Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

6 Commits
Β 
Β 

Repository files navigation

Awesome-Parallel-Decoding

Awesome Stars License: CC BY-NC-ND 4.0 PRWelcome Last Commit

πŸ“’ A curated list of Awesome Parallel Decoding papers to accelerate the inference of (autoregressive) LLMs.

Autoregressive (AR) decoding generates one token per forward pass, leaving the strict token-by-token dependency as the dominant latency bottleneck of modern LLM serving. Parallel decoding refers to the broad family of techniques that shorten this critical path by introducing parallel candidates, parallel verification, parallel reasoning steps, or parallel task orchestration, while ideally preserving the quality (and, for lossless methods, the exact output distribution) of the original model. This repository organizes the literature along the granularity of parallelism β€” from (i) individual tokens, to (ii) reasoning sequences, to (iii) whole tasks β€” and additionally tracks the (iv) model architecture-level and (v) system-level improvements. We also introduce (vi) hybrid advances and (vii) related benchmarks that make parallel decoding practical.

Legend. Venue is given as the publication venue (e.g., ICML'24, NeurIPS'25); preprints use arXiv'YY. The bracketed token at the start of a title is the method's alias/acronym (e.g., [EAGLE]). Author and affiliation fields are filled where reliably verifiable and marked N/A otherwise. A blank Code cell means no official public code repository is currently known.

πŸ“– Contents

Within each section, papers run from older to newer; click a paper to jump to its full row.

LayerSubclassPapers
1. Token-Level Parallel Decoding 1.1 Drafting with the Draft Model Blockwise (NIPS'18) Β· SpecDec (Findings-EMNLP'23) Β· Leviathan (ICML'23) Β· Chen (arXiv'23) Β· BiLD (NeurIPS'23) Β· DistillSpec (ICLR'24) Β· Online SD (ICML'24) Β· Decoding SD (NAACL'25) Β· Multilingual Drafters (EMNLP'24) Β· HASS (ICLR'25) Β· Temp-Centric KD (Findings-EMNLP'24) Β· ParallelSpec (arXiv'24) Β· CTC-drafter (NeurIPS'24) Β· Heterogeneous Vocabularies (ICML'25) Β· CoS (ICML'25) Β· GRIFFIN (NeurIPS'25) Β· CORAL (ACL'25) Β· FR-Spec (ACL'25) Β· Domain Drafters (ICLRW'25) Β· Auto-Task SD (arXiv'25) Β· Mamba Drafters (Findings-EMNLP'25) Β· OmniDraft (NeurIPS'25) Β· SPECTRA (ACL'25) Β· RepSpec (ICLR'26) Β· AdaSPEC (NeurIPS'25) Β· Flatter Tokens (ICLR'26) Β· OnlineSPEC (ICML'26)
1.2 Drafting within the Target Model Draft \& Verify (ACL'24) Β· PaSS (NeurIPSW'23) Β· Medusa (ICML'24) Β· EAGLE (ICML'24) Β· GliDe with a CaPE (ICML'24) Β· Hydra (COLM'24) Β· ReDrafter (arXiv'24) Β· LayerSkip (ACL'24) Β· Kangaroo (NeurIPS'24) Β· MTP (ICML'24) Β· PPD (Findings-EMNLP'25) Β· S3D (arXiv'24) Β· EESD (Findings-ACL'24) Β· EAGLE-2 (EMNLP'24) Β· Amphista (NAACL'25) Β· Clover-2 (arXiv'24) Β· KOALA (CSCWD'25) Β· Draft-on-the-Fly (Findings-EMNLP'24) Β· SWIFT (ICLR'25) Β· Mixture of Attentions (ICLR'25) Β· DeepSeek-V3 (arXiv'24) Β· On MTP (ICLRW'25) Β· EAGLE-3 (NeurIPS'25) Β· Gumiho (ICML'25) Β· DEL (COLM'25) Β· CLaSp (ACL'25) Β· L-MTP (NeurIPS'25) Β· Beagle (arXiv'25) Β· AdaDecode (ICML'25) Β· Your LLM Knows the Future (arXiv'25) Β· CAS-Spec (NeurIPS'25) Β· FractalLLM (Findings-EMNLP'25) Β· Speculative Streaming (EMNLP'25) Β· SparseSpec (MLSys'26) Β· mtp-lm (ICMLW'26) Β· ESP (ICML'26)
1.3 Retrieval-Based and Non-Parametric Drafting Santilli (ACL'23) Β· REST (NAACL'24) Β· Lookahead Decoding (ICML'24) Β· Ouroboros (EMNLP'24) Β· CLLMs (ICML'24) Β· NEST (NeurIPS'24) Β· MSN (EMNLP'24) Β· Token Recycling (ACL'25) Β· SuffixDecoding (NeurIPS'25) Β· SAM Decoding (ACL'25) Β· DReSD (Findings-ACL'25) Β· RAPID (ICML'25) Β· Hierarchy Drafting (Findings-NAACL'25) Β· RASD (Findings-ACL'25) Β· N-Gram Trie (EMNLP'25) Β· Cacheback (EMNLP'25) Β· Jacobi Forcing (ICML'26)
1.4 Draft Structure and Sampling Strategies SpecInfer (ASPLOS'24) Β· SpecTr (NeurIPS'23) Β· MCSD (NLPCC'25) Β· Sequoia (NeurIPS'24) Β· Tree Monte Carlo (ICML'24) Β· Block Verification (ICLR'25) Β· SLiM (Findings-NAACL'24) Β· Dynamic Spec Lookahead (NeurIPSW'24) Β· OPT-Tree (TACL'25) Β· Graph-Structured SD (Findings-ACL'24) Β· DDD (arXiv'24) Β· Multi-Draft Canonical (ICLR'25) Β· Theory of SD (NeurIPS'24) Β· SpecHub (EMNLP'24) Β· SVIP (EMNLP'25) Β· AdaEAGLE (arXiv'24) Β· Judge Decoding (ICLR'25) Β· TETRIS (ACL'25) Β· Optimal Multi-Draft (ICLR'25) Β· Fuzzy SD (Findings-ACL'25) Β· HSDDW (Findings-NAACL'25) Β· Exponential Races (Findings-ACL'25) Β· AASD (EMNLP'25) Β· Traversal Verification (NeurIPS'25) Β· STree (NeurIPS'25) Β· BanditSpec (ICML'25) Β· HeteroSpec (ACL'26) Β· List-Level Coupling (NeurIPS'25) Β· SpecBranch (ICLR'26) Β· Drop-In Adaptation (ACL'25) Β· Pruned Candidate Tree (ACL'25) Β· GTO (ICLR'26) Β· polybasic SD (ICML'25) Β· Not-a-Bandit (ICLR'26) Β· Global Resolution (ICLR'26) Β· Loosely SD (ICLR'26) Β· Lossless Hierarchical SD (ICLR'26) Β· LTD (ICLR'26) Β· SSD (ICLR'26) Β· Cactus (ICLR'26) Β· SpecBlock (arXiv'26) Β· JetSpec (arXiv'26)
1.5 Cross-Technique Integration Speculative Contrastive Decoding (ACL'24) Β· Watermark SD (NeurIPS'24) Β· Constrained Spec Lookaheads (NAACL'25) Β· DREAM (NeurIPS'25) Β· MASSV (Findings-EMNLP'25) Β· Spec-VLA (EMNLP'25) Β· SpecVLM (EMNLP'25) Β· Reward-Shifted SS (EMNLP'25) Β· Safety-Aware SD (EMNLP'25) Β· ViSpec (NeurIPS'25) Β· SPEC-RL (arXiv'25) Β· FastGRPO (ICLR'26) Β· ReSpec (MLSys'26) Β· DAS (MLSys'26) Β· SpecActor (arXiv'25) Β· TLT (ASPLOS'26) Β· RLHFSpec (arXiv'25) Β· SRT (NeurIPSW'25) Β· Watermark SD-2 (ICLR'26)
2. Sequence-Level Parallel Decoding 2.1 Multi-Sequence Parallel Generation FastCoT (arXiv'23) Β· SEED (COLING'25) Β· Speculative RAG (ICLR'25) Β· RSD (ICML'25) Β· Multi-Sample SD (Findings-EMNLP'25) Β· SpecReason (NeurIPS'25) Β· Speculative Thinking (COLM'25) Β· SCoT (Findings-ACL'26) Β· SpecSearch (ICML'25) Β· R2R (NeurIPS'25) Β· Lookahead Reasoning (NeurIPS'25) Β· STAND (EMNLP'25) Β· SpecExit (ICML'26) Β· SpecCoT (Findings-EMNLP'25) Β· SemanticSpec (arXiv'26) Β· SpecGuard (Findings-ACL'26)
2.2 Intra-Sequence Parallel Generation Skeleton-of-Thought (ICLR'24) Β· Graph of Thoughts (AAAI'24) Β· Plato (COLM'25) Β· PASTA (ICML'25) Β· Parallel Decode in One Seq (EMNLP'25) Β· APR (COLM'25) Β· Multiverse (NeurIPS'25) Β· Sprint (NeurIPS'25) Β· PCCoT (EMNLP'25) Β· Parallel Think, Sequential Answer (arXiv'25) Β· Parallel Loop Transformer (arXiv'25)
2.3 Path Exploration and Test-Time Scaling Agent Forest (TMLR'24) Β· Speculative Rejection (NeurIPS'24) Β· DPTS (ACL'25) Β· Group Think (arXiv'25) Β· Adaptive Termination (arXiv'25) Β· A2R (arXiv'25) Β· Parallel-R1 (ICLR'26) Β· ATTS (ICLR'26) Β· Parallel TTS for Latent (ACL'26) Β· DeepPrune (Findings-ACL'26) Β· DTS (ICML'26) Β· Parallel-Probe (ICML'26)
3. Task-Level Parallel Decoding 3.1 Multi-Agent Parallel Collaboration AutoGen (COLM'24) Β· More Agents (TMLR'24) Β· Mixture-of-Agents (ICLR'25) Β· M1-Parallel (ICMLW'25)
3.2 Parallel and Speculative Agent Execution LLMCompiler (ICML'24) Β· LLM-Tool Compiler (arXiv'24) Β· Async Tool Usage (arXiv'24) Β· AsyncLM (arXiv'24) Β· DSP (ICLR'26) Β· Speculative Actions (ICLR'26) Β· Spec Tool Calls (arXiv'25) Β· W\&D (ICLRW'26) Β· RealtimeTool (ICML'26) Β· PASTE (arXiv'26)
4. Architecture-Level Parallel Decoding 4.1 Model Architecture and Training NAT (ICLR'18) Β· Iterative Refinement (EMNLP'18) Β· Latent Transformer (ICML'18) Β· Mask-Predict (EMNLP'19) Β· D3PM (NeurIPS'21) Β· Diffusion-LM (NeurIPS'22) Β· SEDD (ICML'24) Β· RADD (ICLR'25) Β· MD4 (NeurIPS'24) Β· MDLM (NeurIPS'24) Β· DiffuLLaMA (ICLR'25) Β· SDTT (ICLR'25) Β· LLaDA (NeurIPS'25) Β· Block Diffusion (ICLR'25) Β· Duo (ICML'25) Β· Dream 7B (arXiv'25) Β· SDLM (arXiv'25) Β· Fast-dLLM v2 (ICLR'26) Β· RND1 (misc'25) Β· E2D2 (NeurIPS'25) Β· SDAR (Findings-ACL'26) Β· NBDiff (arXiv'25) Β· Efficient-DLM (ICML'26) Β· LLaDA2.0 (arXiv'25) Β· Diffusion In Diffusion (arXiv'26) Β· MBD-LMs (arXiv'26)
4.2 Decoding and Sampling Algorithms EB-Sampler (NeurIPS'25) Β· Fast-dLLM (ICLR'26) Β· SlowFast Sampling (ICLR'26) Β· WINO (ICLR'26) Β· Prophet (ICLR'26) Β· RWS (EMNLP'25) Β· LSD (NeurIPS'25) Β· ADJUST (arXiv'25) Β· Learn2PD (ICLR'26) Β· dParallel (ICLR'26) Β· FreeDave (NeurIPSW'25) Β· LocalLeap (arXiv'25) Β· Saber (ACL'26) Β· LoPA (arXiv'25) Β· SchED (Findings-ACL'26) Β· CadLLM (Findings-ACL'26) Β· Learning Unmasking Policies (ICML'26) Β· Order-Token Search (arXiv'26) Β· d3LLM (ICML'26) Β· RDD (arXiv'26) Β· ReMix (CVPR'26) Β· Info-Gain Sampler (ICML'26) Β· DOS (Findings-ACL'26) Β· Confidence-Based Decoding (arXiv'26)
5. System and Infrastructure 5.1 Serving and Scheduling TriForce (COLM'24) Β· BASS (Findings-ACL'24) Β· DSI (ICLR'25) Β· EMS-SD (NAACL'25) Β· SpecExec (NeurIPS'24) Β· Optimized Speculative Sampling (EMNLP'24) Β· MagicDec (ICLR'25) Β· PEARL (ICLR'25) Β· Dovetail (EMNLP'25) Β· Attention-Level Speculation (ICML'25) Β· EasySpec (NeurIPS'25) Β· AdaSpec (SoCC'25) Β· SubSpec (NeurIPS'25) Β· Mirror-SD (arXiv'25) Β· dInfer (arXiv'25) Β· dLLM-Serve (arXiv'25)
5.2 Algorithm-Infrastructure Co-Design DeFT (ICLR'25) Β· QSpec (EMNLP'25) Β· QuantSpec (ICML'25) Β· SpecPrefill (ICML'25) Β· ML-SpecQD (arXiv'25) Β· SpeCache (ICML'25) Β· MoESD (NeurIPS'25) Β· Cascade (arXiv'25) Β· AsyncSpade (ICML'26) Β· Yggdrasil (NeurIPS'25)
6. Hybrid Methods 6.1 Non-AR Drafter for AR Targets SpecDiff (NAACL'25) Β· DiffuSpec (Findings-ACL'26) Β· SpecDiff-2 (MLSys'26) Β· DEER (arXiv'25) Β· DART (arXiv'26) Β· DFlash (ICML'26) Β· JetSpec (arXiv'26) Β· DSpark (arXiv'26) Β· AdaFlash (arXiv'26)
6.2 Speculation for Other Generative Paradigms CS Drafting (NeurIPS'24) Β· Faster Cascades (ICLR'25) Β· SJD (ICLR'25) Β· LANTERN (ICLR'25) Β· Accelerated Diffusion via Speculative Sampling (ICML'25) Β· ASD (ICML'25) Β· APD (NeurIPS'25) Β· Spiffy (ICMLW'26) Β· SSD (arXiv'25) Β· ODB (arXiv'25)
7. Benchmarks and Evaluation - Speculative Decoding Survey (Findings-ACL'24) Β· Parallel Text Generation Survey (arXiv'25) Β· Parallel Reasoning Survey (arXiv'25) Β· Spec-Bench (misc'24) Β· TTS Benchmark (ICLR'26) Β· dLLM Efficiency Evaluation (arXiv'25) Β· ParallelBench (ICLR'26)

πŸ“™ 1. Token-Level Parallel Decoding

Token-level methods shorten autoregressive generation by proposing, organizing, or verifying multiple token candidates or future positions in parallel.

1.1 Drafting with the Draft Model

An independent draft model proposes candidate tokens that the target LLM verifies; this includes foundational speculative-decoding frameworks, drafter training and distillation, and cross-architecture draft adaptation.

Date Title Authors Code Venue
2018.11 [Blockwise] Blockwise Parallel Decoding for Deep Autoregressive Models Mitchell Stern, Noam Shazeer, Jakob Uszkoreit (@UC Berkeley & Google Brain) [code] NIPS'18
2022.03 [SpecDec] Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation Heming Xia, Tao Ge et al. (@MSRA & PKU) [code] Findings-EMNLP'23
2022.11 Fast Inference from Transformers via Speculative Decoding Yaniv Leviathan, Matan Kalman, Yossi Matias (@Google Research) [code (unofficial)] ICML'23
2023.02 Accelerating Large Language Model Decoding with Speculative Sampling Charlie Chen et al. (@DeepMind) [code (unofficial)] arXiv'23
2023.02 [BiLD] Speculative Decoding with Big Little Decoder Sehoon Kim et al. (@UC Berkeley & ICSI & LBNL) [code] NeurIPS'23
2023.10 [DistillSpec] Improving Speculative Decoding via Knowledge Distillation Yongchao Zhou et al. (@Google Research & Google DeepMind & U Toronto & Princeton & Mila) ICLR'24
2023.10 Online Speculative Decoding Xiaoxuan Liu et al. (@UC Berkeley & UCSD & Google & SJTU) [code] ICML'24
2024.02 Decoding Speculative Decoding Minghao Yan et al. (@UW-Madison) [code] NAACL'25
2024.06 Towards Fast Multilingual LLM Inference: Speculative Decoding and Specialized Drafters Euiin Yi, Taehyeon Kim et al. (@KAIST & KT) [code] EMNLP'24
2024.08 [HASS] Learning Harmonized Representations for Speculative Sampling Lefan Zhang et al. (@Xiaohongshu) [code] ICLR'25
2024.10 Temperature-Centric Investigation of Speculative Decoding with Knowledge Distillation Siru Ouyang et al. (@UIUC & Microsoft) Findings-EMNLP'24
2024.10 [ParallelSpec] Parallel Drafter for Efficient Speculative Decoding Zilin Xiao et al. (@Rice & Tencent AI Lab & UIUC) arXiv'24
2024.12 [CTC-drafter] Speculative Decoding with CTC-based Draft Model for LLM Inference Acceleration Zhuofan Wen et al. (@CAS) [code] NeurIPS'24
2025.02 Accelerating LLM Inference with Lossless Speculative Decoding Algorithms for Heterogeneous Vocabularies Nadav Timor et al. (@Weizmann & Intel & d-Matrix) [code] ICML'25
2025.02 [CoS] Fast Large Language Model Collaborative Decoding via Speculation Jiale Fu et al. (@SEU) [code] ICML'25
2025.02 [GRIFFIN] Effective Token Alignment for Faster Speculative Decoding Shijing Hu et al. (@Fudan & NUS & SMU) [code] NeurIPS'25
2025.02 [CORAL] Learning Consistent Representations across Multi-step Training with Lighter Speculative Drafter Yepeng Weng et al. (@Lenovo Research) ACL'25
2025.02 [FR-Spec] Accelerating Large-Vocabulary Language Models via Frequency-Ranked Speculative Sampling Weilin Zhao, Tengyu Pan et al. (@THU & HIT & BUPT & OpenBMB & WeChat AI/Tencent) [code] ACL'25
2025.03 Training Domain Draft Models for Speculative Decoding: Best Practices and Insights Fenglu Hong et al. (@SambaNova) ICLRW'25
2025.05 Automatic Task Detection and Heterogeneous LLM Speculative Decoding Danying Ge et al. (@BNU) arXiv'25
2025.06 Mamba Drafters for Speculative Decoding Daewon Choi et al. (@KAIST & Amazon & SNU) Findings-EMNLP'25
2025.07 [OmniDraft] A cross-vocabulary, online adaptive drafter for on-device speculative decoding Ramchalam Kinattinkara Ramakrishnan, Zhaocong Yuan et al. (@Qualcomm AI Research) NeurIPS'25
2025.07 [SPECTRA] Faster Large Language Model Inference with Optimized Internal and External Speculation Nguyen-Khang Le et al. (@JAIST) ACL'25
2025.09 [RepSpec] Structural Re-parameterized Draft Model Training for Speculative Decoding Feiye Huo et al. (@PKU & Meituan) ICLR'26
2025.10 [AdaSPEC] Selective Knowledge Distillation for Efficient Speculative Decoders Yuezhou Hu et al. (@UC Berkeley & Georgia Tech & THU) [code] NeurIPS'25
2026.01 Flatter Tokens are More Valuable for Speculative Draft Model Training Jiaming Fan et al. (@SEU & NUIST & Qiyuan Tech) [code] ICLR'26
2026.03 [OnlineSPEC] When Drafts Evolve: Speculative Decoding Meets Online Learning Yu-Yang Qian et al. (@NJU & UCSD) [code] ICML'26

1.2 Drafting within the Target Model

The target model produces its own drafts through layer skipping, early exit, attached prediction heads, or internal feature prediction, removing the need for a separate full draft model.

Date Title Authors Code Venue
2023.09 [Draft & Verify] Lossless Large Language Model Acceleration via Self-Speculative Decoding Jun Zhang et al. (@ZJU & UC Irvine) [code] ACL'24
2023.11 [PaSS] Parallel Speculative Sampling Giovanni Monea, Armand Joulin, Edouard Grave (@EPFL & Apple) NeurIPSW'23
2024.01 [Medusa] Simple LLM Inference Acceleration Framework with Multiple Decoding Heads Tianle Cai et al. (@Princeton & Together AI & UIUC & CMU & UConn) [code] ICML'24
2024.01 [EAGLE] Speculative Sampling Requires Rethinking Feature Uncertainty Yuhui Li et al. (@PKU & Microsoft & University of Waterloo & Vector Institute) [code] ICML'24
2024.02 [GliDe with a CaPE] A Low-Hassle Method to Accelerate Speculative Decoding Cunxiao Du et al. (@SMU & NUS & PolyU & HPC-AI Tech & HIT & Tencent AI Lab) [code] ICML'24
2024.02 [Hydra] Sequentially-Dependent Draft Heads for Medusa Decoding Zachary Ankner et al. (@MIT & MosaicML) [code] COLM'24
2024.03 [ReDrafter] Recurrent Drafter for Fast Speculative Decoding in Large Language Models Yunfei Cheng et al. (@Apple) [code] arXiv'24
2024.04 [LayerSkip] Enabling Early Exit Inference and Self-Speculative Decoding Mostafa Elhoushi et al. (@Meta & U Toronto & CMU & UW-Madison & Dana-Farber) [code] ACL'24
2024.04 [Kangaroo] Lossless Self-Speculative Decoding for Accelerating LLMs via Double Early Exiting Fangcheng Liu et al. (@Huawei & Consumer Business Group, Huawei) [code] NeurIPS'24
2024.04 [MTP] Better & Faster Large Language Models via Multi-token Prediction Fabian Gloeckle et al. (@Meta/FAIR & Γ‰cole des Ponts ParisTech & UniversitΓ© Paris-Saclay) ICML'24
2024.05 [PPD] Hardware-Aware Parallel Prompt Decoding for Memory-Efficient Acceleration of LLM Inference Hao Mark Chen et al. (@Imperial College London & PolyU & Samsung AI Center) [code] Findings-EMNLP'25
2024.05 [S3D] A Simple and Cost-Effective Self-Speculative Decoding Scheme for Low-Memory GPUs Wei Zhong, Manasa Bharadwaj (@LG) arXiv'24
2024.06 [EESD] Speculative Decoding via Early-exiting for Faster LLM Inference with Thompson Sampling Control Mechanism Jiahao Liu, Qifan Wang et al. (@Meituan & Meta) Findings-ACL'24
2024.06 [EAGLE-2] Faster Inference of Language Models with Dynamic Draft Trees Yuhui Li et al. (@PKU & Microsoft & University of Waterloo & Vector Institute) [code] EMNLP'24
2024.06 [Amphista] Bi-directional Multi-head Decoding for Accelerating LLM Inference Zeping Li et al. (@AMD & PKU) NAACL'25
2024.08 [Clover-2] Accurate Inference for Regressive Lightweight Speculative Decoding Bin Xiao et al. (@Baichuan Inc. & BIT) [code] arXiv'24
2024.08 [KOALA] Enhancing Speculative Decoding for LLM via Multi-Layer Draft Heads with Adversarial Learning Kaiqi Zhang, Jing Zhao, Rui Chen (@DUT) CSCWD'25
2024.10 Draft on the Fly: Adaptive Self-Speculative Decoding using Cosine Similarity Michael R. Metel et al. (@Huawei & UdeM) Findings-EMNLP'24
2024.10 [SWIFT] On-the-Fly Self-Speculative Decoding for LLM Inference Acceleration Heming Xia et al. (@PolyU & ZJU & Sea AI Lab) [code] ICLR'25
2024.10 [Mixture of Attentions] Mixture of Attentions For Speculative Decoding Matthieu Zimmer et al. (@Huawei & UCL) [code] ICLR'25
2024.12 [DeepSeek-V3] DeepSeek-V3 Technical Report DeepSeek-AI (@DeepSeek) [code] arXiv'24
2025.02 On multi-token prediction for efficient LLM inference Somesh Mehra, Javier Alonso Garcia, Lukas Mauch (@Sony & EPFL) ICLRW'25
2025.03 [EAGLE-3] Scaling up Inference Acceleration of Large Language Models via Training-Time Test Yuhui Li et al. (@PKU & Microsoft & University of Waterloo & Vector Institute) [code] NeurIPS'25
2025.03 [Gumiho] A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding Jinze Li et al. (@HKU & XJTU & AMD) [code] ICML'25
2025.04 [DEL] Context-Aware Dynamic Exit Layer for Efficient Self-Speculative Decoding Hossein Entezari Zarch et al. (@USC) [code] COLM'25
2025.05 [CLaSp] In-Context Layer Skip for Self-Speculative Decoding Longze Chen et al. (@CAS & UCAS & SUTD & UNSW & Ritzz-AI) ACL'25
2025.05 [L-MTP] Leap Multi-Token Prediction Beyond Adjacent Context for Large Language Models Xiaohao Liu et al. (@NUS & HIT & THU & CAS & CSU) [code] NeurIPS'25
2025.05 [Beagle] Cross-Attention Speculative Decoding Wei Zhong et al. (@LG) arXiv'25
2025.06 [AdaDecode] Accelerating LLM Decoding with Adaptive Layer Parallelism Zhepei Wei et al. (@University of Virginia) [code] ICML'25
2025.07 Your LLM Knows the Future: Uncovering Its Multi-Token Prediction Potential Mohammad Samragh et al. (@Apple) arXiv'25
2025.10 [CAS-Spec] Cascade Adaptive Self-Speculative Decoding for On-the-Fly Lossless Inference Acceleration of LLMs Zhiyuan Ning et al. (@TeleAI & SJTU & HKUST) NeurIPS'25
2025.11 [FractalLLM] Lossless Self-Speculative Decoding with Layer Embedded Self-Compression Juhyeong Kim et al. (@Gachon University) [code] Findings-EMNLP'25
2025.11 [Speculative Streaming] Efficient and Scalable Speculative Decoding with Multi-Stream Attention Nikhil Bhendawade et al. (@Apple) [code] EMNLP'25
2025.12 [SparseSpec] Accelerating Large-Scale Reasoning Model Inference with Sparse Self-Speculative Decoding Yilong Zhao et al. (@UC Berkeley & MIT & UW & NVIDIA & Cornell & THU) [code] MLSys'26
2026.02 [mtp-lm] Multi-Token Prediction via Self-Distillation John Kirchenbauer et al. (@UMD & LLNL & Columbia & Together AI) [code] ICMLW'26
2026.03 [ESP] Efficient Training-Free Multi-Token Prediction via Embedding-Space Probing Raghavv Goel et al. (@Qualcomm AI Research) ICML'26

1.3 Retrieval-Based and Non-Parametric Drafting

Drafts come from retrieved text, n-gram or suffix caches, reusable context patterns, or parameter-free fixed-point/Jacobi procedures.

Date Title Authors Code Venue
2023.05 Accelerating Transformer Inference for Translation via Parallel Decoding Andrea Santilli et al. (@Sapienza University of Rome & University of TΓΌbingen & TΓΌbingen AI Center) [code] ACL'23
2023.11 [REST] Retrieval-Based Speculative Decoding Zhenyu He et al. (@PKU & Princeton) [code] NAACL'24
2024.02 [Lookahead Decoding] Break the Sequential Dependency of LLM Inference Using Lookahead Decoding Yichao Fu et al. (@UCSD & Google & UC Berkeley) [code] ICML'24
2024.02 [Ouroboros] Generating Longer Drafts Phrase by Phrase for Faster Speculative Decoding Weilin Zhao et al. (@THU & Shanghai AI Lab & ModelBest) [code] EMNLP'24
2024.03 [CLLMs] Consistency Large Language Models Siqi Kou et al. (@SJTU & UCSD) [code] ICML'24
2024.05 [NEST] Nearest Neighbor Speculative Decoding for LLM Generation and Attribution Minghan Li et al. (@Cohere & FAIR & University of Chicago & CMU & University of Waterloo) [code] NeurIPS'24
2024.06 [MSN] Make Some Noise: Unlocking Language Model Parallel Inference Capability through Noisy Training Yixuan Wang et al. (@HIT & Du Xiaoman) [code] EMNLP'24
2024.08 [Token Recycling] Turning Trash into Treasure: Accelerating Inference of Large Language Models with Token Recycling Xianzhen Luo et al. (@HIT & Du Xiaoman) [code] ACL'25
2024.11 [SuffixDecoding] Extreme Speculative Decoding for Emerging AI Applications Gabriele Oliaro et al. (@CMU & Snowflake) [code] NeurIPS'25
2024.11 [SAM Decoding] Speculative Decoding via Suffix Automaton Yuxuan Hu et al. (@Renmin University & THU) [code] ACL'25
2025.02 [DReSD] Dense Retrieval for Speculative Decoding Milan Gritta et al. (@Huawei & University of Sheffield) [code] Findings-ACL'25
2025.02 [RAPID] Long-Context Inference with Retrieval-Augmented Speculative Decoding Guanzheng Chen et al. (@NUS & DAMO Academy/Alibaba & Hupan Lab) [code] ICML'25
2025.02 [Hierarchy Drafting] Lossless Acceleration of Large Language Models with Hierarchical Drafting based on Temporal Locality in Speculative Decoding Sukmin Cho et al. (@KAIST) [code] Findings-NAACL'25
2025.03 [RASD] Retrieval-Augmented Speculative Decoding Guofeng Quan et al. (@Alibaba Cloud) Findings-ACL'25
2025.11 Faster In-Context Learning for LLMs via N-Gram Trie Speculative Decoding Jinglin Chen et al. (@Wuhan University & Xiaomi & SJTU) [code] EMNLP'25
2025.11 [Cacheback] Speculative Decoding With Nothing But Cache Zhiyao Ma et al. (@Yale University) [code] EMNLP'25
2025.12 [Jacobi Forcing] Fast and Accurate Causal Parallel Decoding using Jacobi Forcing Lanxiang Hu et al. (@UCSD & SJTU & Snowflake) [code] ICML'26

1.4 Draft Structure and Sampling Strategies

The main contribution concerns candidate-tree structure, multi-draft coupling and acceptance, adaptive draft length, verification order, or theoretical properties of speculative sampling.

Date Title Authors Code Venue
2023.05 [SpecInfer] Accelerating Large Language Model Serving with Tree-based Speculative Inference and Verification Xupeng Miao et al. (@CMU & THU & Stanford & SJTU & PKU & UCSD) [code] ASPLOS'24
2023.10 [SpecTr] Fast Speculative Decoding via Optimal Transport Ziteng Sun et al. (@Google Research) NeurIPS'23
2024.01 [MCSD] Multi-candidate Speculative Decoding Sen Yang et al. (@NJU) [code] NLPCC'25
2024.02 [Sequoia] Scalable and Robust Speculative Decoding Zhuoming Chen et al. (@CMU & Together AI & Yandex & HSE & FAIR) [code] NeurIPS'24
2024.02 Accelerated Speculative Sampling Based on Tree Monte Carlo Zhengmian Hu et al. (@University of Maryland) ICML'24
2024.03 Block Verification Accelerates Speculative Decoding Ziteng Sun et al. (@Google Research) ICLR'25
2024.05 [SLiM] Speculative Decoding with Hypothesis Reduction Chi-Heng Lin et al. (@Samsung Research America) Findings-NAACL'24
2024.05 Dynamic Speculation Lookahead Accelerates Speculative Decoding of Large Language Models Jonathan Mamou et al. (@Intel & Weizmann & Hebrew University) NeurIPSW'24
2024.06 [OPT-Tree] Speculative Decoding with Adaptive Draft Tree Structure Jikai Wang et al. (@Soochow University & Huawei Cloud) [code] TACL'25
2024.07 Graph-Structured Speculative Decoding Zhuocheng Gong et al. (@PKU & Meituan & Tianjin University & RUC) [code] Findings-ACL'24
2024.09 [DDD] Dynamic Depth Decoding: Faster Speculative Decoding for LLMs Oscar Brown et al. (@ML Research Labs & ANU) arXiv'24
2024.10 Multi-Draft Speculative Sampling: Canonical Decomposition and Theoretical Limits Ashish Khisti et al. (@Qualcomm AI Research & University of Toronto) ICLR'25
2024.11 A Theoretical Perspective for Speculative Decoding Algorithm Ming Yin et al. (@Princeton & Northwestern) NeurIPS'24
2024.11 [SpecHub] Provable Acceleration to Multi-Draft Speculative Decoding Ryan Sun et al. (@Lehigh University & Samsung Research America & University of Maryland) [code] EMNLP'24
2024.11 [SVIP] Draft Model Knows When to Stop: Self-Verification Speculative Decoding for Long-Form Generation Ziyin Zhang et al. (@SJTU & Tencent) [code] EMNLP'25
2024.12 [AdaEAGLE] Optimizing Speculative Decoding via Explicit Modeling of Adaptive Draft Structures Situo Zhang et al. (@SJTU) arXiv'24
2025.01 [Judge Decoding] Faster Speculative Sampling Requires Going Beyond Model Alignment Gregor Bachmann et al. (@Meta GenAI & ETH Zurich) ICLR'25
2025.02 [TETRIS] Optimal Draft Token Selection for Batch Speculative Decoding Zhaoxuan Wu, Zijian Zhou et al. (@SMART & NUS & MIT) [code] ACL'25
2025.02 Towards Optimal Multi-draft Speculative Decoding Zhengmian Hu et al. (@University of Maryland & Adobe Research & University of Massachusetts Amherst) ICLR'25
2025.02 Fuzzy Speculative Decoding for a Tunable Accuracy-Runtime Tradeoff Maximilian Holsman et al. (@Duke University) [code] Findings-ACL'25
2025.04 [HSDDW] Hierarchical Speculative Decoding with Dynamic Window Shensian Syu, Hung-yi Lee (@NTU) [code] Findings-NAACL'25
2025.04 Speculative Sampling via Exponential Races Szymon Kobus & Deniz GΓΌndΓΌz (@Imperial College London) Findings-ACL'25
2025.05 [AASD] Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification Jikai Wang et al. (@Soochow University & Huawei Cloud) EMNLP'25
2025.05 Traversal Verification for Speculative Tree Decoding Yepeng Weng et al. (@Lenovo Research & CAS) NeurIPS'25
2025.05 [STree] Speculative Tree Decoding for Hybrid State-Space Models Yangchao Wu et al. (@UCLA & Yale) [code] NeurIPS'25
2025.05 [BanditSpec] Adaptive Speculative Decoding via Bandit Algorithms Yunlong Hou et al. (@NUS & Sea AI Lab & SMU & Yale University) [code] ICML'25
2025.05 [HeteroSpec] Leveraging Contextual Heterogeneity for Efficient Speculative Decoding Siran Liu et al. (@PKU & SCITIX) ACL'26
2025.06 List-Level Distribution Coupling with Applications to Speculative Decoding and Lossy Compression Joseph Rowan et al. (@University of Toronto) NeurIPS'25
2025.06 [SpecBranch] Speculative Decoding via Hybrid Drafting and Rollback-Aware Branch Parallelism Yuhao Shen, Junyi Shen et al. (@ZJU & NUS & USTC) [code] ICLR'26
2025.07 A Drop-In Solution for On-the-Fly Adaptation of Speculative Decoding in Large Language Models Jiesong Liu et al. (@NCSU) ACL'25
2025.07 Faster Speculative Decoding via Effective Draft Decoder with Pruned Candidate Tree Huanran Zheng, Xiaoling Wang (@ECNU) [code] ACL'25
2025.09 [GTO] Bridging Draft Policy Misalignment: Group Tree Optimization for Speculative Decoding Shijing Hu et al. (@Fudan & NUS & SMU) [code] ICLR'26
2025.10 polybasic Speculative Decoding Through a Theoretical Perspective Ruilin Wang et al. (@Xiamen University & ByteDance & Peng Cheng Lab) ICML'25
2025.10 Not-a-Bandit: Provably No-Regret Drafter Selection in Speculative Decoding for LLMs Hongyi Liu et al. (@Rice & AWS & UCSD) [code] ICLR'26
2025.11 Global Resolution: Optimal Multi-Draft Speculative Sampling via Convex Optimization Rahul Krishna Thomas & Arka Pal (@Stanford & Ritual) ICLR'26
2025.11 Training-Free Loosely Speculative Decoding: Accepting Semantically Correct Drafts Beyond Exact Match Jinze Li et al. (@AMD & HKU) [code] ICLR'26
2026.01 Overcoming Joint Intractability with Lossless Hierarchical Speculative Decoding Yuxuan Zhou et al. (@Alibaba & UW) [code] ICLR'26
2026.03 [LTD] Learning To Draft: Adaptive Speculative Decoding with Reinforcement Learning Jiebin Zhang et al. (@PKU & MSRA) [code] ICLR'26
2026.03 [SSD] Speculative Speculative Decoding Tanishq Kumar, Tri Dao, Avner May (@Stanford & Princeton & Together AI) [code] ICLR'26
2026.04 [Cactus] Accelerating Auto-Regressive Decoding with Constrained Acceptance Speculative Sampling Yongchang Hao, Lili Mou (@University of Alberta & Amii & Canada CIFAR AI Chair) [code] ICLR'26
2026.05 [SpecBlock] Block-Iterative Speculative Decoding with Dynamic Tree Drafting Weijie Shi et al. (@HKUST & MetaX & ZJNU & Soochow University) [code] arXiv'26
2026.06 [JetSpec] Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting Lanxiang Hu et al. (@UCSD & NJU & StepFun) [code] arXiv'26

1.5 Cross-Technique Integration

Speculative decoding is integrated with another technique or application domain, including multimodal generation, quality/alignment/safety controls, watermarking, and accelerated reinforcement-learning rollouts.

Date Title Authors Code Venue
2023.11 Speculative Contrastive Decoding Hongyi Yuan et al. (@THU & Alibaba) ACL'24
2024.10 Inevitable Trade-off between Watermark Strength and Speculative Sampling Efficiency for Language Models Zhengmian Hu & Heng Huang (@University of Maryland) NeurIPS'24
2024.12 Constrained Decoding with Speculative Lookaheads Nishanth Nakshatri et al. (@Purdue University & AWS AI Labs) NAACL'25
2025.05 [DREAM] Drafting with Refined Target Features and Entropy-Adaptive Cross-Attention Fusion for Multimodal Speculative Decoding Yunhai Hu et al. (@NYU & UPenn & Cerebras Systems) [code] NeurIPS'25
2025.05 [MASSV] Multimodal Adaptation and Self-Data Distillation for Speculative Decoding of Vision-Language Models Mugilan Ganesan et al. (@University of Waterloo & Cerebras Systems) Findings-EMNLP'25
2025.07 [Spec-VLA] Speculative Decoding for Vision-Language-Action Models with Relaxed Acceptance Songsheng Wang et al. (@University of Macau & Infinigence AI & THU & Zhongguancun Academy) [code] EMNLP'25
2025.08 [SpecVLM] Enhancing Speculative Decoding of Video LLMs via Verifier-Guided Token Pruning Yicheng Ji et al. (@ZJU & PolyU & BUPT) [code] EMNLP'25
2025.08 Reward-Shifted Speculative Sampling Is An Efficient Test-Time Weak-to-Strong Aligner Bolian Li et al. (@Purdue University) [code] EMNLP'25
2025.08 Speculative Safety-Aware Decoding Xuekang Wang et al. (@BIT & CAS) [code] EMNLP'25
2025.09 [ViSpec] Accelerating Vision-Language Models with Vision-Aware Speculative Decoding Jialiang Kang et al. (@PKU & Huawei) [code] NeurIPS'25
2025.09 [SPEC-RL] Accelerating On-Policy Reinforcement Learning with Speculative Rollouts Bingshuai Liu et al. (@Xiamen University & Shopee & THU) [code] arXiv'25
2025.09 [FastGRPO] Accelerating Policy Optimization via Concurrency-aware Speculative Decoding and Online Draft Learning Yizhou Zhang et al. (@Lanzhou University & HKU & NUS) [code] ICLR'26
2025.10 [ReSpec] Towards Optimizing Speculative Decoding in Reinforcement Learning Systems Qiaoling Chen et al. (@NTU & THU & NUS & Qiji Zhifeng & Shanghai Innovation Institute) MLSys'26
2025.11 [DAS] Beat the long tail: Distribution-Aware Speculative Decoding for RL Training Zelei Shao et al. (@UIUC & Together AI & UCSD & Prime Intellect) MLSys'26
2025.11 [SpecActor] Fast LLM Post-training via Decoupled and Fastest-of-N Speculation Rongxin Cheng et al. (@SJTU & ByteDance) arXiv'25
2025.11 [TLT] Taming the Long-Tail: Efficient Reasoning RL Training with Adaptive Drafter Qinghao Hu et al. (@MIT & NVIDIA & ETH Zurich & MIT-IBM AI Lab & UMass Amherst) [code] ASPLOS'26
2025.12 [RLHFSpec] Breaking the Efficiency Bottleneck in RLHF Training via Adaptive Drafting Siqi Wang et al. (@Beihang University) arXiv'25
2026.01 [SRT] Accelerating Reinforcement Learning via Speculative Rollout with Tree-Structured Cache Chi-Chih Chang et al. (@Cornell & UIUC & THU & UW & ByteDance) NeurIPSW'25
2026.02 Improving the Trade-off Between Watermark Strength and Speculative Sampling Efficiency for Language Models Weiqing He et al. (@University of Pennsylvania) [code] ICLR'26

πŸ“™ 2. Sequence-Level Parallel Decoding

Sequence-level methods expose parallelism over semantic units, reasoning segments, partial chains, or complete candidate trajectories.

2.1 Multi-Sequence Parallel Generation

Several reasoning continuations or complete answer sequences are generated in parallel and then verified, corrected, selected, or aggregated at the semantic level.

Date Title Authors Code Venue
2023.11 [FastCoT] Fast Chain-of-Thought: A Glance of Future from Parallel Decoding Leads to Answers Faster Hongxuan Zhang et al. (@NJU & Ant Group) arXiv'23
2024.06 [SEED] Accelerating Reasoning Tree Construction via Scheduled Speculative Decoding Zhenglin Wang et al. (@SEU) [code] COLING'25
2024.07 [Speculative RAG] Enhancing Retrieval Augmented Generation through Drafting Zilong Wang et al. (@UCSD & Google) ICLR'25
2025.01 [RSD] Reward-Guided Speculative Decoding for Efficient LLM Reasoning Baohao Liao et al. (@University of Amsterdam & Salesforce AI Research) [code] ICML'25
2025.03 Speculative Decoding for Multi-Sample Inference Yiwei Li et al. (@BIT & Xiaohongshu) Findings-EMNLP'25
2025.04 [SpecReason] Fast and Accurate Inference-Time Compute via Speculative Reasoning Rui Pan et al. (@Princeton & CMU) [code] NeurIPS'25
2025.04 [Speculative Thinking] Enhancing Small-Model Reasoning with Large Model Guidance at Inference Time Wang Yang et al. (@Case Western Reserve University & Carnegie Mellon University) [code] COLM'25
2025.04 [SCoT] Efficient Reasoning for LLMs through Speculative Chain-of-Thought Jikai Wang et al. (@Soochow University & CUHK & Meta Stone & Shanghai AI Lab) [code] Findings-ACL'26
2025.05 [SpecSearch] Accelerating Large Language Model Reasoning via Speculative Search Zhihai Wang et al. (@USTC & Huawei & Tianjin University) [code] ICML'25
2025.05 [R2R] Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing Tianyu Fu et al. (@THU & Infinigence AI & SJTU) [code] NeurIPS'25
2025.06 [Lookahead Reasoning] Scaling Speculative Decoding with Lookahead Reasoning Yichao Fu et al. (@UCSD & SJTU & UIUC) [code] NeurIPS'25
2025.06 [STAND] Accelerated Test-Time Scaling with Model-Free Speculative Sampling Woomin Song et al. (@KAIST & Amazon AGI & AirSignal) EMNLP'25
2025.09 [SpecExit] Accelerating Large Reasoning Model via Speculative Exit Rubing Yang et al. (@Tencent) [code] ICML'26
2025.11 [SpecCoT] Accelerating Chain-of-Thought Reasoning through Speculative Exploration Junhan Shi et al. (@THU & Peng Cheng Lab & Xidian University) Findings-EMNLP'25
2026.02 [SemanticSpec] Beyond Tokens: Semantic-Aware Speculative Decoding for Efficient Inference by Probing Internal States Ximing Dong et al. (@Huawei & University of Manitoba & Queen's University) arXiv'26
2026.04 [SpecGuard] From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning Kiran Purohit et al. (@IIT Kharagpur & Adobe Research) Findings-ACL'26

2.2 Intra-Sequence Parallel Generation

One output is decomposed into structurally independent sections, subproblems, or graph nodes that can be generated concurrently.

Date Title Authors Code Venue
2023.07 [Skeleton-of-Thought] Prompting LLMs for Efficient Parallel Generation Xuefei Ning et al. (@THU & Microsoft & KU Leuven & Infinigence-AI) [code] ICLR'24
2023.08 [Graph of Thoughts] Solving Elaborate Problems with Large Language Models Maciej Besta et al. (@ETH Zurich & Warsaw University of Technology & Cledar) [code] AAAI'24
2024.02 [Plato] Plan to Efficiently Decode for Large Language Model Inference Shuowei Jin et al. (@University of Michigan & UC Berkeley & CMU & Duke & USC) COLM'25
2025.02 [PASTA] Learning to Keep a Promise: Scaling Language Model Decoding Parallelism with Learned Asynchronous Decoding Tian Jin et al. (@MIT & Google Research & Google DeepMind) ICML'25
2025.03 Accelerate Parallelizable Reasoning via Parallel Decoding within One Sequence Yijiong Yu (@THU & OpenCSG) [code] EMNLP'25
2025.04 [APR] Learning Adaptive Parallel Reasoning with Language Models Jiayi Pan et al. (@UC Berkeley & UCSF) [code] COLM'25
2025.06 [Multiverse] Your Language Models Secretly Decide How to Parallelize and Merge Generation Xinyu Yang et al. (@CMU & NVIDIA) [code] NeurIPS'25
2025.06 [Sprint] Enabling Interleaved Planning and Parallelized Execution in Reasoning Models Emil Biju et al. (@Stanford & Microsoft & Google) [code] NeurIPS'25
2025.06 [PCCoT] Parallel Continuous Chain-of-Thought with Jacobi Iteration Haoyi Wu et al. (@ShanghaiTech) [code] EMNLP'25
2025.09 Parallel Thinking, Sequential Answering: Bridging NAR and AR for Efficient Reasoning Qihang Ai, Haiyun Jiang (@NTU & SJTU) arXiv'25
2025.10 Parallel Loop Transformer for Efficient Test-Time Computation Scaling Bohong Wu et al. (@ByteDance) arXiv'25

2.3 Path Exploration and Test-Time Scaling

Parallel test-time compute is allocated across branching reasoning paths, with explicit control of exploration, pruning, early termination, and answer aggregation.

Date Title Authors Code Venue
2024.02 More Agents Is All You Need Junyou Li et al. (@Tencent) [code] TMLR'24
2024.10 [Speculative Rejection] Fast Best-of-N Decoding via Speculative Rejection Hanshi Sun et al. (@CMU & University of Virginia & UC Berkeley & Princeton & Fudan University & Google DeepMind) [code] NeurIPS'24
2025.02 [DPTS] Dynamic Parallel Tree Search for Efficient LLM Reasoning Yifu Ding et al. (@Beihang University & NTU & Wuhan University) [code] ACL'25
2025.05 [Group Think] Multiple Concurrent Reasoning Agents Collaborating at Token Level Granularity Chan-Jan Hsu et al. (@MediaTek Research) arXiv'25
2025.07 Adaptive Termination for Multi-round Parallel Reasoning: An Universal Semantic Entropy-Guided Framework Zenan Xu et al. (@Tencent & CUHK) arXiv'25
2025.09 [A2R] An Asymmetric Two-Stage Reasoning Framework for Parallel Reasoning Ziqi Wang et al. (@USTC & Baidu & USYD) arXiv'25
2025.09 [Parallel-R1] Towards Parallel Thinking via Reinforcement Learning Tong Zheng et al. (@Tencent AI Lab & University of Maryland & UNC Chapel Hill & City HKU & Washington University in St. Louis) [code] ICLR'26
2025.09 [ATTS] Asynchronous Test-Time Scaling via Conformal Prediction Jing Xiong et al. (@HKU & Huawei & UCL & CUHK) [code] ICLR'26
2025.10 Parallel Test-Time Scaling for Latent Reasoning Models Runyang You et al. (@PolyU & USTC & HIT Shenzhen & Shandong Jianzhu University) [code] ACL'26
2025.10 [DeepPrune] Parallel Scaling without Inter-trace Redundancy Shangqing Tu et al. (@THU & ShanghaiTech University) [code] Findings-ACL'26
2025.11 [DTS] Enhancing Large Reasoning Models via Decoding Tree Sketching Zicheng Xu et al. (@Johns Hopkins University & UNC Charlotte & Rice University & University of Minnesota) [code] ICML'26
2026.02 [Parallel-Probe] Towards Efficient Parallel Thinking via 2D Probing Tong Zheng et al. (@University of Maryland & Washington University in St. Louis & UNC Chapel Hill & City HKU & Tongji University) [code] ICML'26

πŸ“™ 3. Task-Level Parallel Decoding

Task-level methods coordinate concurrent agents, plans, tool calls, functions, or environment actions.

3.1 Multi-Agent Parallel Collaboration

Multiple LLM agents or teams solve and deliberate concurrently, then combine their results through voting, synthesis, or coordination.

Date Title Authors Code Venue
2023.08 [AutoGen] Enabling Next-Gen LLM Applications via Multi-Agent Conversations Qingyun Wu et al. (@Microsoft & PSU & UW & Xidian University) [code] COLM'24
2024.02 More Agents Is All You Need Junyou Li et al. (@Tencent) [code] TMLR'24
2024.06 [Mixture-of-Agents] Mixture-of-Agents Enhances Large Language Model Capabilities Junlin Wang et al. (@Duke University & Together AI & University of Chicago & Stanford University) [code] ICLR'25
2025.07 [M1-Parallel] Optimizing Sequential Multi-Step Tasks with Parallel LLM Agents Enhao Zhang et al. (@UW & Microsoft) ICMLW'25

3.2 Parallel and Speculative Agent Execution

Independent tools, functions, or predicted future actions are planned, dispatched, or speculatively executed in parallel.

Date Title Authors Code Venue
2023.12 [LLMCompiler] An LLM Compiler for Parallel Function Calling Sehoon Kim et al. (@UC Berkeley & ICSI & LBNL) [code] ICML'24
2024.05 [LLM-Tool Compiler] An LLM-Tool Compiler for Fused Parallel Function Calling Simranjit Singh et al. (@Microsoft & Southern Illinois University) arXiv'24
2024.10 Asynchronous Tool Usage for Real-Time Agents Antonio A. Ginart et al. (@Salesforce AI Research) arXiv'24
2024.12 [AsyncLM] Asynchronous LLM Function Calling In Gim, Seung-seob Lee, Lin Zhong (@Yale University) arXiv'24
2025.09 [DSP] Dynamic Speculative Agent Planning Yilin Guan et al. (@Johns Hopkins University & University of Alberta & University of British Columbia & Avey Research Center & Google DeepMind & UCSB) [code] ICLR'26
2025.10 [Speculative Actions] A Lossless Framework for Faster AI Agents Naimeng Ye et al. (@Columbia University) [code] ICLR'26
2025.12 Optimizing Agentic Language Model Inference via Speculative Tool Calls Daniel Nichols et al. (@UMD & LLNL) arXiv'25
2026.02 [W&D] Scaling Parallel Tool Calling for Efficient Deep Research Agents Xiaoqiang Lin et al. (@Salesforce AI Research) [code] ICLRW'26
2026.03 [RealtimeTool] Parallel Decoding for Real-Time LLM Function Calling Xiaoxin Shi et al. (@SJTU & Shanghai Innovation Institute & Fudan University) [code] ICML'26
2026.03 [PASTE] Parallelizing Tool Execution and LLM Generation for Low-Latency Agent Serving Yifan Sui et al. (@SJTU & Microsoft & Stevens Institute of Technology & Google & HKUST) arXiv'26

πŸ“™ 4. Architecture-Level Parallel Decoding

Architecture-level methods redesign training or generation so strict left-to-right token causality is no longer the only decoding path.

4.1 Model Architecture and Training

The work designs, trains, converts, or scales architectures that support non-left-to-right generation, including masked diffusion, block diffusion, and semi-autoregressive models.

Date Title Authors Code Venue
2017.11 [NAT] Non-Autoregressive Neural Machine Translation Jiatao Gu et al. (@Salesforce Research & HKU) [code] ICLR'18
2018.02 Deterministic Non-Autoregressive Neural Sequence Modeling by Iterative Refinement Jason Lee, Elman Mansimov, Kyunghyun Cho (@New York University & CIFAR) [code] EMNLP'18
2018.03 [Latent Transformer] Fast Decoding in Sequence Models Using Discrete Latent Variables Łukasz Kaiser et al. (@Google Brain) [code] ICML'18
2019.04 [Mask-Predict] Parallel Decoding of Conditional Masked Language Models Marjan Ghazvininejad et al. (@FAIR) [code] EMNLP'19
2021.07 [D3PM] Structured Denoising Diffusion Models in Discrete State-Spaces Jacob Austin et al. (@Google Research) [code] NeurIPS'21
2022.05 [Diffusion-LM] Diffusion-LM Improves Controllable Text Generation Xiang Lisa Li et al. (@Stanford University) [code] NeurIPS'22
2023.10 [SEDD] Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution Aaron Lou, Chenlin Meng, Stefano Ermon (@Stanford University) [code] ICML'24
2024.06 [RADD] Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data Jingyang Ou et al. (@RUC & Huawei) [code] ICLR'25
2024.06 [MD4] Simplified and Generalized Masked Diffusion for Discrete Data Jiaxin Shi et al. (@Google DeepMind) [code] NeurIPS'24
2024.06 [MDLM] Simple and Effective Masked Diffusion Language Models Subham Sekhar Sahoo et al. (@Cornell Tech) [code] NeurIPS'24
2024.10 [DiffuLLaMA] Scaling Diffusion Language Models via Adaptation from Autoregressive Models Shansan Gong et al. (@HKU & UIUC & Apple & Tencent AI Lab) [code] ICLR'25
2024.10 [SDTT] Beyond Autoregression: Fast LLMs via Self-Distillation Through Time Justin Deschenaux, Caglar Gulcehre (@EPFL / CLAIRE Lab) [code] ICLR'25
2025.02 [LLaDA] Large Language Diffusion Models Shen Nie et al. (@Renmin University of China & Ant Group) [code] NeurIPS'25
2025.03 [Block Diffusion] Interpolating Between Autoregressive and Diffusion Language Models Marianne Arriola et al. (@Cornell Tech & Stanford & Cohere) [code] ICLR'25
2025.06 [Duo] The Diffusion Duality Subham Sekhar Sahoo et al. (@Cornell Tech & EPFL & Cohere) [code] ICML'25
2025.08 [Dream 7B] Diffusion Large Language Models Jiacheng Ye et al. (@HKU & Huawei) [code] arXiv'25
2025.09 [SDLM] Sequential Diffusion Language Models Yangzhou Liu et al. (@Shanghai AI Lab & NJU & THU & Fudan University & CUHK & Soochow University & Donghua University) [code] arXiv'25
2025.09 [Fast-dLLM v2] Efficient Block-Diffusion LLM Chengyue Wu et al. (@HKU & NVIDIA & MIT) [code] ICLR'26
2025.10 [RND1] Training Diffusion Language Models at Scale using Autoregressive Models Radical Numerics Inc. (@Radical Numerics Inc.) [code] misc'25
2025.10 [E2D2] Encoder-Decoder Diffusion Language Models for Efficient Training and Inference Marianne Arriola et al. (@Cornell University) [code] NeurIPS'25
2025.10 [SDAR] A Synergistic Diffusion-AutoRegression Paradigm for Scalable Sequence Generation Shuang Cheng et al. (@Shanghai AI Lab & ZJU & UMD & SJTU & THU) [code] Findings-ACL'26
2025.12 [NBDiff] From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMs Yuchuan Tian et al. (@PKU & Huawei) [code] arXiv'25
2025.12 [Efficient-DLM] From Autoregressive to Diffusion Language Models, and Beyond in Speed Yonggan Fu et al. (@NVIDIA & Georgia Tech & UChicago & HKU & MIT) ICML'26
2025.12 [LLaDA2.0] Scaling Up Diffusion Language Models to 100B Tiwei Bie et al. (@Ant Group & Renmin University of China & ZJU & Westlake University & HKUST) [code] arXiv'25
2026.01 [Diffusion In Diffusion] Reclaiming Global Coherence in Semi-Autoregressive Diffusion Linrui Ma et al. (@Huawei) arXiv'26
2026.06 [MBD-LMs] Multi-Block Diffusion Language Models Yijie Jin et al. (@SJTU & XJTU & Huawei) [code] arXiv'26

4.2 Decoding and Sampling Algorithms

Inference algorithms accelerate trained non-autoregressive or diffusion language models through unmasking policies, confidence schedules, trajectory methods, or error correction.

Date Title Authors Code Venue
2025.05 [EB-Sampler] Accelerated Sampling from Masked Diffusion Models via Entropy Bounded Unmasking Heli Ben-Hamu et al. (@Meta) NeurIPS'25
2025.05 [Fast-dLLM] Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding Chengyue Wu et al. (@HKU & NVIDIA & MIT) [code] ICLR'26
2025.06 [SlowFast Sampling] Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principles Qingyan Wei et al. (@SJTU & Shanghai AI Lab) [code] ICLR'26
2025.07 [WINO] Wide-In, Narrow-Out: Revokable Decoding for Efficient and Effective DLLMs Feng Hong et al. (@SJTU & Apple) [code] ICLR'26
2025.08 [Prophet] Diffusion Language Models Know the Answer Before Decoding Pengxiang Li et al. (@PolyU HK & Dartmouth & Surrey & Sun Yat-sen University & ELLIS Institute TΓΌbingen & MPI-IS & TΓΌbingen AI Center) [code] ICLR'26
2025.09 [RWS] Reward-Weighted Sampling: Enhancing Non-Autoregressive Characteristics in Masked Diffusion LLMs Daehoon Gwak et al. (@KAIST & SKKU) EMNLP'25
2025.09 [LSD] Learnable Sampler Distillation for Discrete Diffusion Models Feiyang Fu et al. (@UESTC) [code] NeurIPS'25
2025.09 [ADJUST] Enabling Approximate Joint Sampling in Diffusion LMs Parikshit Bansal et al. (@UT-Austin) arXiv'25
2025.09 [Learn2PD] Learning to Parallel: Accelerating Diffusion Large Language Models via Learnable Parallel Decoding Wenrui Bao et al. (@UCF & Mobi.AI & HKUST) [code] ICLR'26
2025.09 [dParallel] Learnable Parallel Decoding for dLLMs Zigeng Chen et al. (@NUS) [code] ICLR'26
2025.10 [FreeDave] Free Draft-and-Verification: Toward Lossless Parallel Decoding for Diffusion Large Language Models Shutong Wu et al. (@UW-Madison) [code] NeurIPSW'25
2025.10 [LocalLeap] Accelerating Diffusion LLM Inference via Local Determinism Propagation Fanheng Kong et al. (@Kuaishou) [code] arXiv'25
2025.10 [Saber] Efficient Sampling with Adaptive Acceleration and Backtracking Enhanced Remasking for Diffusion Language Model in Code Generation Yihong Dong et al. (@PKU) [code] ACL'26
2025.12 [LoPA] Scaling dLLM Inference via Lookahead Parallel Decoding Chenkai Xu et al. (@SJTU) [code] arXiv'25
2025.12 [SchED] Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules Amr Mohamed et al. (@MBZUAI & Ecole Polytechnique) [code] Findings-ACL'26
2025.12 [CadLLM] Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibration Jucheng Shen et al. (@Rice & Intel & UT-Austin) [code] Findings-ACL'26
2025.12 Learning Unmasking Policies for Diffusion Language Models Metod Jazbec et al. (@Apple & UvA & MIT) [code] ICML'26
2026.01 [Order-Token Search] Improving Diffusion Language Model Decoding through Joint Search in Generation Order and Token Space Yangyi Shen et al. (@Stanford & ZJU) arXiv'26
2026.01 [d3LLM] Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation Yu-Yang Qian et al. (@UCSD & NJU & SJTU) [code] ICML'26
2026.02 [RDD] Reversible Diffusion Decoding for Diffusion Language Models Xinyun Wang et al. (@ECNU & THU & Oxford & ZJU & Rakuten Singapore) arXiv'26
2026.02 [ReMix] Rejection Mixing: Fast Semantic Propagation of Mask Tokens for Efficient DLLM Inference Yushi Ye et al. (@SJTU & Apple) [code] CVPR'26
2026.02 [Info-Gain Sampler] Improving Sampling for Masked Diffusion Models via Information Gain Kaisen Yang et al. (@THU & MIT & SJTU & Beihang University) [code] ICML'26
2026.03 [DOS] Dependency-Oriented Sampler for Masked Diffusion Language Models Xueyu Zhou et al. (@PolyU HK) Findings-ACL'26
2026.03 Confidence-Based Decoding is Provably Efficient for Diffusion Language Models Changxiao Cai et al. (@UMich & CUHK) arXiv'26

πŸ“™ 5. System and Infrastructure

System and infrastructure work turns algorithmic parallelism into reliable latency, throughput, memory, and SLO gains through serving, scheduling, and algorithm–runtime co-design.

5.1 Serving and Scheduling

Runtime and serving techniques optimize batching, GPU scheduling, pipelines, distributed execution, caching, or SLO-aware adaptation for parallel decoding.

Date Title Authors Code Venue
2024.04 [TriForce] Lossless Acceleration of Long Sequence Generation with Hierarchical Speculative Decoding Hanshi Sun et al. (@CMU & FAIR) [code] COLM'24
2024.04 [BASS] Batched Attention-optimized Speculative Sampling Haifeng Qian et al. (@AWS) Findings-ACL'24
2024.05 [DSI] Distributed Speculative Inference (DSI): Speculation Parallelism for Provably Faster Lossless Language Model Inference Nadav Timor et al. (@Weizmann & Intel & Texas A&M University) [code] ICLR'25
2024.05 [EMS-SD] Efficient Multi-sample Speculative Decoding for Accelerating Large Language Models Yunsheng Ni et al. (@Huawei) [code] NAACL'25
2024.06 [SpecExec] Massively Parallel Speculative Decoding For Interactive LLM Inference on Consumer Devices Ruslan Svirschevski et al. (@Yandex & HSE & Together AI & CMU & FAIR) [code] NeurIPS'24
2024.06 Optimized Speculative Sampling for GPU Hardware Accelerators Dominik Wagner et al. (@TH NΓΌrnberg & KAIST & Intel) [code] EMNLP'24
2024.08 [MagicDec] Breaking the Latency-Throughput Tradeoff for Long Context Generation with Speculative Decoding Ranajoy Sadhukhan et al. (@CMU & Moffett AI & Together AI) [code] ICLR'25
2024.08 [PEARL] Parallel Speculative Decoding with Adaptive Draft Length Tianyu Liu et al. (@USTC & Tencent & Shanghai AI Lab) [code] ICLR'25
2024.12 [Dovetail] A CPU/GPU Heterogeneous Speculative Decoding for LLM inference Libo Zhang et al. (@NUDT) [code] EMNLP'25
2025.01 Attention-Level Speculation Jack Cai et al. (@University of Toronto & Tenstorrent & University of Waterloo) [code] ICML'25
2025.02 [EasySpec] Layer-Parallel Speculative Decoding for Efficient Multi-GPU Utilization Yize Wu et al. (@CAS & UCAS) [code] NeurIPS'25
2025.03 [AdaSpec] Adaptive Speculative Decoding for Fast, SLO-Aware Large Language Model Serving Kaiyu Huang et al. (@Tongji University & HUST & CUHK Shenzhen) [code] SoCC'25
2025.09 [SubSpec] Speculate Deep and Accurate: Lossless and Training-Free Acceleration for Offloaded LLMs via Substitute Speculative Decoding Pei-Shuo Wang et al. (@NYCU & Cornell) [code] NeurIPS'25
2025.10 [Mirror-SD] Mirror Speculative Decoding: Breaking the Serial Barrier in LLM Inference Nikhil Bhendawade et al. (@Apple) arXiv'25
2025.10 [dInfer] An Efficient Inference Framework for Diffusion Language Models Yuxin Ma et al. (@Ant Group & ZJU & Westlake University & RUC & UCAS & SJTU) [code] arXiv'25
2025.12 [dLLM-Serve] Taming the Memory Footprint Crisis: System Design for Production Diffusion LLM Serving Jiakun Fan et al. (@Virginia Tech) [code] arXiv'25

5.2 Algorithm-Infrastructure Co-Design

The decoding algorithm is co-designed with deployment constraints or systems techniques such as quantization, KV-cache management, MoE execution, and hardware-aware verification.

Date Title Authors Code Venue
2024.04 [DeFT] Decoding with Flash Tree-attention for Efficient Tree-structured LLM Inference Jinwei Yao et al. (@Westlake University & ZJU & CMU & UIUC & HKUST) [code] ICLR'25
2024.10 [QSpec] Speculative Decoding with Complementary Quantization Schemes Juntao Zhao et al. (@HKU & HIT) [code] EMNLP'25
2025.02 [QuantSpec] Self-Speculative Decoding with Hierarchical Quantized KV Cache Rishabh Tiwari et al. (@UC Berkeley & Apple & ICSI & LBNL) [code] ICML'25
2025.02 [SpecPrefill] Speculative Prefill: Turbocharging TTFT with Lightweight and Training-Free Token Importance Estimation Jingyu Liu et al. (@University of Chicago & CMU) [code] ICML'25
2025.03 [ML-SpecQD] Multi-Level Speculative Decoding with Quantized Drafts Evangelos Georganas et al. (@Intel) arXiv'25
2025.03 [SpeCache] Speculative Key-Value Caching for Efficient Generation of LLMs Shibo Jie et al. (@PKU & Huawei) ICML'25
2025.05 [MoESD] Unveil Speculative Decoding' s Potential for Accelerating Sparse MoE Zongle Huang et al. (@THU & Huawei) NeurIPS'25
2025.06 [Cascade] Utility-Driven Speculative Decoding for Mixture-of-Experts Anish Saxena et al. (@Georgia Tech & NVIDIA) arXiv'25
2025.10 [AsyncSpade] Efficient Test-Time Scaling with Asynchronous Sparse Decoding Shuqing Luo et al. (@UNC Chapel Hill & Johns Hopkins & UCLA) [code] ICML'26
2025.12 [Yggdrasil] Bridging Dynamic Speculation and Static Runtime for Latency-Optimal Tree-Based LLM Decoding Yue Guan, Changming Yu et al. (@SJTU & Qizhi Institute & UCSD) NeurIPS'25

Related Serving Engines and Libraries

These software projects support speculative or parallel decoding but are not paper records in the taxonomy, so they are not included in the classified-paper counts above.

Name Type Documented Support Code
vLLM Serving engine EAGLE / EAGLE-3, MTP, DFlash, draft-model SD, n-gram, suffix decoding [code]
SGLang Serving engine EAGLE-2 / EAGLE-3, MTP, DFlash, draft-model SD, n-gram [code]
TensorRT-LLM Serving engine Medusa, EAGLE, ReDrafter, n-gram, draft-model SD, lookahead decoding [code]
TGI Serving engine Medusa, n-gram speculation [code]
llama.cpp Serving engine Draft-model speculative decoding [code]
NVIDIA Triton Serving engine Draft-model SD, EAGLE-3 tutorials [code]
Speculators Library Unified abstraction & training for SD drafters (vLLM-native) [code]
TorchSpec Training framework PyTorch-native large-scale SD drafter training [code]

πŸ“™ 6. Hybrid Methods

Hybrid methods make the deliberate combination of multiple decoding or generative paradigms their central contribution.

6.1 Non-AR Drafter for AR Targets

A non-autoregressive or semi-autoregressive language model drafts multiple tokens in parallel for verification by an autoregressive target.

Date Title Authors Code Venue
2024.08 [SpecDiff] Speculative Diffusion Decoding: Accelerating Language Generation through Diffusion Jacob K. Christopher et al. (@University of Virginia & LLNL) NAACL'25
2025.10 [DiffuSpec] Unlocking Diffusion Language Models for Speculative Decoding Guanghao Li, Zhihui Fu et al. (@THU & SUSTech & OPPO Research Institute) [code (unofficial)] Findings-ACL'26
2025.11 [SpecDiff-2] Scaling Diffusion Drafter Alignment For Faster Speculative Decoding Jameson Sandler et al. (@University of Virginia) MLSys'26
2025.12 [DEER] Draft with Diffusion, Verify with Autoregressive Models Zicong Cheng, Guo-Wei Yang et al. (@THU & Proxseer Inc & SJTU) [code] arXiv'25
2026.01 [DART] Diffusion-Inspired Speculative Decoding for Fast LLM Inference Fuliang Liu, Xue Li et al. (@NJU & Alibaba) [code] arXiv'26
2026.02 [DFlash] Block Diffusion for Flash Speculative Decoding Jian Chen, Yesheng Liang, Zhijian Liu (@UCSD) [code] ICML'26
2026.06 [JetSpec] Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting Lanxiang Hu et al. (@UCSD) [code] arXiv'26
2026.07 [DSpark] Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation Xin Cheng et al. (@DeepSeek-AI & PKU) [code] arXiv'26
2026.07 [AdaFlash] Adaptive Speculative Decoding via On-Policy Distilled Diffusion Drafters Yu-Yang Qian et al. (@NJU & Huawei) arXiv'26

6.2 Speculation for Other Generative Paradigms

The draft-then-verify principle is transferred to diffusion language models, model cascades, continuous diffusion models, and visual autoregressive generation.

Date Title Authors Code Venue
2023.12 [CS Drafting] Cascade Speculative Drafting for Even Faster LLM Inference Ziyi Chen et al. (@UIUC) [code] NeurIPS'24
2024.05 Faster Cascades via Speculative Decoding Harikrishna Narasimhan et al. (@Google Research) ICLR'25
2024.10 [SJD] Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding Yao Teng, Han Shi et al. (@HKU & Huawei & CUHK & THU & SJTU & Infinigence AI) [code] ICLR'25
2024.10 [LANTERN] Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding Doohyuk Jang, Sihwan Park et al. (@KAIST & Intel & AITRICS) [code] ICLR'25
2025.01 Accelerated Diffusion Models via Speculative Sampling Valentin De Bortoli, Alexandre Galashov et al. (@Google DeepMind) ICML'25
2025.05 [ASD] Diffusion Models are Secretly Exchangeable: Parallelizing DDPMs via Auto Speculation Hengyuan Hu, Aniket Das et al. (@Stanford) ICML'25
2025.06 [APD] Accelerating Diffusion LLMs via Adaptive Parallel Decoding Daniel Israel et al. (@UCLA) [code] NeurIPS'25
2025.09 [Spiffy] Structuring The Future: Diffusion LLM Speculative Decoding via Calibrated Draft Graphs Sudhanshu Agrawal, Risheek Garrepalli et al. (@Qualcomm AI Research) ICMLW'26
2025.10 [SSD] Self Speculative Decoding for Diffusion Large Language Models Yifeng Gao, Ziang Ji et al. (@SJTU & USTC & Xidian University & Shanghai AI Lab & Huawei) arXiv'25
2025.11 [ODB] Orchestrating Dual-Boundaries: An Arithmetic Intensity Inspired Acceleration Framework for Diffusion Language Models Linye Wei, Wenjue Chen et al. (@PKU) [code] arXiv'25

Related hybrid resource. Speculative Cascades β€” A hybrid approach for smarter, faster LLM inference (Google Research blog, 2025) connects routing, selection, and speculation into one production-oriented framework.

πŸ“™ 7. Benchmarks and Evaluation

This layer organizes and evaluates the field through related surveys, benchmark suites, and diagnostic studies.

Related surveys.

Date Title Authors Code Venue
2024.01 Unlocking Efficiency in Large Language Model Inference: A Comprehensive Survey of Speculative Decoding Heming Xia et al. (@PolyU & PKU & MSRA & Alibaba) [code] Findings-ACL'24
2025.08 A Survey on Parallel Text Generation: From Parallel Decoding to Diffusion Language Models Lingzhe Zhang, Liancheng Fang et al. (@PKU & UIC & THU & XPENG & Alibaba & HKUST(GZ)) [list] arXiv'25
2025.10 A Survey on Parallel Reasoning Ziqi Wang, Boye Niu et al. (@USTC & Baidu & USYD) [list] arXiv'25

Benchmark suites.

Date Title Authors Code Venue
2024.01 [Spec-Bench] A Comprehensive Benchmark and Unified Evaluation Platform for Speculative Decoding Heming Xia et al. (@PolyU & PKU & MSRA & Alibaba) [code] misc'24
2025.09 Scaling Up, Speeding Up: A Benchmark of Speculative Decoding for Efficient LLM Test-Time Scaling Shengyin Sun, Yiming Li et al. (@CityU HK & Huawei) [code] ICLR'26

Diagnostic studies.

Date Title Authors Code Venue
2025.10 How Efficient Are Diffusion Language Models? A Critical Examination of Efficiency Evaluation Practices Han Peng, Peiyu Liu et al. (@RUC & UIBE & THU & CityU HK) arXiv'25
2025.10 [ParallelBench] Understanding the Trade-offs of Parallel Decoding in Diffusion LLMs Wonjun Kang, Kevin Galim et al. (@FuriosaAI & UW-Madison & Microsoft & UC Berkeley & SNU & KRAFTON AI) [code] ICLR'26

Related Awesome Lists


🀝 Contributing

Contributions are very welcome! If you would like to add a paper, fix metadata (venue, affiliation, links), or propose a new category, please open a pull request or an issue. When adding a paper, please keep entries in chronological order within each section and follow the existing table format: Date | [Alias] Title (linked to paper) | Authors (@Affiliation) | Code (+ stars badge) | Venue. Where information is uncertain, please mark it as N/A rather than guessing.


πŸ“ Citation

🌟 If you find this repository useful, please consider citing our survey paper: 🌟

@misc{survey-parallel-decoding,
  title        = {Parallel Decoding of Language Models: A Survey},
  author       = {Yu-Yang Qian and Jia-Chen Liu and Hao-Cong Wu and Lanxiang Hu and Hao Zhang and Peng Zhao},
  year         = {2026},
  note         = {\url{https://github.com/ZinYY/Awesome-Parallel-Decoding}}
}

Our survey paper is coming soon, stay tuned!

About

Parallel Decoding of Language Models: A Survey

Topics

Resources

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors