推理加速论文趋势 2023–2026

主要结论

• 推理加速论文从 2023 年的 54 篇增长到 2026 年的 91 篇，三年间扩大到约 1.7 倍。
• NeurIPS 是发推理加速论文最多的会议，2023–2026 年共 90 篇，占四个会议总量的 35%。
• 交叉方向里涨得最快的是深度学习 · large language models：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 9 篇。
• 在「效率与系统」的 5 个主题里，它的 2026 年论文数排第 5；同组增长最快的硬件协同是 4.4 倍，推理加速是 1.7 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026A Method-Class Divide in Sub-4-Bit Quantization:\\Iterative vs.\ Single-Pass Sensitivity to Data Composition
• ICML 2026Consistency Deep Equilibrium Models
• ICLR 2026OrderDP: A Theoretically Guaranteed Lossless Dynamic Data Pruning Framework
• CVPR 2026GeoRK2 Geometry-Guided Runge-Kutta Integration for Diffusion Transformer Acceleration
• NeurIPS 2025PolarQuant: Leveraging Polar Transformation for Key Cache Quantization and Decoding Acceleration
• ICML 2025RePaViT: Scalable Vision Transformer Acceleration via Structural Reparameterization on Feedforward Network Layers
• ICLR 2025Progressive distillation induces an implicit curriculum
• CVPR 2025Adapting Dense Matching for Homography Estimation with Grid-based Acceleration
• NeurIPS 2024Provable Acceleration of Nesterov's Accelerated Gradient for Asymmetric Matrix Factorization and Linear Neural Networks
• ICML 2024Diversified Batch Selection for Training Acceleration
• ICLR 2024Embarrassingly Simple Dataset Distillation
• CVPR 2024Data-Free Quantization via Pseudo-label Filtering
• NeurIPS 2023Continuous-time Analysis of Anchor Acceleration
• ICML 2023Auxiliary Modality Learning with Generalized Curriculum Distillation
• ICLR 2023Accurate Neural Training with 4-bit Matrix Multiplications at Standard Formats
• CVPR 2023NIPQ: Noise Proxy-Based Integrated Pseudo-Quantization
• NeurIPS 2026AcceleGrad#: Adaptive Geometry-Aware Acceleration
• ICML 2026PsumQuant: In-line Post-training Partial Sum Quantizer for Energy Efficient NPU Inference
• ICLR 2026NLI : Non-uniform Linear Interpolation Approximation of Nonlinear Operations for Efficient LLMs Inference
• CVPR 2026SODA Sensitivity-Oriented Dynamic Acceleration for Diffusion Transformer

全部推理加速论文

共 259 篇，第 3 / 6 页

101. ICLR 2025Effective Interplay between Sparsity and Quantization: From Theory to Practice
102. ICLR 2025Elucidating the Preconditioning in Consistency Distillation
103. ICLR 2025Methods for Convex $(L_0,L_1)$-Smooth Optimization: Clipping, Acceleration, and Adaptivity
104. ICLR 2025Model Equality Testing: Which Model is this API Serving?
105. ICLR 2025Nesterov acceleration in benignly non-convex landscapes
106. ICLR 2025One Step Diffusion via Shortcut Models
107. ICLR 2025Optimizing $(L_0, L_1)$-Smooth Functions by Gradient Methods
108. ICLR 2025PALMBENCH: A COMPREHENSIVE BENCHMARK OF COMPRESSED LARGE LANGUAGE MODELS ON MOBILE PLATFORMS
109. ICLR 2025PEARL: Parallel Speculative Decoding with Adaptive Draft Length
110. ICLR 2025PFDiff: Training-Free Acceleration of Diffusion Models Combining Past and Future Scores
111. ICLR 2025Progressive distillation induces an implicit curriculum
112. ICLR 2025Quamba: A Post-Training Quantization Recipe for Selective State Space Models
113. ICLR 2025Restructuring Vector Quantization with the Rotation Trick
114. ICLR 2025RTop-K: Ultra-Fast Row-Wise Top-K Selection for Neural Network Acceleration on GPUs
115. ICLR 2025SageAttention: Accurate 8-Bit Attention for Plug-and-play Inference Acceleration
116. ICLR 2025Scaling FP8 training to trillion-token LLMs
117. ICLR 2025SGD with memory: fundamental properties and stochastic acceleration
118. ICLR 2025SVDQuant: Absorbing Outliers by Low-Rank Component for 4-Bit Diffusion Models
119. ICML 2025any4: Learned 4-bit Numeric Representation for LLMs
120. ICML 2025BanditSpec: Adaptive Speculative Decoding via Bandit Algorithms
121. ICML 2025COGNATE: Acceleration of Sparse Tensor Programs on Emerging Hardware using Transfer Learning
122. ICML 2025Compress then Serve: Serving Thousands of LoRA Adapters with Little Overhead
123. ICML 2025Constant Stepsize Local GD for Logistic Regression: Acceleration by Instability
124. ICML 2025Distilling the Knowledge in Data Pruning
125. ICML 2025Ensemble Distribution Distillation via Flow Matching
126. ICML 2025Fast Large Language Model Collaborative Decoding via Speculation
127. ICML 2025Learning to Keep a Promise: Scaling Language Model Decoding Parallelism with Learned Asynchronous Decoding
128. ICML 2025Matryoshka Quantization
129. ICML 2025Metadata Conditioning Accelerates Language Model Pre-training
130. ICML 2025On Exact Bit-level Reversible Transformers Without Changing Architecture
131. ICML 2025Optimizing Large Language Model Training Using FP4 Quantization
132. ICML 2025Quamba2: A Robust and Scalable Post-training Quantization Framework for Selective State Space Models
133. ICML 2025RePaViT: Scalable Vision Transformer Acceleration via Structural Reparameterization on Feedforward Network Layers
134. ICML 2025SADA: Stability-guided Adaptive Diffusion Acceleration
135. ICML 2025SageAttention2: Efficient Attention with Thorough Outlier Smoothing and Per-thread INT4 Quantization
136. ICML 2025Scalable Model Merging with Progressive Layer-wise Distillation
137. ICML 2025SKIM: Any-bit Quantization Pushing The Limits of Post-Training Quantization
138. ICML 2025Sparsing Law: Towards Large Language Models with Greater Activation Sparsity
139. ICML 2025Streamline Without Sacrifice - Squeeze out Computation Redundancy in LMM
140. ICML 2025When Dynamic Data Selection Meets Data Augmentation: Achieving Enhanced Training Acceleration
141. NeurIPS 2025$\text{S}^2$Q-VDiT: Accurate Quantized Video Diffusion Transformer with Salient Data and Sparse Token Distillation
142. NeurIPS 2025Accelerating Parallel Diffusion Model Serving with Residual Compression
143. NeurIPS 2025Advancing Wasserstein Convergence Analysis of Score-Based Models: Insights from Discretization and Second-Order Acceleration
144. NeurIPS 2025Breaking AR’s Sampling Bottleneck: Provable Acceleration via Diffusion Language Models
145. NeurIPS 2025CAS-Spec: Cascade Adaptive Self-Speculative Decoding for On-the-Fly Lossless Inference Acceleration of LLMs
146. NeurIPS 2025EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test
147. NeurIPS 2025EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models
148. NeurIPS 2025FALQON: Accelerating LoRA Fine-tuning with Low-Bit Floating-Point Arithmetic
149. NeurIPS 2025FP4 All the Way: Fully Quantized Training of Large Language Models
150. NeurIPS 2025How to build a consistency model: Learning flow maps via self-distillation

常见问题

2023–2026 年四大会议一共有多少篇推理加速论文？

共 259 篇，按年份分别是 2023 年 54 篇、2024 年 50 篇、2025 年 64 篇、2026 年 91 篇。

NeurIPS 2026 有多少篇推理加速论文？

NeurIPS 2026 有 33 篇推理加速论文，上一届是 15 篇。

ICLR 2026 有多少篇推理加速论文？

ICLR 2026 有 17 篇推理加速论文，上一届是 18 篇。

ICML 2026 有多少篇推理加速论文？

ICML 2026 有 26 篇推理加速论文，上一届是 22 篇。

CVPR 2026 有多少篇推理加速论文？

CVPR 2026 有 15 篇推理加速论文，上一届是 9 篇。

哪个会议的推理加速论文最多？

NeurIPS，2023–2026 年共 90 篇。