{"total":107,"rows":[{"id":"2410.10781","title":"When Attention Sink Emerges in Language Models: An Empirical View","created":"2024-10-14","cited_by_count":208,"area":"inference"},{"id":"2404.14294","title":"A Survey on Efficient Inference for Large Language Models","created":"2024-04-22","cited_by_count":205,"area":"inference"},{"id":"2401.06104","title":"Transformers are Multi-State RNNs","created":"2024-01-11","cited_by_count":120,"area":"inference"},{"id":"2402.12374","title":"Sequoia: Scalable, Robust, and Hardware-aware Speculative Decoding","created":"2024-02-19","cited_by_count":103,"area":"inference"},{"id":"2312.15234","title":"Towards Efficient Generative Large Language Model Serving: A Survey from Algorithms to Systems","created":"2023-12-23","cited_by_count":102,"area":"inference"},{"id":"2506.06266","title":"Cartridges: Lightweight and general-purpose long context representations via self-study","created":"2025-06-06","cited_by_count":86,"area":"inference"},{"id":"2402.09398","title":"Get More with LESS: Synthesizing Recurrence with KV Cache Compression for Efficient LLM Inference","created":"2024-02-14","cited_by_count":75,"area":"inference"},{"id":"2510.00636","title":"Expected Attention: KV Cache Compression by Estimating Attention from Future Queries Distribution","created":"2025-10-01","cited_by_count":74,"area":"inference"},{"id":"2410.01699","title":"Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding","created":"2024-10-02","cited_by_count":66,"area":"inference"},{"id":"2412.10319","title":"SCBench: A KV Cache-Centric Analysis of Long-Context Methods","created":"2024-12-13","cited_by_count":60,"area":"inference"},{"id":"2407.01527","title":"KV Cache Compression, But What Must We Give in Return? A Comprehensive Benchmark of Long Context Capable Approaches","created":"2024-07-01","cited_by_count":42,"area":"inference"},{"id":"2410.04466","title":"Large Language Model Inference Acceleration: A Comprehensive Hardware Perspective","created":"2024-10-06","cited_by_count":38,"area":"inference"},{"id":"2409.18962","title":"Exploring Token Pruning in Vision State Space Models","created":"2024-09-27","cited_by_count":28,"area":"inference"},{"id":"2410.20771","title":"MrT5: Dynamic Token Merging for Efficient Byte-level Language Models","created":"2024-10-28","cited_by_count":28,"area":"inference"},{"id":"2502.13063","title":"Cramming 1568 Tokens into a Single Vector and Back Again: Exploring the Limits of Embedding Space Capacity","created":"2025-02-18","cited_by_count":28,"area":"inference"},{"id":"2503.18292","title":"Jenga: Effective Memory Management for Serving LLM with Heterogeneity","created":"2025-03-24","cited_by_count":25,"area":"inference"},{"id":"2410.05589","title":"ParallelSpec: Parallel Drafter for Efficient Speculative Decoding","created":"2024-10-08","cited_by_count":22,"area":"inference"},{"id":"2504.19449","title":"R-Sparse: Rank-Aware Activation Sparsity for Efficient LLM Inference","created":"2025-04-28","cited_by_count":21,"area":"inference"},{"id":"2406.05317","title":"LoCoCo: Dropping In Convolutions for Long Context Compression","created":"2024-06-08","cited_by_count":19,"area":"inference"},{"id":"2407.15504","title":"Fundamental Limits of Prompt Compression: A Rate-Distortion Framework for Black-Box Language Models","created":"2024-07-22","cited_by_count":19,"area":"inference"},{"id":"2505.17272","title":"Zebra-Llama: Towards Extremely Efficient Hybrid Models","created":"2025-05-22","cited_by_count":18,"area":"inference"},{"id":"2507.08143","title":"Compactor: Calibrated Query-Agnostic KV Cache Compression with Approximate Leverage Scores","created":"2025-07-10","cited_by_count":17,"area":"inference"},{"id":"2603.03251","title":"Speculative Speculative Decoding","created":"2026-03-03","cited_by_count":15,"area":"inference"},{"id":"2411.03786","title":"The N-Grammys: Accelerating Autoregressive Inference with Learning-Free Batched Speculation","created":"2024-11-06","cited_by_count":14,"area":"inference"},{"id":"2406.01698","title":"Demystifying AI Platform Design for Distributed Inference of Next-Generation LLM models","created":"2024-06-03","cited_by_count":12,"area":"inference"},{"id":"2501.14808","title":"HyGen: Efficient LLM Serving via Elastic Online-Offline Request Co-location","created":"2025-01-15","cited_by_count":10,"area":"inference"},{"id":"2505.14969","title":"STree: Speculative Tree Decoding for Hybrid State-Space Models","created":"2025-05-20","cited_by_count":10,"area":"inference"},{"id":"2405.14105","title":"Distributed Speculative Inference (DSI): Speculation Parallelism for Provably Faster Lossless Language Model Inference","created":"2024-05-23","cited_by_count":9,"area":"inference"},{"id":"2410.01485","title":"A Little Goes a Long Way: Efficient Long Context Training and Inference with Partial Contexts","created":"2024-10-02","cited_by_count":9,"area":"inference"},{"id":"2410.23079","title":"BUZZ: Beehive-structured Sparse KV Cache with Segmented Heavy Hitters for Efficient LLM Inference","created":"2024-10-30","cited_by_count":9,"area":"inference"},{"id":"2502.12574","title":"HeadInfer: Memory-Efficient LLM Inference by Head-wise Offloading","created":"2025-02-18","cited_by_count":9,"area":"inference"},{"id":"2503.10135","title":"Gumiho: A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding","created":"2025-03-13","cited_by_count":9,"area":"inference"},{"id":"2503.16163","title":"SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs","created":"2025-03-20","cited_by_count":9,"area":"inference"},{"id":"2509.15763","title":"UniGist: Towards General and Hardware-aligned Sequence-level Long Context Compression","created":"2025-09-19","cited_by_count":9,"area":"inference"},{"id":"2406.06110","title":"Recurrent Context Compression: Efficiently Expanding the Context Window of LLM","created":"2024-06-10","cited_by_count":8,"area":"inference"},{"id":"2505.22135","title":"RAD: Redundancy-Aware Distillation for Hybrid Models via Self-Speculative Decoding","created":"2025-05-28","cited_by_count":8,"area":"inference"},{"id":"2402.09360","title":"HiRE: High Recall Approximate Top-$k$ Estimation for Efficient LLM Inference","created":"2024-02-14","cited_by_count":7,"area":"inference"},{"id":"2405.17951","title":"Efficient Time Series Processing for Transformers and State-Space Models through Token Merging","created":"2024-05-28","cited_by_count":7,"area":"inference"},{"id":"2605.05971","title":"Training Transformers for KV Cache Compressibility","created":"2026-05-07","cited_by_count":7,"area":"inference"},{"id":"2606.20537","title":"Execution-State Capsules: Graph-Bound Execution-State Checkpoint and Restore for Low-Latency, Small-Batch, On-Device Physical-AI Serving","created":"2026-06-18","cited_by_count":7,"area":"inference"},{"id":"2604.20920","title":"Simplified Sparse Attention via Gist Tokens","created":"2026-04-22","cited_by_count":6,"area":"inference"},{"id":"2503.11816","title":"Key, Value, Compress: A Systematic Exploration of KV Cache Compression Techniques","created":"2025-03-14","cited_by_count":5,"area":"inference"},{"id":"2503.22196","title":"EdgeInfinite: A Memory-Efficient Infinite-Context Transformer for Edge Devices","created":"2025-03-28","cited_by_count":5,"area":"inference"},{"id":"2509.19873","title":"SpecMamba: Accelerating Mamba Inference on FPGA with Speculative Decoding","created":"2025-09-24","cited_by_count":5,"area":"inference"},{"id":"2602.05191","title":"Double-P: Hierarchical Top-P Sparse Attention for Long-Context LLMs","created":"2026-02-05","cited_by_count":5,"area":"inference"},{"id":"2605.08234","title":"When Does Value-Aware KV Eviction Help? A Fixed-Contract Diagnostic for Non-Monotone Cache Compression","created":"2026-05-07","cited_by_count":5,"area":"inference"},{"id":"2406.07528","title":"QuickLLaMA: Query-aware Inference Acceleration for Large Language Models","created":"2024-06-11","cited_by_count":4,"area":"inference"},{"id":"2412.19325","title":"Performance Control in Early Exiting to Deploy Large Models at the Same Cost of Smaller Ones","created":"2024-12-26","cited_by_count":4,"area":"inference"},{"id":"2503.11132","title":"X-EcoMLA: Upcycling Pre-Trained Attention into MLA for Efficient and Extreme KV Compression","created":"2025-03-14","cited_by_count":4,"area":"inference"},{"id":"2602.07397","title":"Scout Before You Attend: Sketch-and-Walk Sparse Attention for Efficient LLM Inference","created":"2026-02-07","cited_by_count":4,"area":"inference"},{"id":"2605.05219","title":"Sparse Prefix Caching for Hybrid and Recurrent LLM Serving","created":"2026-04-17","cited_by_count":4,"area":"inference"},{"id":"2606.09659","title":"End-to-End Context Compression at Scale","created":"2026-06-08","cited_by_count":4,"area":"inference"},{"id":"2607.16673","title":"SpecLA: Efficient Speculative Decoding for Linear-Attention Models","created":"2026-07-18","cited_by_count":4,"area":"inference"},{"id":"2502.02493","title":"EasySpec: Layer-Parallel Speculative Decoding for Efficient Multi-GPU Utilization","created":"2025-02-04","cited_by_count":3,"area":"inference"},{"id":"2504.03661","title":"MILLION: Mastering Long-Context LLM Inference Via Outlier-Immunized KV Product Quantization","created":"2025-03-12","cited_by_count":3,"area":"inference"},{"id":"2510.20787","title":"Alleviating Forgetfulness of Linear Attention by Hybrid Sparse Attention and Contextualized Learnable Token Eviction","created":"2025-10-23","cited_by_count":3,"area":"inference"},{"id":"2601.05149","title":"Multi-Scale Local Speculative Decoding for Image Generation","created":"2026-01-08","cited_by_count":3,"area":"inference"},{"id":"2603.05931","title":"A Persistent-State Dataflow Accelerator for Memory-Bound Linear Attention Decode on FPGA","created":"2026-03-06","cited_by_count":3,"area":"inference"},{"id":"2605.09329","title":"Test-Time Speculation","created":"2026-05-10","cited_by_count":3,"area":"inference"},{"id":"2605.19049","title":"KVBuffer: IO-aware Serving for Linear Attention","created":"2026-05-18","cited_by_count":3,"area":"inference"},{"id":"2504.19191","title":"WuNeng: Hybrid State with Attention","created":"2025-04-27","cited_by_count":2,"area":"inference"},{"id":"2507.09768","title":"Knowing When to Quit: Probabilistic Early Exits for Speech Separation","created":"2025-07-13","cited_by_count":2,"area":"inference"},{"id":"2511.14806","title":"MergeDNA: Context-aware Genome Modeling with Dynamic Tokenization through Token Merging","created":"2025-11-17","cited_by_count":2,"area":"inference"},{"id":"2602.18196","title":"RAT+: Train Dense, Infer Sparse -- Recurrence Augmented Attention for Dilated Inference","created":"2026-02-20","cited_by_count":2,"area":"inference"},{"id":"2603.02188","title":"Multi-Head Low-Rank Attention","created":"2026-03-02","cited_by_count":2,"area":"inference"},{"id":"2605.06221","title":"UniPrefill: Universal Long-Context Prefill Acceleration via Block-wise Dynamic Sparsification","created":"2026-05-07","cited_by_count":2,"area":"inference"},{"id":"2605.13784","title":"Attention Once Is All You Need: Efficient Streaming Inference with Stateful Transformers","created":"2026-05-13","cited_by_count":2,"area":"inference"},{"id":"2605.22884","title":"Tensor Cache: Eviction-conditioned Associative Memory for Transformers","created":"2026-05-21","cited_by_count":2,"area":"inference"},{"id":"2606.02775","title":"AURA: Action-Gated Memory for Robot Policies at Constant VRAM","created":"2026-06-01","cited_by_count":2,"area":"inference"},{"id":"2606.06467","title":"You Only Index Once: Cross-Layer Sparse Attention with Shared Routing","created":"2026-06-04","cited_by_count":2,"area":"inference"},{"id":"2608.28444","title":"Sliding-window beats linear attention","created":"2026-08-28","cited_by_count":2,"area":"inference"},{"id":"2609.04434","title":"What Attention Recalls and Recurrence Controls in Hybrid Language Models","created":"2026-09-03","cited_by_count":2,"area":"inference"},{"id":"2610.13778","title":"ReplaySSM: Cache SSM Inputs, Not State","created":"2026-10-09","cited_by_count":2,"area":"inference"},{"id":"2508.00370","title":"EdgeInfinite-Instruct: Bridging SFT-Based Optimization and NPU-Level Efficiency for Edge Devices","created":"2025-08-01","cited_by_count":1,"area":"inference"},{"id":"2512.09238","title":"Training-free Context-adaptive Attention for Efficient Long Context Modeling","created":"2025-12-10","cited_by_count":1,"area":"inference"},{"id":"2602.06471","title":"Revisiting the Shape Convention of Transformer Language Models","created":"2026-02-06","cited_by_count":1,"area":"inference"},{"id":"2603.06642","title":"SR-TTT Does Not Learn Retrieval: A Correction and Mechanistic Post-Mortem of Surprisal-Aware Residual Test-Time Training","created":"2026-02-26","cited_by_count":1,"area":"inference"},{"id":"2603.10353","title":"S-HPLB: Efficient LLM Attention Serving via Sparsity-Aware Head Parallelism Load Balance","created":"2026-03-11","cited_by_count":1,"area":"inference"},{"id":"2603.23701","title":"The Diminishing Returns of Early-Exit Decoding in Modern LLMs","created":"2026-03-24","cited_by_count":1,"area":"inference"},{"id":"2604.08426","title":"KV Cache Offloading for Context-Intensive Tasks","created":"2026-04-09","cited_by_count":1,"area":"inference"},{"id":"2604.25570","title":"Vision SmolMamba: Spike-Guided Token Pruning for Energy-Efficient Spiking State-Space Vision Models","created":"2026-04-28","cited_by_count":1,"area":"inference"},{"id":"2605.09630","title":"Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models","created":"2026-05-10","cited_by_count":1,"area":"inference"},{"id":"2606.09862","title":"Blurry Window Attention","created":"2026-05-31","cited_by_count":1,"area":"inference"},{"id":"2606.16429","title":"Taylor-Calibrate: Principled Initialization for Hybrid Linear Attention Distillation","created":"2026-06-15","cited_by_count":1,"area":"inference"},{"id":"2608.15533","title":"DeltaLog: Deferred Materialization of Recurrent States for Linear Attention Decoding","created":"2026-08-16","cited_by_count":1,"area":"inference"},{"id":"2505.08600","title":"Automatic Task Detection and Heterogeneous LLM Speculative Decoding","created":"2025-05-13","cited_by_count":0,"area":"inference"},{"id":"2505.20698","title":"Sparsified State-Space Models are Efficient Highway Networks","created":"2025-05-27","cited_by_count":0,"area":"inference"},{"id":"2507.09514","title":"QuarterMap: Efficient Post-Training Token Pruning for Visual State Space Models","created":"2025-07-13","cited_by_count":0,"area":"inference"},{"id":"2507.19409","title":"Modality Agnostic Efficient Long Range Encoder","created":"2025-07-25","cited_by_count":0,"area":"inference"},{"id":"2508.13599","title":"Towards Efficient Vision State Space Models via Token Merging","created":"2025-08-19","cited_by_count":0,"area":"inference"},{"id":"2509.03377","title":"TRACE: Unlocking Effective CXL Bandwidth via Lossless Compression and Precision Scaling","created":"2025-09-03","cited_by_count":0,"area":"inference"},{"id":"2604.07716","title":"Breaking the KV Cache Bottleneck: Fan Duality Model Achieves O(1) Decode Memory with Superior Associative Recall","created":"2026-04-09","cited_by_count":0,"area":"inference"},{"id":"2605.12471","title":"KV-Fold: One-Step KV-Cache Recurrence for Long-Context Inference","created":"2026-05-12","cited_by_count":0,"area":"inference"},{"id":"2605.20936","title":"DASH: Fast Differentiable Architecture Search for Hybrid Attention in Minutes on a Single GPU","created":"2026-05-20","cited_by_count":0,"area":"inference"},{"id":"2605.26558","title":"Cassandra: Enabling Reasoning LLMs at Edge via Self-Speculative Decoding","created":"2026-05-26","cited_by_count":0,"area":"inference"},{"id":"2606.18056","title":"ConSA: Controllable Sparsity in Hybrid Attention via Learnable Allocation","created":"2026-06-16","cited_by_count":0,"area":"inference"},{"id":"2607.19368","title":"Spectral-LSH: Sub-Quadratic Prompt Compression via Krylov-Projected Locality-Sensitive Hashing","created":"2026-06-12","cited_by_count":0,"area":"inference"},{"id":"2607.21535","title":"Windowed-MTP: Removing the Full-Context Draft-KV Tax at Million-Token Context","created":"2026-07-23","cited_by_count":0,"area":"inference"},{"id":"2608.19395","title":"HYDRA: A Heterogeneous Chiplet DSE Framework for Serving Dynamic Hybrid LLM Workloads","created":"2026-08-19","cited_by_count":0,"area":"inference"},{"id":"2608.30439","title":"Event-Driven Language Models with Sparse Neural Activity for Neuromorphic Hardware","created":"2026-08-31","cited_by_count":0,"area":"inference"}],"next_cursor":"100"}