "fields_of_study": ["Physics", "Computer Science"], "published_date": "2024-06-03", "url": "https://arxiv.org/abs/2406.01701", "pdf_url": "https://arxiv.org/pdf/2406.01701v3", "arxiv_id": "2406.01701", "doi": "10.22331/q-2026-03-20-2033", "citation_count": 4, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "Quantum", "quality_score": 0.1747}
{"id": "935b4da18d05ec11f70059c8a9b4e7a8ebca75c36f59561389ef81144faac39e", "sources": ["arxiv", "semantic_scholar"], "title": "S3D: A Simple and Cost-Effective Self-Speculative Decoding Scheme for Low-Memory GPUs", "abstract": "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. Given limited memory and the necessity of quantization, a high-performing model on a high-end GPU can slow down by up to 7 times. To this end, we propose Skippy Simultaneous Speculative Decoding (or S3D), a cost-effective self-speculative SD method based on simultaneous multi-token decoding and mid-layer skipping. When compared against recent effective open-source SD systems, our method has achieved one of the top performance-memory ratios while requiring minimal architecture changes and training data. Leveraging our memory efficiency, we created a smaller yet more effective SD model based on Phi-3. It is 1.4 to 2 times faster than the quantized EAGLE model and operates in half-precision while using less VRAM.", "authors": ["Wei Zhong", "Manasa Bharadwaj"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-30", "url": "https://arxiv.org/abs/2405.20314", "pdf_url": "https://arxiv.org/pdf/2405.20314v2", "arxiv_id": "2405.20314", "doi": "10.48550/arXiv.2405.20314", "citation_count": 10, "influential_citation_count": 0, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2603}
{"id": "356d9d45e8d7e97dd240cd314d1c3c84b5ec0bd11169ac96e41b4e0b6a67260e", "sources": ["arxiv", "semantic_scholar"], "title": "SpecDec++: Boosting Speculative Decoding via Adaptive Candidate Lengths", "abstract": "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 use simple heuristics to choose K, which may result in sub-optimal performance. We study the choice of the candidate length K and formulate it as a Markov Decision Process. We theoretically show that the optimal policy of this Markov decision process takes the form of a threshold policy, i.e., the current speculation should stop and be verified when the probability of getting a rejection exceeds a threshold value. Motivated by this theory, we propose SpecDec++, an enhanced version of speculative decoding that adaptively determines the candidate length on the fly. We augment the draft model with a trained acceptance prediction head to predict the conditional acceptance probability of the candidate tokens. SpecDec++ will stop the current speculation when the predicted probability that at least one token gets rejected exceeds a threshold. We implement SpecDec++ and apply it to the llama-2-chat 7B & 70B model pair. Our adaptive method achieves a 2.04x speedup on the Alpaca dataset (7.2% improvement over the baseline speculative decoding). On the GSM8K and HumanEval datasets, our method achieves a 2.26x speedup (9.4% improvement) and 2.23x speedup (11.1% improvement), respectively. The code of this paper is available at https://github.com/Kaffaljidhmah2/SpecDec_pp.", "authors": ["Kaixuan Huang", "Xudong Guo", "Mengdi Wang"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-30", "url": "https://arxiv.org/abs/2405.19715", "pdf_url": "https://arxiv.org/pdf/2405.19715v3", "arxiv_id": "2405.19715", "doi": "10.48550/arXiv.2405.19715", "citation_count": 56, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/Kaffaljidhmah2/SpecDec_pp", "venue": "arXiv.org", "quality_score": 0.439}
{"id": "5fc24943753a416bc170e82fa881004a68ec663c429acf2880295913352ba6ab", "sources": ["arxiv", "semantic_scholar"], "title": "Nearest Neighbor Speculative Decoding for LLM Generation and Attribution", "abstract": "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 often exhibit slow inference speeds and produce non-fluent texts. In this paper, we introduce Nearest Neighbor Speculative Decoding (NEST), a novel semi-parametric language modeling approach that is capable of incorporating real-world text spans of arbitrary length into the LM generations and providing attribution to their sources. NEST performs token-level retrieval at each inference step to compute a semi-parametric mixture distribution and identify promising span continuations in a corpus. It then uses an approximate speculative decoding procedure that accepts a prefix of the retrieved span or generates a new token. NEST significantly enhances the generation quality and attribution rate of the base LM across a variety of knowledge-intensive tasks, surpassing the conventional kNN-LM method and performing competitively with in-context retrieval augmentation. In addition, NEST substantially improves the generation speed, achieving a 1.8x speedup in inference time when applied to Llama-2-Chat 70B. Code will be released at https://github.com/facebookresearch/NEST/tree/main.", "authors": ["Minghan Li", "Xilun Chen", "Ari Holtzman", "Beidi Chen", "Jimmy Lin", "Wen-tau Yih", "Xi Victoria Lin"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-29", "url": "https://arxiv.org/abs/2405.19325", "pdf_url": "https://arxiv.org/pdf/2405.19325v3", "arxiv_id": "2405.19325", "doi": "10.48550/arXiv.2405.19325", "citation_count": 26, "influential_citation_count": 3, "has_code": true, "code_url": "https://github.com/facebookresearch/NEST/tree/main", "venue": "Neural Information Processing Systems", "quality_score": 0.3578}
{"id": "6acae4981190d44b8a78f2cf32317774fcb65daa0c990c40314f76b15bcd050b", "sources": ["arxiv", "semantic_scholar"], "title": "Faster Cascades via Speculative Decoding", "abstract": "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 speculative decoding uses speculative execution to primarily invoke the larger model in parallel verification mode. These mechanisms offer different benefits: empirically, cascades offer better cost-quality trade-offs, often even outperforming the large model, while theoretically, speculative decoding offers a guarantee of quality-neutrality. In this paper, we leverage the best of both these approaches by designing new speculative cascading techniques that implement their deferral rule through speculative execution. We characterize the optimal deferral rule for our speculative cascades, and employ a plug-in approximation to the optimal rule. Experiments with Gemma and T5 models on a range of language benchmarks show that our approach yields better cost quality trade-offs than cascading and speculative decoding baselines.", "authors": ["Harikrishna Narasimhan", "Wittawat Jitkrittum", "Ankit Singh Rawat", "Seungyeon Kim", "Neha Gupta", "Aditya Krishna Menon", "Sanjiv Kumar"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-29", "url": "https://arxiv.org/abs/2405.19261", "pdf_url": "https://arxiv.org/pdf/2405.19261v2", "arxiv_id": "2405.19261", "doi": "10.48550/arXiv.2405.19261", "citation_count": 35, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.3891}
{"id": "53ffe10d24c9bd1a3d69f53a2fe98e676e2c5cad7420b803975f92bab5f3f9a2", "sources": ["arxiv", "semantic_scholar"], "title": "Hardware-Aware Parallel Prompt Decoding for Memory-Efficient Acceleration of LLM Inference", "abstract": "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. Crucially, they often neglect other metrics essential for real-life deployments, such as memory consumption and training cost. To overcome these limitations, we propose a novel parallel prompt decoding that requires only $0.0002$% trainable parameters, enabling efficient training on a single A100-40GB GPU in just 16 hours. Inspired by the human natural language generation process, $PPD$ approximates outputs generated at future timesteps in parallel by using multiple prompt tokens. This approach partially recovers the missing conditional dependency information necessary for multi-token generation, resulting in up to a 28% higher acceptance rate for long-range predictions. Furthermore, we present a hardware-aware dynamic sparse tree technique that adaptively optimizes this decoding scheme to fully leverage the computational capacities on different GPUs. Through extensive experiments across LLMs ranging from MobileLlama to Vicuna-13B on a wide range of benchmarks, our approach demonstrates up to 2.49$\\times$ speedup and maintains a minimal runtime memory overhead of just $0.0004$%. More importantly, our parallel prompt decoding can serve as an orthogonal optimization for synergistic integration with existing speculative decoding, showing up to $1.22\\times$ further speed improvement. Our code is available at https://github.com/hmarkc/parallel-prompt-decoding.", "authors": ["Hao Mark Chen", "Wayne Luk", "Ka Fai Cedric Yiu", "Rui Li", "Konstantin Mishchenko", "Stylianos I. Venieris", "Hongxiang Fan"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-28", "url": "https://arxiv.org/abs/2405.18628", "pdf_url": "https://arxiv.org/pdf/2405.18628v3", "arxiv_id": "2405.18628", "doi": "10.48550/arXiv.2405.18628", "citation_count": 18, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/hmarkc/parallel-prompt-decoding", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.3197}
{"id": "5bbe9626cc1cf15881a604c0999978a672132bba1c5e3e2521bf65a7433d5549", "sources": ["arxiv", "semantic_scholar"], "title": "Distributed Speculative Inference (DSI): Speculation Parallelism for Provably Faster Lossless Language Model Inference", "abstract": "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), requiring no training or architectural modifications, and it preserves the target distribution. Prior studies on SI have demonstrated empirical speedups over non-SI--but rely on sufficiently fast and accurate drafters, which are often unavailable in practice. We identify a gap where SI can be slower than non-SI if drafters are too slow or inaccurate. We close this gap by proving that DSI is faster than both SI and non-SI--given any drafters. DSI is therefore not only faster than SI, but also unlocks the acceleration of LMs for which SI fails. DSI leverages speculation parallelism (SP), a novel type of task parallelism, to orchestrate target and drafter instances that overlap in time, establishing a new foundational tradeoff between computational resources and latency. Our simulations show that DSI is 1.29-1.92x faster than SI in single-node setups for various off-the-shelf LMs and tasks. We open-source all our code.", "authors": ["Nadav Timor", "Jonathan Mamou", "Daniel Korat", "Moshe Berchansky", "Oren Pereg", "Moshe Wasserblat", "Tomer Galanti", "Michal Gordon", "David Harel"], "categories": ["cs.DC", "cs.AI", "cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-23", "url": "https://arxiv.org/abs/2405.14105", "pdf_url": "https://arxiv.org/pdf/2405.14105v5", "arxiv_id": "2405.14105", "doi": null, "citation_count": 9, "influential_citation_count": 0, "has_code": true, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.25}
{"id": "3ae87cc8276a115349987a9b658e679ebc310bb5807fb517372f2c4b2bd93424", "sources": ["arxiv", "semantic_scholar"], "title": "EMS-SD: Efficient Multi-sample Speculative Decoding for Accelerating Large Language Models", "abstract": "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. Vanilla method adds padding tokens in order to ensure that the number of new tokens remains consistent across samples. However, this increases the computational and memory access overhead, thereby reducing the speedup ratio. We propose a novel method that can resolve the issue of inconsistent tokens accepted by different samples without necessitating an increase in memory or computing overhead. Furthermore, our proposed method can handle the situation where the prediction tokens of different samples are inconsistent without the need to add padding tokens. Sufficient experiments demonstrate the efficacy of our method. Our code is available at https://github.com/niyunsheng/EMS-SD.", "authors": ["Yunsheng Ni", "Chuanjian Liu", "Yehui Tang", "Kai Han", "Yunhe Wang"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-13", "url": "https://arxiv.org/abs/2405.07542", "pdf_url": "https://arxiv.org/pdf/2405.07542v2", "arxiv_id": "2405.07542", "doi": "10.48550/arXiv.2405.07542", "citation_count": 2, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/niyunsheng/EMS-SD", "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.1193}
{"id": "9c044b20b3f753e2057a7cf0bfbcf3b4942ed264594acd96b0fb0ab7a6da68b5", "sources": ["arxiv", "semantic_scholar"], "title": "Dynamic Speculation Lookahead Accelerates Speculative Decoding of Large Language Models", "abstract": "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 (static SL) is suboptimal. We introduce DISCO (DynamIc SpeCulation lookahead Optimization), a novel method for dynamically selecting the SL. Our experiments with four datasets show that DISCO reaches an average speedup of 10% compared to the best static SL baseline, while generating the exact same text.", "authors": ["Jonathan Mamou", "Oren Pereg", "Daniel Korat", "Moshe Berchansky", "Nadav Timor", "Moshe Wasserblat", "Roy Schwartz"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-07", "url": "https://arxiv.org/abs/2405.04304", "pdf_url": "https://arxiv.org/pdf/2405.04304v5", "arxiv_id": "2405.04304", "doi": null, "citation_count": 18, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.3197}
{"id": "6e6a31fb0f13afbc376c2f0d03feb04b6190ed5fe638c0456f835644fc662c42", "sources": ["arxiv", "semantic_scholar"], "title": "Clover: Regressive Lightweight Speculative Decoding with Sequential Knowledge", "abstract": "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 batch of tokens is actually computed. Consequently, the GPU spends most of its time on memory transfer instead of computation. Recently, parallel decoding, a type of speculative decoding algorithms, is becoming more popular and has demonstrated impressive efficiency improvement in generation. It introduces extra decoding heads to large models, enabling them to predict multiple subsequent tokens simultaneously and verify these candidate continuations in a single decoding step. However, this approach deviates from the training objective of next token prediction used during pre-training, resulting in a low hit rate for candidate tokens. In this paper, we propose a new speculative decoding algorithm, Clover, which integrates sequential knowledge into the parallel decoding process. This enhancement improves the hit rate of speculators and thus boosts the overall efficiency. Clover transmits the sequential knowledge from pre-speculated tokens via the Regressive Connection, then employs an Attention Decoder to integrate these speculated tokens. Additionally, Clover incorporates an Augmenting Block that modifies the hidden states to better align with the purpose of speculative generation rather than next token prediction. The experiment results demonstrate that Clover outperforms the baseline by up to 91% on Baichuan-Small and 146% on Baichuan-Large, respectively, and exceeds the performance of the previously top-performing method, Medusa, by up to 37% on Baichuan-Small and 57% on Baichuan-Large, respectively.", "authors": ["Bin Xiao", "Chunan Shi", "Xiaonan Nie", "Fan Yang", "Xiangwei Deng", "Lei Su", "Weipeng Chen", "Bin Cui"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-05-01", "url": "https://arxiv.org/abs/2405.00263", "pdf_url": "https://arxiv.org/pdf/2405.00263v1", "arxiv_id": "2405.00263", "doi": "10.48550/arXiv.2405.00263", "citation_count": 11, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2698}
{"id": "988206e09b0b62e53fdb357e771c69307ba14973e9457e00ac4eb041a9eaf696", "sources": ["arxiv", "semantic_scholar"], "title": "Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting", "abstract": "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 from early exiting, we propose a novel self-speculative decoding framework \\emph{Kangaroo}, which uses a fixed shallow sub-network as a self-draft model, with the remaining layers serving as the larger target model. We train a lightweight and efficient adapter module on top of the sub-network to bridge the gap between the sub-network and the full model's representation ability. It is noteworthy that the inference latency of the self-draft model may no longer be negligible compared to the large model, necessitating strategies to increase the token acceptance rate while minimizing the drafting steps of the small model. To address this challenge, we introduce an additional early exiting mechanism for generating draft tokens. Specifically, we halt the small model's subsequent prediction during the drafting phase once the confidence level for the current token falls below a certain threshold. Extensive experiments on the Spec-Bench demonstrate the effectiveness of Kangaroo. Under single-sequence verification, Kangaroo achieves speedups up to $1.68\\times$ on Spec-Bench, outperforming Medusa-1 with 88.7\\% fewer additional parameters (67M compared to 591M). The code for Kangaroo is available at https://github.com/Equationliu/Kangaroo.", "authors": ["Fangcheng Liu", "Yehui Tang", "Zhenhua Liu", "Yunsheng Ni", "Kai Han", "Yunhe Wang"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-04-29", "url": "https://arxiv.org/abs/2404.18911", "pdf_url": "https://arxiv.org/pdf/2404.18911v1", "arxiv_id": "2404.18911", "doi": "10.48550/arXiv.2404.18911", "citation_count": 57, "influential_citation_count": 6, "has_code": true, "code_url": "https://github.com/Equationliu/Kangaroo", "venue": "arXiv.org", "quality_score": 0.4409}
{"id": "681e54debd120618f1188d9050fd375d4cc2aba9011575ae6de3f87b025444b7", "sources": ["arxiv", "semantic_scholar"], "title": "LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding", "abstract": "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 inference, we show that this training recipe increases the accuracy of early exit at earlier layers, without adding any auxiliary layers or modules to the model. Third, we present a novel self-speculative decoding solution where we exit at early layers and verify and correct with remaining layers of the model. Our proposed self-speculative decoding approach has less memory footprint than other speculative decoding approaches and benefits from shared compute and activations of the draft and verification stages. We run experiments on different Llama model sizes on different types of training: pretraining from scratch, continual pretraining, finetuning on specific data domain, and finetuning on specific task. We implement our inference solution and show speedups of up to 2.16x on summarization for CNN/DM documents, 1.82x on coding, and 2.0x on TOPv2 semantic parsing task. We open source our code and checkpoints at https://github.com/facebookresearch/LayerSkip.", "authors": ["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"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-04-25", "url": "https://arxiv.org/abs/2404.16710", "pdf_url": "https://arxiv.org/pdf/2404.16710v4", "arxiv_id": "2404.16710", "doi": "10.18653/v1/2024.acl-long.681", "citation_count": 269, "influential_citation_count": 19, "has_code": true, "code_url": "https://github.com/facebookresearch/LayerSkip", "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.6505}
{"id": "0611f3f579c30fc8e483c693d7710d3991274d1c259a9f7a0ad60cea8101ae8c", "sources": ["arxiv", "semantic_scholar"], "title": "Beyond the Speculative Game: A Survey of Speculative Execution in Large Language Models", "abstract": "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) per day. The bottleneck is mainly due to the autoregressive innateness of LLMs, where tokens can only be generated sequentially during decoding. To alleviate the bottleneck, the idea of speculative execution, which originates from the field of computer architecture, is introduced to LLM decoding in a \\textit{draft-then-verify} style. Under this regime, a sequence of tokens will be drafted in a fast pace by utilizing some heuristics, and then the tokens shall be verified in parallel by the LLM. As the costly sequential inference is parallelized, LLM decoding speed can be significantly boosted. Driven by the success of LLMs in recent couple of years, a growing literature in this direction has emerged. Yet, there lacks a position survey to summarize the current landscape and draw a roadmap for future development of this promising area. To meet this demand, we present the very first survey paper that reviews and unifies literature of speculative execution in LLMs (e.g., blockwise parallel decoding, speculative decoding, etc.) in a comprehensive framework and a systematic taxonomy. Based on the taxonomy, we present a critical review and comparative analysis of the current arts. Finally we highlight various key challenges and future directions to further develop the area.", "authors": ["Chen Zhang", "Zhuorui Liu", "Dawei Song"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-04-23", "url": "https://arxiv.org/abs/2404.14897", "pdf_url": "https://arxiv.org/pdf/2404.14897v1", "arxiv_id": "2404.14897", "doi": "10.48550/arXiv.2404.14897", "citation_count": 10, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.2603}
{"id": "486f06f046f7434f0b6a3518eab5a304f6d019bfc36589f3a92c7eb39bd7e757", "sources": ["arxiv", "semantic_scholar"], "title": "TriForce: Lossless Acceleration of Long Sequence Generation with Hierarchical Speculative Decoding", "abstract": "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 sequence length. Due to the auto-regressive nature of LLMs, the entire KV cache will be loaded for every generated token, resulting in low utilization of computational cores and high latency. While various compression methods for KV cache have been proposed to alleviate this issue, they suffer from degradation in generation quality. We introduce TriForce, a hierarchical speculative decoding system that is scalable for long sequence generation. This approach leverages the original model weights and dynamic sparse KV cache via retrieval as a draft model, which serves as an intermediate layer in the hierarchy and is further speculated by a smaller model to reduce its drafting latency. TriForce not only facilitates impressive speedups for Llama2-7B-128K, achieving up to 2.31$\\times$ on an A100 GPU but also showcases scalability in handling even longer contexts. For the offloading setting on two RTX 4090 GPUs, TriForce achieves 0.108s/token$\\unicode{x2014}$only half as slow as the auto-regressive baseline on an A100, which attains 7.78$\\times$ on our optimized offloading system. Additionally, TriForce performs 4.86$\\times$ than DeepSpeed-Zero-Inference on a single RTX 4090 GPU. TriForce's robustness is highlighted by its consistently outstanding performance across various temperatures. The code is available at https://github.com/Infini-AI-Lab/TriForce.", "authors": ["Hanshi Sun", "Zhuoming Chen", "Xinyu Yang", "Yuandong Tian", "Beidi Chen"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-04-18", "url": "https://arxiv.org/abs/2404.11912", "pdf_url": "https://arxiv.org/pdf/2404.11912v3", "arxiv_id": "2404.11912", "doi": "10.48550/arXiv.2404.11912", "citation_count": 105, "influential_citation_count": 5, "has_code": true, "code_url": "https://github.com/Infini-AI-Lab/TriForce", "venue": "arXiv.org", "quality_score": 0.5063}
{"id": "38f2329b1ef98400df42f77b25c01a1f73c62d4d5dedc31c9f84aae3b52a2877", "sources": ["arxiv", "semantic_scholar"], "title": "On Speculative Decoding for Multimodal Large Language Models", "abstract": "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 the LLaVA 7B model. We show that a language-only model can serve as a good draft model for speculative decoding with LLaVA 7B, bypassing the need for image tokens and their associated processing components from the draft model. Our experiments across three different tasks show that speculative decoding can achieve a memory-bound speedup of up to 2.37$\\times$ using a 115M parameter language model that we trained from scratch. Additionally, we introduce a compact LLaVA draft model incorporating an image adapter, which shows marginal performance gains in image captioning while maintaining comparable results in other tasks.", "authors": ["Mukul Gagrani", "Raghavv Goel", "Wonseok Jeon", "Junyoung Park", "Mingu Lee", "Christopher Lott"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-04-13", "url": "https://arxiv.org/abs/2404.08856", "pdf_url": "https://arxiv.org/pdf/2404.08856v1", "arxiv_id": "2404.08856", "doi": "10.48550/arXiv.2404.08856", "citation_count": 32, "influential_citation_count": 4, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3796}
{"id": "b0e014a46b5ee6e40ba56180a410ddb821c9b41f5eeb39921eb6ecb4583342b8", "sources": ["arxiv", "semantic_scholar"], "title": "DeFT: Decoding with Flash Tree-attention for Efficient Tree-structured LLM Inference", "abstract": "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 inefficient due to improper partitioning of queries and KV cache during attention calculation. This leads to two main issues: (1) a lack of memory access (IO) reuse for KV cache of shared prefixes, and (2) poor load balancing.As a result, there is redundant KV cache IO between GPU global memory and shared memory, along with low GPU utilization. To address these challenges, we propose DeFT(Decoding with Flash Tree-Attention), a hardware-efficient attention algorithm with prefix-aware and load-balanced KV cache partitions. DeFT reduces the number of read/write operations of KV cache during attention calculation through KV-Guided Grouping, a method that avoids repeatedly loading KV cache of shared prefixes in attention computation. Additionally, we propose Flattened Tree KV Splitting, a mechanism that ensures even distribution of the KV cache across partitions with little computation redundancy, enhancing GPU utilization during attention computations. By reducing 73-99% KV cache IO and nearly 100% IO for partial results during attention calculation, DeFT achieves up to 2.23/3.59x speedup in the end-to-end/attention latency across three practical tree-based workloads compared to state-of-the-art attention algorithms. Our code is available at https://github.com/LINs-lab/DeFT.", "authors": ["Jinwei Yao", "Kaiqi Chen", "Kexun Zhang", "Jiaxuan You", "Binhang Yuan", "Zeke Wang", "Tao Lin"], "categories": ["cs.CL", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-03-30", "url": "https://arxiv.org/abs/2404.00242", "pdf_url": "https://arxiv.org/pdf/2404.00242v4", "arxiv_id": "2404.00242", "doi": null, "citation_count": 14, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/LINs-lab/DeFT", "venue": "International Conference on Learning Representations", "quality_score": 0.294}
{"id": "1d4fbe25611f739b2a42a7212b17961281f0d81a44127644286252d422782029", "sources": ["arxiv", "semantic_scholar"], "title": "SDSAT: Accelerating LLM Inference through Speculative Decoding with Semantic Adaptive Tokens", "abstract": "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 involve: 1) Fine-tune the model by incorporating semantic adaptive tokens that possess flexible decoding capabilities without changing its structure, allowing them to generate high-quality draft tokens. 2) By employing a training method that does not affect the standard tokens, the model can acquire parallel decoding abilities atop its original framework with minimal training overhead. 3) We have designed the \"two-step-draft-then-verify\" generation strategies using both greedy search and nucleus sampling. Experiments conducted on the CodeLlama-13B and 7B models have yielded speed increases of over 3.5X and 3.0X, respectively. Please refer to https://github.com/hasuoshenyun/SDSAT.", "authors": ["Chengbo Liu", "Yong Zhu"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-03-27", "url": "https://arxiv.org/abs/2403.18647", "pdf_url": "https://arxiv.org/pdf/2403.18647v2", "arxiv_id": "2403.18647", "doi": "10.48550/arXiv.2403.18647", "citation_count": 2, "influential_citation_count": 1, "has_code": true, "code_url": "https://github.com/hasuoshenyun/SDSAT", "venue": "arXiv.org", "quality_score": 0.1505}
{"id": "27202f7f1e8f0f0b1f7f473545aaed629420bfc2110e52590e1798b175574b0b", "sources": ["arxiv", "semantic_scholar"], "title": "Block Verification Accelerates Speculative Decoding", "abstract": "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. In prior works, draft verification is performed independently token-by-token. Surprisingly, we show that this approach is not optimal. We propose Block Verification, a simple draft verification algorithm that verifies the entire block jointly and provides additional wall-clock speedup. We prove that the proposed mechanism is optimal in the expected number of tokens produced each iteration and specifically is never worse than the standard token-level verification. Empirically, block verification provides modest but consistent wall-clock speedups over the standard token verification algorithm of 5%-8% in a range of tasks and datasets. Given that block verification does not increase code complexity, maintains the strong lossless guarantee of the standard speculative decoding verification algorithm, cannot deteriorate performance, and, in fact, consistently improves it, it can be used as a good default in speculative decoding implementations.", "authors": ["Ziteng Sun", "Uri Mendlovic", "Yaniv Leviathan", "Asaf Aharoni", "Jae Hun Ro", "Ahmad Beirami", "Ananda Theertha Suresh"], "categories": ["cs.LG", "cs.CL", "cs.DS", "cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2024-03-15", "url": "https://arxiv.org/abs/2403.10444", "pdf_url": "https://arxiv.org/pdf/2403.10444v3", "arxiv_id": "2403.10444", "doi": null, "citation_count": 25, "influential_citation_count": 4, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.3537}
{"id": "3d23fefddad90bbce7ee7a548a2a9970ea0da484b3adffaf6c1b83167c1162b1", "sources": ["arxiv", "semantic_scholar"], "title": "Recurrent Drafter for Fast Speculative Decoding in Large Language Models", "abstract": "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 states, (2) applying a dynamic tree attention algorithm over beam search results to eliminate duplicated prefixes in candidate sequences, and (3) training through knowledge distillation from the LLM. ReDrafter accelerates Vicuna inference in MT-Bench by up to 2.8x with a PyTorch implementation on Nvidia H100 GPUs. To demonstrate its practicality in real environments, we also validated its effectiveness for on-device applications by implementing the approach in MLX and benchmarking performance on Metal GPUs in Apple Silicon chips, achieving up to 2.3x speedup.", "authors": ["Yunfei Cheng", "Aonan Zhang", "Xuanyu Zhang", "Chong Wang", "Yi Wang"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-03-14", "url": "https://arxiv.org/abs/2403.09919", "pdf_url": "https://arxiv.org/pdf/2403.09919v5", "arxiv_id": "2403.09919", "doi": "10.48550/arXiv.2403.09919", "citation_count": 33, "influential_citation_count": 6, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4225}
{"id": "d09913b1792ce60e9925a92c4df817dbde664fd5342f45bee4e811c0af2ac801", "sources": ["arxiv", "semantic_scholar"], "title": "Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs", "abstract": "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, since draft models are often unavailable in the modern open-source LLM families, e.g., for Llama 2 7B, training a high-quality draft model is required to enable inference acceleration via speculative decoding. In this paper, we propose a simple draft model training framework for direct alignment to chat-capable target models. With the proposed framework, we train Llama 2 Chat Drafter 115M, a draft model for Llama 2 Chat 7B or larger, with only 1.64\\% of the original size. Our training framework only consists of pretraining, distillation dataset generation, and finetuning with knowledge distillation, with no additional alignment procedure. For the finetuning step, we use instruction-response pairs generated by target model for distillation in plausible data distribution, and propose a new Total Variation Distance++ (TVD++) loss that incorporates variance reduction techniques inspired from the policy gradient method in reinforcement learning. Our empirical results show that Llama 2 Chat Drafter 115M with speculative decoding achieves up to 2.3 block efficiency and 2.4$\\times$ speed-up relative to autoregressive decoding on various tasks with no further task-specific fine-tuning.", "authors": ["Raghavv Goel", "Mukul Gagrani", "Wonseok Jeon", "Junyoung Park", "Mingu Lee", "Christopher Lott"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-29", "url": "https://arxiv.org/abs/2403.00858", "pdf_url": "https://arxiv.org/pdf/2403.00858v4", "arxiv_id": "2403.00858", "doi": "10.48550/arXiv.2403.00858", "citation_count": 15, "influential_citation_count": 1, "has_code": true, "code_url": null, "venue": "arXiv.org", "quality_score": 0.301}
{"id": "c4e2410b3603dd01c8663b5c3afc115ea094d3fd9080050889f87fb49129bfdf", "sources": ["arxiv", "semantic_scholar"], "title": "Minions: Accelerating Large Language Model Inference with Aggregated Speculative Execution", "abstract": "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 to speed up LLM inference, they typically require fine-tuning the LLM, incurring significant time and economic costs. Meanwhile, speculative decoding has been proposed to use small speculative models (SSMs) to accelerate the inference of LLM. However, the low acceptance rate of SSM and the high verification cost of LLM prohibit further performance improvement of inference. In this paper, we propose Minions, an LLM inference system that accelerates LLM inference with a collective and adaptive speculative generation. Specifically, Minions proposes a majority-voted mechanism to leverage multiple SSMs to jointly speculate the outputs of LLM, which improves the inference performance without introducing prohibitive computation costs for LLM. To better trade off the number of tokens speculated from SSM and the verification cost of LLM, Minions proposes an adaptive mechanism to dynamically determine the optimal speculation length of SSM, which can achieve better inference performance across different models, datasets, and hyper-parameters. In addition, Minions decouples the SSM decoding and LLM verification efficiently and adopts a pipelined execution mechanism to further improve the inference performance of LLM. By comparing with the state-of-the-art LLM inference systems, we demonstrate that Minions can achieve higher inference throughput and lower inference time.", "authors": ["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"], "categories": ["cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-24", "url": "https://arxiv.org/abs/2402.15678", "pdf_url": "https://arxiv.org/pdf/2402.15678v2", "arxiv_id": "2402.15678", "doi": null, "citation_count": 3, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.1505}
{"id": "41151d071dbc3846638ecebb019e55afca3581d45e84ea734cb43fd8179f5011", "sources": ["arxiv"], "title": "Seeing is Believing: Mitigating Hallucination in Large Vision-Language Models via CLIP-Guided Decoding", "abstract": "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 additional datasets, or incorporating complex external tools. We first perform empirical analysis on sentence-level LVLM hallucination, finding that CLIP similarity to the image acts as a stronger and more robust indicator of hallucination compared to token likelihoods. Motivated by this, we introduce our CLIP-Guided Decoding (CGD) approach, a straightforward but effective training-free approach to reduce object hallucination at decoding time. CGD uses CLIP to guide the model's decoding process by enhancing visual grounding of generated text with the image. Experiments demonstrate that CGD effectively mitigates object hallucination across multiple LVLM families while preserving the utility of text generation. Codes are available at https://github.com/d-ailin/CLIP-Guided-Decoding.", "authors": ["Ailin Deng", "Zhirui Chen", "Bryan Hooi"], "categories": ["cs.CV", "cs.AI", "cs.CL", "cs.LG", "cs.MM"], "fields_of_study": [], "published_date": "2024-02-23", "url": "https://arxiv.org/abs/2402.15300", "pdf_url": "https://arxiv.org/pdf/2402.15300v2", "arxiv_id": "2402.15300", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/d-ailin/CLIP-Guided-Decoding", "venue": null, "quality_score": 0.0}
{"id": "d1dacab66163f4ef1e9d64269e5ab32b32fd603165eca5fc5693a79fe426aa68", "sources": ["arxiv", "semantic_scholar"], "title": "Recursive Speculative Decoding: Accelerating LLM Inference via Sampling Without Replacement", "abstract": "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 a single-sequence speculative decoding. However, those works independently generate tokens at each level of the tree, not leveraging the tree's entire diversifiability. Besides, their empirical superiority has been shown for fixed length of sequences, implicitly granting more computational resource to LLM for the tree-based methods. None of the existing works has conducted empirical studies with fixed target computational budgets despite its importance to resource-bounded devices. We present Recursive Speculative Decoding (RSD), a novel tree-based method that samples draft tokens without replacement and maximizes the diversity of the tree. During RSD's drafting, the tree is built by either Gumbel-Top-$k$ trick that draws tokens without replacement in parallel or Stochastic Beam Search that samples sequences without replacement while early-truncating unlikely draft sequences and reducing the computational cost of LLM. We empirically evaluate RSD with Llama 2 and OPT models, showing that RSD outperforms the baseline methods, consistently for fixed draft sequence length and in most cases for fixed computational budgets at LLM.", "authors": ["Wonseok Jeon", "Mukul Gagrani", "Raghavv Goel", "Junyoung Park", "Mingu Lee", "Christopher Lott"], "categories": ["cs.LG", "cs.AI"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-21", "url": "https://arxiv.org/abs/2402.14160", "pdf_url": "https://arxiv.org/pdf/2402.14160v2", "arxiv_id": "2402.14160", "doi": "10.48550/arXiv.2402.14160", "citation_count": 30, "influential_citation_count": 4, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3728}
{"id": "bd4d8d65a9924c0431db4e9c75daac6f0b52302d226efbf600978c094845c269", "sources": ["arxiv", "semantic_scholar"], "title": "Ouroboros: Generating Longer Drafts Phrase by Phrase for Faster Speculative Decoding", "abstract": "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 such a drafting-verification framework, drafting efficiency has become a bottleneck in the final speedup of speculative decoding. Therefore, generating longer drafts at less cost can lead to better decoding speedup. To achieve this, we introduce Ouroboros, which can generate draft phrases to parallelize the drafting process and meanwhile lengthen drafts in a training-free manner. The experimental results on various typical text generation tasks show that Ouroboros can achieve speedups of up to $2.8\\times$ over speculative decoding and $3.9\\times$ over vanilla decoding, without fine-tuning draft and target models. The source code of Ouroboros is available at https://github.com/thunlp/Ouroboros.", "authors": ["Weilin Zhao", "Yuxiang Huang", "Xu Han", "Wang Xu", "Chaojun Xiao", "Xinrong Zhang", "Yewei Fang", "Kaihuo Zhang", "Zhiyuan Liu", "Maosong Sun"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-21", "url": "https://arxiv.org/abs/2402.13720", "pdf_url": "https://arxiv.org/pdf/2402.13720v3", "arxiv_id": "2402.13720", "doi": "10.18653/v1/2024.emnlp-main.742", "citation_count": 27, "influential_citation_count": 5, "has_code": true, "code_url": "https://github.com/thunlp/Ouroboros", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.3891}
{"id": "d47bafaa2e9fc61df6f7e08a7ef3dcf7be4f6986318aa9bab2d15f95e9df8cf9", "sources": ["arxiv", "semantic_scholar"], "title": "Distillation Contrastive Decoding: Improving LLMs Reasoning with Contrastive Decoding and Distillation", "abstract": "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-thought Prompting and advanced distillation techniques, including Dropout and Quantization. This approach effectively addresses the limitations of Contrastive Decoding (CD), which typically requires both an expert and an amateur model, thus increasing computational resource demands. By integrating contrastive prompts with distillation, DCD obviates the need for an amateur model and reduces memory usage. Our evaluations demonstrate that DCD significantly enhances LLM performance across a range of reasoning benchmarks, surpassing both CD and existing methods in the GSM8K and StrategyQA datasets.", "authors": ["Phuc Phan", "Hieu Tran", "Long Phan"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-21", "url": "https://arxiv.org/abs/2402.14874", "pdf_url": "https://arxiv.org/pdf/2402.14874v2", "arxiv_id": "2402.14874", "doi": "10.48550/arXiv.2402.14874", "citation_count": 17, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3138}
{"id": "768003fca07b400151cd526e84da9febe2d2beb31fe9044a8be540b68a5ff7c4", "sources": ["arxiv", "semantic_scholar"], "title": "Sequoia: Scalable, Robust, and Hardware-aware Speculative Decoding", "abstract": "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, and adapt to different hyperparameters and hardware. This paper introduces Sequoia, a scalable, robust, and hardware-aware algorithm for speculative decoding. To attain better scalability, Sequoia introduces a dynamic programming algorithm to find the optimal tree structure for the speculated tokens. To achieve robust speculative performance, Sequoia uses a novel sampling and verification method that outperforms prior work across different decoding temperatures. Finally, Sequoia introduces a hardware-aware tree optimizer that maximizes speculative performance by automatically selecting the token tree size and depth for a given hardware platform. Evaluation shows that Sequoia improves the decoding speed of Llama2-7B, Llama2-13B, and Vicuna-33B on an A100 by up to $4.04\\times$, $3.73\\times$, and $2.27\\times$. For offloading setting on L40, Sequoia achieves as low as 0.56 s/token for exact Llama2-70B inference latency, which is $9.96\\times$ on our optimized offloading system (5.6 s/token), $9.7\\times$ than DeepSpeed-Zero-Inference, $19.5\\times$ than Huggingface Accelerate.", "authors": ["Zhuoming Chen", "Avner May", "Ruslan Svirschevski", "Yuhsun Huang", "Max Ryabinin", "Zhihao Jia", "Beidi Chen"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-19", "url": "https://arxiv.org/abs/2402.12374", "pdf_url": "https://arxiv.org/pdf/2402.12374v3", "arxiv_id": "2402.12374", "doi": "10.48550/arXiv.2402.12374", "citation_count": 95, "influential_citation_count": 11, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5396}
{"id": "67ef487bb6d1c263e7e18ebf9b01a0015d7139fb1d21d27bee057a62d5181125", "sources": ["arxiv", "semantic_scholar"], "title": "Generation Meets Verification: Accelerating Large Language Model Inference with Smart Parallel Auto-Correct Decoding", "abstract": "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-autoregressive inference and speculative decoding capabilities, SPACE uniquely enables autoregressive LLMs to parallelize token generation and verification. This is realized through a specialized semi-autoregressive supervised fine-tuning process that equips existing LLMs with the ability to simultaneously predict multiple tokens. Additionally, an auto-correct decoding algorithm facilitates the simultaneous generation and verification of token sequences within a single model invocation. Through extensive experiments on a range of LLMs, SPACE has demonstrated inference speedup ranging from 2.7x-4.0x on HumanEval-X while maintaining output quality.", "authors": ["Hanling Yi", "Feng Lin", "Hongbin Li", "Peiyang Ning", "Xiaotian Yu", "Rong Xiao"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-19", "url": "https://arxiv.org/abs/2402.11809", "pdf_url": "https://arxiv.org/pdf/2402.11809v3", "arxiv_id": "2402.11809", "doi": "10.48550/arXiv.2402.11809", "citation_count": 26, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.3578}
{"id": "f7bfd7fd916df004684bfad76c94639ef631a600067f0f3b0a59341d5db7264c", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Streaming: Fast LLM Inference without Auxiliary Models", "abstract": "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 downstream tasks grows, these draft models add significant complexity to inference systems. We propose Speculative Streaming, a single-model speculative decoding method that fuses drafting into the target model by changing the fine-tuning objective from next token prediction to future n-gram prediction. Speculative Streaming speeds up decoding by 1.8 - 3.1X in a diverse set of tasks, such as Summarization, Structured Queries, and Meaning Representation, without sacrificing generation quality. Additionally, Speculative Streaming is parameter-efficient. It achieves on-par/higher speed-ups than Medusa-style architectures while using ~10000X fewer extra parameters, making it well-suited for resource-constrained devices.", "authors": ["Nikhil Bhendawade", "Irina Belousova", "Qichen Fu", "Henry Mason", "Mohammad Rastegari", "Mahyar Najibi"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-16", "url": "https://arxiv.org/abs/2402.11131", "pdf_url": "https://arxiv.org/pdf/2402.11131v1", "arxiv_id": "2402.11131", "doi": "10.48550/arXiv.2402.11131", "citation_count": 44, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.4133}
{"id": "dff0644adcdb708365e373b19ad2f33f82ce9be491fe7db1c504f6221ba5e466", "sources": ["arxiv", "semantic_scholar"], "title": "Any-Precision LLM: Low-Cost Deployment of Multiple, Different-Sized LLMs", "abstract": "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 with deploying multiple LLMs of varying sizes despite its practical significance. Thus, this paper introduces \\emph{any-precision LLM}, extending the concept of any-precision DNN to LLMs. Addressing challenges in any-precision LLM, we propose a lightweight method for any-precision quantization of LLMs, leveraging a post-training quantization framework, and develop a specialized software engine for its efficient serving. As a result, our solution significantly reduces the high costs of deploying multiple, different-sized LLMs by overlaying LLMs quantized to varying bit-widths, such as 3, 4, ..., $n$ bits, into a memory footprint comparable to a single $n$-bit LLM. All the supported LLMs with varying bit-widths demonstrate state-of-the-art model quality and inference throughput, proving itself to be a compelling option for deployment of multiple, different-sized LLMs. Our code is open-sourced and available online.", "authors": ["Yeonhong Park", "Jake Hyun", "SangLyul Cho", "Bonggeun Sim", "Jae W. Lee"], "categories": ["cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-16", "url": "https://arxiv.org/abs/2402.10517", "pdf_url": "https://arxiv.org/pdf/2402.10517v4", "arxiv_id": "2402.10517", "doi": "10.48550/arXiv.2402.10517", "citation_count": 54, "influential_citation_count": 11, "has_code": true, "code_url": "https://github.com/SNU-ARC/any-precision-llm", "venue": "International Conference on Machine Learning", "quality_score": 0.5396}
{"id": "546d766e1e4387ef530e3151eb52bd42a1c79f069847f3f24de216c474c5d889", "sources": ["arxiv", "semantic_scholar"], "title": "GliDe with a CaPE: A Low-Hassle Method to Accelerate Speculative Decoding", "abstract": "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 is a modified draft model architecture that reuses the cached keys and values from the target LLM, while CaPE is a proposal expansion method that uses the draft model's confidence scores to help select additional candidate tokens for verification. Extensive experiments on different benchmarks demonstrate that our proposed GliDe draft model significantly reduces the expected decoding latency. Additional evaluation using walltime reveals that GliDe can accelerate Vicuna models up to 2.17x and further extend the improvement to 2.61x with CaPE. We will release our code, data, and the trained draft models.", "authors": ["Cunxiao Du", "Jing Jiang", "Xu Yuanchen", "Jiawei Wu", "Sicheng Yu", "Yongqi Li", "Shenggui Li", "Kai Xu", "Liqiang Nie", "Zhaopeng Tu", "Yang You"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-03", "url": "https://arxiv.org/abs/2402.02082", "pdf_url": "https://arxiv.org/pdf/2402.02082v1", "arxiv_id": "2402.02082", "doi": "10.48550/arXiv.2402.02082", "citation_count": 78, "influential_citation_count": 4, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 0.4744}
{"id": "5afeb01a5e46fba807f459ae1e691d411dd2127eb024b6e49a25152ef04caefe", "sources": ["arxiv", "semantic_scholar"], "title": "Break the Sequential Dependency of LLM Inference Using Lookahead Decoding", "abstract": "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 obtain and unable to generalize. In this paper, we introduce Lookahead decoding, an exact, parallel decoding algorithm that accelerates LLM decoding without needing auxiliary models or data stores. It allows trading per-step log(FLOPs) to reduce the number of total decoding steps, is more parallelizable on single or multiple modern accelerators, and is compatible with concurrent memory-efficient attention (e.g., FlashAttention). Our implementation of Lookahead decoding can speed up autoregressive decoding by up to 1.8x on MT-bench and 4x with strong scaling on multiple GPUs in code completion tasks. Our code is avialable at https://github.com/hao-ai-lab/LookaheadDecoding", "authors": ["Yichao Fu", "Peter Bailis", "Ion Stoica", "Hao Zhang"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-03", "url": "https://arxiv.org/abs/2402.02057", "pdf_url": "https://arxiv.org/pdf/2402.02057v1", "arxiv_id": "2402.02057", "doi": "10.48550/arXiv.2402.02057", "citation_count": 315, "influential_citation_count": 35, "has_code": true, "code_url": "https://github.com/hao-ai-lab/LookaheadDecoding", "venue": "International Conference on Machine Learning", "quality_score": 0.7782}
{"id": "4d4e11f693ae1f37f62f30b2b926422543110fa1dd5f341bd0bff2b0728d2534", "sources": ["arxiv", "semantic_scholar"], "title": "Decoding Speculative Decoding", "abstract": "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 speculative decoding heavily depends on the choice of the draft model. In this work, we perform a detailed study comprising over 350 experiments with LLaMA-65B and OPT-66B using speculative decoding and delineate the factors that affect the performance gain provided by speculative decoding. Our experiments indicate that the performance of speculative decoding depends heavily on the latency of the draft model, and the draft model's capability in language modeling does not correlate strongly with its performance in speculative decoding. Based on these insights we explore a new design space for draft models and design hardware-efficient draft models for speculative decoding. Our newly designed draft model can provide 111% higher throughput than existing draft models and our approach generalizes further to all LLaMA models (1/2/3.1) and supervised fine-tuned models.", "authors": ["Minghao Yan", "Saurabh Agarwal", "Shivaram Venkataraman"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-02-02", "url": "https://arxiv.org/abs/2402.01528", "pdf_url": "https://arxiv.org/pdf/2402.01528v4", "arxiv_id": "2402.01528", "doi": "10.48550/arXiv.2402.01528", "citation_count": 39, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.4005}
{"id": "dfd9d429a28a236017b071e728afd8feee8abc64132fd4bd0b116dc6c0b17581", "sources": ["arxiv", "semantic_scholar"], "title": "Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads", "abstract": "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 speculative decoding have been suggested to address this issue, their implementation is impeded by the challenges associated with acquiring and maintaining a separate draft model. In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. By leveraging parallel processing, Medusa substantially reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. Medusa-2: Medusa is fine-tuned together with the backbone LLM, enabling better prediction accuracy of Medusa heads and higher speedup but needing a special training recipe that preserves the backbone model's capabilities. Moreover, we propose several extensions that improve or expand the utility of Medusa, including a self-distillation to handle situations where no training data is available and a typical acceptance scheme to boost the acceptance rate while maintaining generation quality. We evaluate Medusa on models of various sizes and training procedures. Our experiments demonstrate that Medusa-1 can achieve over 2.2x speedup without compromising generation quality, while Medusa-2 further improves the speedup to 2.3-3.6x.", "authors": ["Tianle Cai", "Yuhong Li", "Zhengyang Geng", "Hongwu Peng", "Jason D. Lee", "Deming Chen", "Tri Dao"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-01-19", "url": "https://arxiv.org/abs/2401.10774", "pdf_url": "https://arxiv.org/pdf/2401.10774v3", "arxiv_id": "2401.10774", "doi": "10.48550/arXiv.2401.10774", "citation_count": 753, "influential_citation_count": 147, "has_code": true, "code_url": "https://github.com/FasterDecoding/Medusa", "venue": "International Conference on Machine Learning", "quality_score": 1.0}
{"id": "3739a5bf6bc169148adb1c108e3f1338f2c82f67c08799ecc6cef741383ee896", "sources": ["arxiv", "semantic_scholar"], "title": "Unlocking Efficiency in Large Language Model Inference: A Comprehensive Survey of Speculative Decoding", "abstract": "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 autoregressive decoding, Speculative Decoding facilitates the simultaneous decoding of multiple tokens per step, thereby accelerating inference. This paper presents a comprehensive overview and analysis of this promising decoding paradigm. We begin by providing a formal definition and formulation of Speculative Decoding. Then, we organize in-depth discussions on its key facets, such as drafter selection and verification strategies. Furthermore, we present a comparative analysis of leading methods under third-party testing environments. We aim for this work to serve as a catalyst for further research on Speculative Decoding, ultimately contributing to more efficient LLM inference.", "authors": ["Heming Xia", "Zhe Yang", "Qingxiu Dong", "Peiyi Wang", "Yongqi Li", "Tao Ge", "Tianyu Liu", "Wenjie Li", "Zhifang Sui"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-01-15", "url": "https://arxiv.org/abs/2401.07851", "pdf_url": "https://arxiv.org/pdf/2401.07851v3", "arxiv_id": "2401.07851", "doi": "10.48550/arXiv.2401.07851", "citation_count": 283, "influential_citation_count": 50, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.8538}
{"id": "3b5e717fc383731e127e738c5842430f05d1b04a76cc8beb48d2e9c575923e4d", "sources": ["arxiv", "semantic_scholar"], "title": "Small LLMs Are Weak Tool Learners: A Multi-LLM Agent", "abstract": "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 accurately but also excel in task planning, tool invocation, and result summarization. While traditional works focus on training a single LLM with all these capabilities, performance limitations become apparent, particularly with smaller models. To overcome these challenges, we propose a novel approach that decomposes the aforementioned capabilities into a planner, caller, and summarizer. Each component is implemented by a single LLM that focuses on a specific capability and collaborates with others to accomplish the task. This modular framework facilitates individual updates and the potential use of smaller LLMs for building each capability. To effectively train this framework, we introduce a two-stage training paradigm. First, we fine-tune a backbone LLM on the entire dataset without discriminating sub-tasks, providing the model with a comprehensive understanding of the task. Second, the fine-tuned LLM is used to instantiate the planner, caller, and summarizer respectively, which are continually fine-tuned on respective sub-tasks. Evaluation across various tool-use benchmarks illustrates that our proposed multi-LLM framework surpasses the traditional single-LLM approach, highlighting its efficacy and advantages in tool learning.", "authors": ["Weizhou Shen", "Chenliang Li", "Hongzhan Chen", "Ming Yan", "Xiaojun Quan", "Hehong Chen", "Ji Zhang", "Fei Huang"], "categories": ["cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-01-14", "url": "https://arxiv.org/abs/2401.07324", "pdf_url": "https://arxiv.org/pdf/2401.07324v3", "arxiv_id": "2401.07324", "doi": "10.48550/arXiv.2401.07324", "citation_count": 124, "influential_citation_count": 6, "has_code": true, "code_url": "https://github.com/X-PLUG/Multi-LLM-Agent", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.5242}
{"id": "9289d4a06c883c5d803b3c4a71e69c87d4c36977648eb360689f1b92910b96af", "sources": ["arxiv", "semantic_scholar"], "title": "Multi-Candidate Speculative Decoding", "abstract": "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 the target model. However, the acceptance rate of candidate tokens receives limitations from several factors, such as the model, the dataset, and the decoding setup. This paper proposes sampling multiple candidates from a draft model and then organising them in batches for verification. We design algorithms for efficient multi-candidate verification while maintaining the distribution of the target model. Our approach shows significant improvements in acceptance rates on multiple datasets and models, consistently outperforming standard speculative decoding.", "authors": ["Sen Yang", "Shujian Huang", "Xinyu Dai", "Jiajun Chen"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2024-01-12", "url": "https://arxiv.org/abs/2401.06706", "pdf_url": "https://arxiv.org/pdf/2401.06706v1", "arxiv_id": "2401.06706", "doi": "10.48550/arXiv.2401.06706", "citation_count": 37, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": "Natural Language Processing and Chinese Computing", "quality_score": 0.3949}
{"id": "172cdfcaed73f43ec06d4e8d3885d211bc109844b8346580d8b9003583c9170a", "sources": ["arxiv", "semantic_scholar"], "title": "Cascade Speculative Drafting for Even Faster LLM Inference", "abstract": "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 runs, ultimately improving efficiency. However, the drafting process in speculative decoding includes slow autoregressive generation and allocates equal time to generating tokens, irrespective of their importance. These inefficiencies collectively contribute to the suboptimal performance of speculative decoding. To further improve LLM inference, we introduce Cascade Speculative Drafting (CS Drafting), a speculative execution algorithm that incorporates two types of cascades. The Vertical Cascade eliminates autoregressive generation from neural models, while the Horizontal Cascade optimizes time allocation in drafting for improved efficiency. Combining both cascades, CS Drafting achieves greater speedup compared to the baselines in our experiments, while preserving the same output distribution as the target model.", "authors": ["Ziyi Chen", "Xiaocong Yang", "Jiacheng Lin", "Chenkai Sun", "Kevin Chen-Chuan Chang", "Jie Huang"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2023-12-18", "url": "https://arxiv.org/abs/2312.11462", "pdf_url": "https://arxiv.org/pdf/2312.11462v5", "arxiv_id": "2312.11462", "doi": "10.48550/arXiv.2312.11462", "citation_count": 96, "influential_citation_count": 8, "has_code": false, "code_url": null, "venue": "Neural Information Processing Systems", "quality_score": 0.4967}
{"id": "418682902e811217ef1402456ecf3ad7ba498f0194e64f4f76d6592a981b8d61", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Contrastive Decoding", "abstract": "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), a straightforward yet powerful decoding approach that leverages predictions from smaller language models~(LMs) to achieve both decoding acceleration and quality improvement. Extensive evaluations and analyses on four diverse language tasks demonstrate the effectiveness of SCD, showing that decoding efficiency and quality can compatibly benefit from one smaller LM.", "authors": ["Hongyi Yuan", "Keming Lu", "Fei Huang", "Zheng Yuan", "Chang Zhou"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2023-11-15", "url": "https://arxiv.org/abs/2311.08981", "pdf_url": "https://arxiv.org/pdf/2311.08981v2", "arxiv_id": "2311.08981", "doi": "10.48550/arXiv.2311.08981", "citation_count": 8, "influential_citation_count": 1, "has_code": false, "code_url": null, "venue": "Annual Meeting of the Association for Computational Linguistics", "quality_score": 0.2386}
{"id": "2d0641d2877fb244ffa00863f126db4b56cb63a10035476a394e51ee41f0d0e3", "sources": ["arxiv", "semantic_scholar"], "title": "REST: Retrieval-Based Speculative Decoding", "abstract": "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 draft language model for speculative decoding, REST harnesses the power of retrieval to generate draft tokens. This method draws from the reservoir of existing knowledge, retrieving and employing relevant tokens based on the current context. Its plug-and-play nature allows for seamless integration and acceleration of any language models, all without necessitating additional training. When benchmarked on 7B and 13B language models in a single-batch setting, REST achieves a significant speedup of 1.62X to 2.36X on code or text generation. The code of REST is available at https://github.com/FasterDecoding/REST.", "authors": ["Zhenyu He", "Zexuan Zhong", "Tianle Cai", "Jason D. Lee", "Di He"], "categories": ["cs.CL", "cs.AI", "cs.IR", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2023-11-14", "url": "https://arxiv.org/abs/2311.08252", "pdf_url": "https://arxiv.org/pdf/2311.08252v2", "arxiv_id": "2311.08252", "doi": "10.48550/arXiv.2311.08252", "citation_count": 162, "influential_citation_count": 16, "has_code": true, "code_url": "https://github.com/FasterDecoding/REST", "venue": "North American Chapter of the Association for Computational Linguistics", "quality_score": 0.6152}
{"id": "005924600d1d561a41e58f127eb55b2a4c77af234b6386ae573eba5a025ee598", "sources": ["arxiv", "semantic_scholar"], "title": "Leveraging Speculative Sampling and KV-Cache Optimizations Together for Generative AI using OpenVINO", "abstract": "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 can be used together with model-based optimizations (e.g. quantization) to provide an optimized solution. Both sampling methods make use of KV caching. A Jupyter notebook and some sample executions are provided.", "authors": ["Haim Barad", "Ekaterina Aidova", "Yury Gorbachev"], "categories": ["cs.LG", "cs.AI", "cs.PF"], "fields_of_study": ["Computer Science"], "published_date": "2023-11-08", "url": "https://arxiv.org/abs/2311.04951", "pdf_url": "https://arxiv.org/pdf/2311.04951v2", "arxiv_id": "2311.04951", "doi": "10.48550/arXiv.2311.04951", "citation_count": 1, "influential_citation_count": 0, "has_code": true, "code_url": "https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/speculative-sampling", "venue": "arXiv.org", "quality_score": 0.0753}
{"id": "59426672d48f139b9674b376fb424828577342832d3a069035d365dd21f86da9", "sources": ["arxiv", "semantic_scholar"], "title": "The Synergy of Speculative Decoding and Batching in Serving Large Language Models", "abstract": "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 are two techniques to improve GPU hardware utilization in LLM inference. To study their synergy, we implement a prototype implementation and perform an extensive characterization analysis on various LLM models and GPU architectures. We observe that the optimal speculation length depends on the batch size used. We analyze the key observation and build a quantitative model to explain it. Based on our analysis, we propose a new adaptive speculative decoding strategy that chooses the optimal speculation length for different batch sizes. Our evaluations show that our proposed method can achieve equal or better performance than the state-of-the-art speculation decoding schemes with fixed speculation length.", "authors": ["Qidong Su", "Christina Giannoula", "Gennady Pekhimenko"], "categories": ["cs.LG", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2023-10-28", "url": "https://arxiv.org/abs/2310.18813", "pdf_url": "https://arxiv.org/pdf/2310.18813v1", "arxiv_id": "2310.18813", "doi": "10.48550/arXiv.2310.18813", "citation_count": 21, "influential_citation_count": 2, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3356}
{"id": "c7a03c3f5d42878235e16256009225ab25138f3e54853397f3fae81b3a993f5b", "sources": ["arxiv", "semantic_scholar"], "title": "SpecTr: Fast Speculative Decoding via Optimal Transport", "abstract": "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 model to sample a $\\textit{draft}$ (block or sequence of tokens), and then score all tokens in the draft by the large language model in parallel. A subset of the tokens in the draft are accepted (and the rest rejected) based on a statistical method to guarantee that the final output follows the distribution of the large model. In this work, we provide a principled understanding of speculative decoding through the lens of optimal transport (OT) with $\\textit{membership cost}$. This framework can be viewed as an extension of the well-known $\\textit{maximal-coupling}$ problem. This new formulation enables us to generalize the speculative decoding method to allow for a set of $k$ candidates at the token-level, which leads to an improved optimal membership cost. We show that the optimal draft selection algorithm (transport plan) can be computed via linear programming, whose best-known runtime is exponential in $k$. We then propose a valid draft selection algorithm whose acceptance probability is $(1-1/e)$-optimal multiplicatively. Moreover, it can be computed in time almost linear with size of domain of a single token. Using this $new draft selection$ algorithm, we develop a new autoregressive sampling algorithm called $\\textit{SpecTr}$, which provides speedup in decoding while ensuring that there is no quality degradation in the decoded output. We experimentally demonstrate that for state-of-the-art large language models, the proposed approach achieves a wall clock speedup of 2.13X, a further 1.37X speedup over speculative decoding on standard benchmarks.", "authors": ["Ziteng Sun", "Ananda Theertha Suresh", "Jae Hun Ro", "Ahmad Beirami", "Himanshu Jain", "Felix Yu"], "categories": ["cs.LG", "cs.CL", "cs.DS", "cs.IT"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2023-10-23", "url": "https://arxiv.org/abs/2310.15141", "pdf_url": "https://arxiv.org/pdf/2310.15141v2", "arxiv_id": "2310.15141", "doi": "10.48550/arXiv.2310.15141", "citation_count": 151, "influential_citation_count": 18, "has_code": false, "code_url": null, "venue": "Neural Information Processing Systems", "quality_score": 0.6394}
{"id": "4a5e48eaf98f79409c10cbbc00f3eba9bf1e1fe42857d9571ff0f2454c36c96b", "sources": ["arxiv", "semantic_scholar"], "title": "SPEED: Speculative Pipelined Execution for Efficient Decoding", "abstract": "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 with these models. This is particularly pronounced due to the autoregressive nature of generative LLM inference, where tokens are generated sequentially since each token depends on all previous output tokens. It is therefore challenging to achieve any token-level parallelism, making inference extremely memory-bound. In this work, we propose SPEED, which improves inference efficiency by speculatively executing multiple future tokens in parallel with the current token using predicted values based on early-layer hidden states. For Transformer decoders that employ parameter sharing, the memory operations for the tokens executing in parallel can be amortized, which allows us to accelerate generative LLM inference. We demonstrate the efficiency of our method in terms of latency reduction relative to model accuracy and demonstrate how speculation allows for training deeper decoders with parameter sharing with minimal runtime overhead.", "authors": ["Coleman Hooper", "Sehoon Kim", "Hiva Mohammadzadeh", "Hasan Genc", "Kurt Keutzer", "Amir Gholami", "Sophia Shao"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2023-10-18", "url": "https://arxiv.org/abs/2310.12072", "pdf_url": "https://arxiv.org/pdf/2310.12072v2", "arxiv_id": "2310.12072", "doi": "10.48550/arXiv.2310.12072", "citation_count": 52, "influential_citation_count": 6, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.4311}
{"id": "0a0048d6d75dc0f0af4f2cac013af9142883a3c23daf8f01dd01ea43ca72e28f", "sources": ["arxiv", "semantic_scholar"], "title": "DistillSpec: Improving Speculative Decoding via Knowledge Distillation", "abstract": "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 that is well-aligned with the target model is challenging. To tackle this issue, we propose DistillSpec that uses knowledge distillation to better align the draft model with the target model, before applying SD. DistillSpec makes two key design choices, which we demonstrate via systematic study to be crucial to improving the draft and target alignment: utilizing on-policy data generation from the draft model, and tailoring the divergence function to the task and decoding strategy. Notably, DistillSpec yields impressive 10 - 45% speedups over standard SD on a range of standard benchmarks, using both greedy and non-greedy sampling. Furthermore, we combine DistillSpec with lossy SD to achieve fine-grained control over the latency vs. task performance trade-off. Finally, in practical scenarios with models of varying sizes, first using distillation to boost the performance of the target model and then applying DistillSpec to train a well-aligned draft model can reduce decoding latency by 6-10x with minimal performance drop, compared to standard decoding without distillation.", "authors": ["Yongchao Zhou", "Kaifeng Lyu", "Ankit Singh Rawat", "Aditya Krishna Menon", "Afshin Rostamizadeh", "Sanjiv Kumar", "Jean-François Kagy", "Rishabh Agarwal"], "categories": ["cs.CL", "cs.AI", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2023-10-12", "url": "https://arxiv.org/abs/2310.08461", "pdf_url": "https://arxiv.org/pdf/2310.08461v2", "arxiv_id": "2310.08461", "doi": "10.48550/arXiv.2310.08461", "citation_count": 159, "influential_citation_count": 12, "has_code": false, "code_url": null, "venue": "International Conference on Learning Representations", "quality_score": 0.557}
{"id": "8f8b1b6229e18341df51a3b2a856ff108a87eba39bf21ebd3324ce8f0c01549a", "sources": ["arxiv", "semantic_scholar"], "title": "Online Speculative Decoding", "abstract": "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 and a significant capability gap between the draft and target models. We introduce online speculative decoding to address this challenge. The main idea is to continuously update the (multiple) draft model(s) on observed user query data. Adapting to query distribution mitigates the shifts between the training distribution of the draft model and the query distribution, enabling the draft model to more accurately predict the target model's outputs. We develop a prototype of online speculative decoding based on knowledge distillation and evaluate it using both synthetic and real query data. The results show a substantial increase in the token acceptance rate by 0.1 to 0.65, bringing 1.42x to 2.17x latency reduction. Our code is available at https://github.com/LiuXiaoxuanPKU/OSD.", "authors": ["Xiaoxuan Liu", "Lanxiang Hu", "Peter Bailis", "Alvin Cheung", "Zhijie Deng", "Ion Stoica", "Hao Zhang"], "categories": ["cs.AI", "cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2023-10-11", "url": "https://arxiv.org/abs/2310.07177", "pdf_url": "https://arxiv.org/pdf/2310.07177v4", "arxiv_id": "2310.07177", "doi": "10.48550/arXiv.2310.07177", "citation_count": 112, "influential_citation_count": 9, "has_code": true, "code_url": "https://github.com/LiuXiaoxuanPKU/OSD", "venue": "International Conference on Machine Learning", "quality_score": 0.5133}
{"id": "80046499ae7fe34b7d3ceeeb93e6c84cb79fd7a37e04b7e1c0346ec84371dfe2", "sources": ["arxiv"], "title": "Draft & Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding", "abstract": "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, which is achieved by selectively skipping certain intermediate layers during drafting. Subsequently, the verification stage employs the original LLM to validate those draft output tokens in one forward pass. This process ensures the final output remains identical to that produced by the unaltered LLM. Moreover, the proposed method requires no additional neural network training and no extra memory footprint, making it a plug-and-play and cost-effective solution for inference acceleration. Benchmarks with LLaMA-2 and its variants demonstrated a speedup up to 1.99$\\times$.", "authors": ["Jun Zhang", "Jue Wang", "Huan Li", "Lidan Shou", "Ke Chen", "Gang Chen", "Sharad Mehrotra"], "categories": ["cs.CL"], "fields_of_study": [], "published_date": "2023-09-15", "url": "https://arxiv.org/abs/2309.08168", "pdf_url": "https://arxiv.org/pdf/2309.08168v2", "arxiv_id": "2309.08168", "doi": null, "citation_count": 0, "influential_citation_count": 0, "has_code": false, "code_url": null, "venue": null, "quality_score": 0.0}
{"id": "49216a6b94fcca5887cbfa932cb17e60ca294b4a213b49b3e9b1a6269de650e5", "sources": ["arxiv", "semantic_scholar"], "title": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "abstract": "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 generates one token at a time per request. The varying prefill and decode times also lead to imbalance across micro-batches when using pipeline parallelism, resulting in further inefficiency due to bubbles. We present SARATHI to address these challenges. SARATHI employs chunked-prefills, which splits a prefill request into equal sized chunks, and decode-maximal batching, which constructs a batch using a single prefill chunk and populates the remaining slots with decodes. During inference, the prefill chunk saturates GPU compute, while the decode requests 'piggyback' and cost up to an order of magnitude less compared to a decode-only batch. Chunked-prefills allows constructing multiple decode-maximal batches from a single prefill request, maximizing coverage of decodes that can piggyback. Furthermore, the uniform compute design of these batches ameliorates the imbalance between micro-batches, significantly reducing pipeline bubbles. Our techniques yield significant improvements in inference performance across models and hardware. For the LLaMA-13B model on A6000 GPU, SARATHI improves decode throughput by up to 10x, and accelerates end-to-end throughput by up to 1.33x. For LLaMa-33B on A100 GPU, we achieve 1.25x higher end-to-end-throughput and up to 4.25x higher decode throughput. When used with pipeline parallelism on GPT-3, SARATHI reduces bubbles by 6.29x, resulting in an end-to-end throughput improvement of 1.91x.", "authors": ["Amey Agrawal", "Ashish Panwar", "Jayashree Mohan", "Nipun Kwatra", "Bhargav S. Gulavani", "Ramachandran Ramjee"], "categories": ["cs.LG", "cs.DC"], "fields_of_study": ["Computer Science"], "published_date": "2023-08-31", "url": "https://arxiv.org/abs/2308.16369", "pdf_url": "https://arxiv.org/pdf/2308.16369v1", "arxiv_id": "2308.16369", "doi": "10.48550/arXiv.2308.16369", "citation_count": 231, "influential_citation_count": 19, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.6505}
{"id": "8114250c7ba56938a075f69d04d50662d43f8a0a16cc9c40f32b7e0785db9223", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating LLM Inference with Staged Speculative Decoding", "abstract": "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 speculative decoding. First, we restructure the speculative batch as a tree, which reduces generation costs and increases the expected tokens per batch. Second, we add a second stage of speculative decoding. Taken together, we reduce single-batch decoding latency by 3.16x with a 762M parameter GPT-2-L model while perfectly preserving output quality.", "authors": ["Benjamin Spector", "Chris Re"], "categories": ["cs.AI", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2023-08-08", "url": "https://arxiv.org/abs/2308.04623", "pdf_url": "https://arxiv.org/pdf/2308.04623v1", "arxiv_id": "2308.04623", "doi": "10.48550/arXiv.2308.04623", "citation_count": 178, "influential_citation_count": 9, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.5632}
{"id": "6e04394cce44d057e3bcbd2217a25675868387cf0c1bc36a6cd6b8435d406631", "sources": ["arxiv", "semantic_scholar"], "title": "RETA-LLM: A Retrieval-Augmented Large Language Model Toolkit", "abstract": "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 LLMs). Applying this strategy, LLMs can generate more factual texts in response to user input according to the relevant content retrieved by IR systems from external corpora as references. In addition, by incorporating external knowledge, retrieval-augmented LLMs can answer in-domain questions that cannot be answered by solely relying on the world knowledge stored in parameters. To support research in this area and facilitate the development of retrieval-augmented LLM systems, we develop RETA-LLM, a {RET}reival-{A}ugmented LLM toolkit. In RETA-LLM, we create a complete pipeline to help researchers and users build their customized in-domain LLM-based systems. Compared with previous retrieval-augmented LLM systems, RETA-LLM provides more plug-and-play modules to support better interaction between IR systems and LLMs, including {request rewriting, document retrieval, passage extraction, answer generation, and fact checking} modules. Our toolkit is publicly available at https://github.com/RUC-GSAI/YuLan-IR/tree/main/RETA-LLM.", "authors": ["Jiongnan Liu", "Jiajie Jin", "Zihan Wang", "Jiehan Cheng", "Zhicheng Dou", "Ji-Rong Wen"], "categories": ["cs.IR"], "fields_of_study": ["Computer Science"], "published_date": "2023-06-08", "url": "https://arxiv.org/abs/2306.05212", "pdf_url": "https://arxiv.org/pdf/2306.05212v1", "arxiv_id": "2306.05212", "doi": "10.48550/arXiv.2306.05212", "citation_count": 61, "influential_citation_count": 2, "has_code": true, "code_url": "https://github.com/RUC-GSAI/YuLan-IR/tree/main/RETA-LLM", "venue": "arXiv.org", "quality_score": 0.4481}
{"id": "ae9d74b24d12158c81e74194281d8cce682e379fc765f2defb62ff138cab96a0", "sources": ["arxiv", "semantic_scholar"], "title": "AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration", "abstract": "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 deployment challenges. We propose Activation-aware Weight Quantization (AWQ), a hardware-friendly approach for LLM low-bit weight-only quantization. AWQ finds that not all weights in an LLM are equally important. Protecting only 1% salient weights can greatly reduce quantization error. To identify salient weight channels, we should refer to the activation distribution, not weights. To avoid the hardware-inefficient mix-precision quantization, we mathematically derive that scaling up the salient channels can reduce the quantization error. AWQ employs an equivalent transformation to scale the salient weight channels to protect them. The scale is determined by collecting the activation statistics offline. AWQ does not rely on any backpropagation or reconstruction, so it generalizes to different domains and modalities without overfitting the calibration set. AWQ outperforms existing work on various language modeling and domain-specific benchmarks (coding and math). Thanks to better generalization, it achieves excellent quantization performance for instruction-tuned LMs and, for the first time, multi-modal LMs. Alongside AWQ, we implement TinyChat, an efficient and flexible inference framework tailored for 4-bit on-device LLM/VLMs. With kernel fusion and platform-aware weight packing, TinyChat offers more than 3x speedup over the Huggingface FP16 implementation on both desktop and mobile GPUs. It also democratizes the deployment of the 70B Llama-2 model on mobile GPUs.", "authors": ["Ji Lin", "Jiaming Tang", "Haotian Tang", "Shang Yang", "Wei-Ming Chen", "Wei-Chen Wang", "Guangxuan Xiao", "Xingyu Dang", "Chuang Gan", "Song Han"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2023-06-01", "url": "https://arxiv.org/abs/2306.00978", "pdf_url": "https://arxiv.org/pdf/2306.00978v6", "arxiv_id": "2306.00978", "doi": "10.1145/3714983.3714987", "citation_count": 1454, "influential_citation_count": 217, "has_code": true, "code_url": "https://github.com/mit-han-lab/llm-awq", "venue": "Conference on Machine Learning and Systems", "quality_score": 1.0}
{"id": "3e23452c9fc523b106c8710f61fc345800751af58ed7ce7f0fa2af1b0a718721", "sources": ["arxiv", "semantic_scholar"], "title": "SpecInfer: Accelerating Generative Large Language Model Serving with Tree-based Speculative Inference and Verification", "abstract": "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 each represent a candidate token sequence. The correctness of all candidate token sequences represented by a token tree is verified against the LLM in parallel using a novel tree-based parallel decoding mechanism. SpecInfer uses an LLM as a token tree verifier instead of an incremental decoder, which significantly reduces the end-to-end latency and computational requirement for serving generative LLMs while provably preserving model quality. Our evaluation shows that SpecInfer outperforms existing LLM serving systems by 1.5-2.8x for distributed LLM inference and by 2.6-3.5x for offloading-based LLM inference, while preserving the same generative performance. SpecInfer is publicly available at https://github.com/flexflow/FlexFlow/", "authors": ["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"], "categories": ["cs.CL", "cs.DC", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2023-05-16", "url": "https://arxiv.org/abs/2305.09781", "pdf_url": "https://arxiv.org/pdf/2305.09781v4", "arxiv_id": "2305.09781", "doi": "10.1145/3620666.3651335", "citation_count": 368, "influential_citation_count": 55, "has_code": true, "code_url": "https://github.com/flexflow/FlexFlow/", "venue": "International Conference on Architectural Support for Programming Languages and Operating Systems", "quality_score": 0.8741}
{"id": "fb6008aa313ec9941aa1e1a1e593086f2dccf2fcedde4e9dbee3ce4576e8af0b", "sources": ["arxiv", "semantic_scholar"], "title": "Accelerating Large Language Model Decoding with Speculative Sampling", "abstract": "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 comparable to that of sampling a single token from the larger target model. This is combined with a novel modified rejection sampling scheme which preserves the distribution of the target model within hardware numerics. We benchmark speculative sampling with Chinchilla, a 70 billion parameter language model, achieving a 2-2.5x decoding speedup in a distributed setup, without compromising the sample quality or making modifications to the model itself.", "authors": ["Charlie Chen", "Sebastian Borgeaud", "Geoffrey Irving", "Jean-Baptiste Lespiau", "Laurent Sifre", "John Jumper"], "categories": ["cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2023-02-02", "url": "https://arxiv.org/abs/2302.01318", "pdf_url": "https://arxiv.org/pdf/2302.01318v1", "arxiv_id": "2302.01318", "doi": "10.48550/arXiv.2302.01318", "citation_count": 908, "influential_citation_count": 108, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 1.0}
{"id": "1adec7fc5b299c3e07eba043ad02b9090d6e8249d04ad85781ab58da756bcca6", "sources": ["arxiv", "semantic_scholar"], "title": "Fast Inference from Transformers via Speculative Decoding", "abstract": "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 of our approach lie the observations that (1) hard language-modeling tasks often include easier subtasks that can be approximated well by more efficient models, and (2) using speculative execution and a novel sampling method, we can make exact decoding from the large models faster, by running them in parallel on the outputs of the approximation models, potentially generating several tokens concurrently, and without changing the distribution. Our method can accelerate existing off-the-shelf models without retraining or architecture changes. We demonstrate it on T5-XXL and show a 2X-3X acceleration compared to the standard T5X implementation, with identical outputs.", "authors": ["Yaniv Leviathan", "Matan Kalman", "Yossi Matias"], "categories": ["cs.LG", "cs.CL"], "fields_of_study": ["Computer Science"], "published_date": "2022-11-30", "url": "https://arxiv.org/abs/2211.17192", "pdf_url": "https://arxiv.org/pdf/2211.17192v2", "arxiv_id": "2211.17192", "doi": "10.48550/arXiv.2211.17192", "citation_count": 1646, "influential_citation_count": 279, "has_code": false, "code_url": null, "venue": "International Conference on Machine Learning", "quality_score": 1.0}
{"id": "da28afc96931be423a3c8200370b6484947e29f2a8f0b3fde8a1a79ee9ff0ace", "sources": ["arxiv", "semantic_scholar"], "title": "Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation", "abstract": "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 Spec-Verification -- a reliable method for verifying the drafted tokens efficiently in the decoding paradigm. Experimental results on various seq2seq tasks including machine translation and abstractive summarization show our approach can achieve around $5\\times$ speedup for the popular Transformer architectures with comparable generation quality to beam search decoding, refreshing the impression that the draft-then-verify paradigm introduces only $1.4\\times$$\\sim$$2\\times$ speedup. In addition to the remarkable speedup, we also demonstrate 3 additional advantages of SpecDec, revealing its practical value for accelerating generative models in real-world applications. Our models and codes are available at https://github.com/hemingkx/SpecDec.", "authors": ["Heming Xia", "Tao Ge", "Peiyi Wang", "Si-Qing Chen", "Furu Wei", "Zhifang Sui"], "categories": ["cs.CL", "cs.LG"], "fields_of_study": ["Computer Science"], "published_date": "2022-03-30", "url": "https://arxiv.org/abs/2203.16487", "pdf_url": "https://arxiv.org/pdf/2203.16487v6", "arxiv_id": "2203.16487", "doi": "10.18653/v1/2023.findings-emnlp.257", "citation_count": 183, "influential_citation_count": 12, "has_code": true, "code_url": "https://github.com/hemingkx/SpecDec", "venue": "Conference on Empirical Methods in Natural Language Processing", "quality_score": 0.5662}
{"id": "e965fbcd433831dc0def0d0120e14b36cb2dd4d0b53cff938308d76ba597b58a", "sources": ["arxiv", "semantic_scholar"], "title": "Causal Inference Using Tractable Circuits", "abstract": "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 interpreted in a causality context as follows. One can compile a non-parametric causal graph into an arithmetic circuit that supports inference in time linear in the circuit size. The circuit is also non-parametric so it can be used to estimate parameters from data and to further reason (in linear time) about the causal graph parametrized by these estimates. Moreover, the circuit size can sometimes be bounded even when the treewidth of the causal graph is not, leading to tractable inference on models that have been deemed intractable previously. This has been enabled by a new technique that can exploit causal mechanisms computationally but without needing to know their identities (the classical setup in causal inference). Our goal is to provide a causality-oriented exposure to these new results and to speculate on how they may potentially contribute to more scalable and versatile causal inference.", "authors": ["Adnan Darwiche"], "categories": ["cs.AI", "cs.CC", "cs.LG", "cs.LO", "stat.ME"], "fields_of_study": ["Computer Science", "Mathematics"], "published_date": "2022-02-07", "url": "https://arxiv.org/abs/2202.02891", "pdf_url": "https://arxiv.org/pdf/2202.02891v1", "arxiv_id": "2202.02891", "doi": null, "citation_count": 23, "influential_citation_count": 3, "has_code": false, "code_url": null, "venue": "arXiv.org", "quality_score": 0.3451}