{"total":126,"rows":[{"id":"2211.17192","title":"Fast Inference from Transformers via Speculative Decoding","created":"2022-11-30","cited_by_count":1653,"area":"inference"},{"id":"2210.03350","title":"Measuring and Narrowing the Compositionality Gap in Language Models","created":"2022-10-07","cited_by_count":1075,"area":"lr"},{"id":"2203.14680","title":"Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space","created":"2022-03-28","cited_by_count":532,"area":null},{"id":"2102.04906","title":"Dynamic Neural Networks: A Survey","created":"2021-02-09","cited_by_count":408,"area":null},{"id":"2112.07658","title":"AdaViT: Adaptive Tokens for Efficient Vision Transformer","created":"2021-12-14","cited_by_count":362,"area":null},{"id":"2207.07061","title":"Confident Adaptive Language Modeling","created":"2022-07-14","cited_by_count":310,"area":"inference"},{"id":"2310.02226","title":"Think before you speak: Training Language Models With Pause Tokens","created":"2023-10-03","cited_by_count":264,"area":"lr"},{"id":"2402.16363","title":"LLM Inference Unveiled: Survey and Roofline Model Insights","created":"2024-02-26","cited_by_count":152,"area":"inference"},{"id":"2106.09647","title":"Deep Learning Through the Lens of Example Difficulty","created":"2021-06-17","cited_by_count":141,"area":null},{"id":"2410.20290","title":"Fast Best-of-N Decoding via Speculative Rejection","created":"2024-10-26","cited_by_count":112,"area":null},{"id":"2406.15627","title":"Benchmarking Uncertainty Quantification Methods for Large Language Models with LM-Polygraph","created":"2024-06-21","cited_by_count":105,"area":null},{"id":"2303.09435","title":"Jump to Conclusions: Short-Cutting Transformers With Linear Transformations","created":"2023-03-16","cited_by_count":90,"area":null},{"id":"2209.00099","title":"Efficient Methods for Natural Language Processing: A Survey","created":"2022-08-31","cited_by_count":78,"area":null},{"id":"2106.05022","title":"Adaptive Inference through Early-Exit Networks: Design, Challenges and Directions","created":"2021-06-09","cited_by_count":75,"area":"inference"},{"id":"2104.08803","title":"Consistent Accelerated Inference via Confident Adaptive Transformers","created":"2021-04-18","cited_by_count":74,"area":null},{"id":"2403.07652","title":"Harder Tasks Need More Experts: Dynamic Routing in MoE Models","created":"2024-03-12","cited_by_count":72,"area":"moe"},{"id":"2308.11696","title":"Efficient Benchmarking of Language Models","created":"2023-08-22","cited_by_count":71,"area":null},{"id":"2312.04916","title":"EE-LLM: Large-Scale Training and Inference of Early-Exit Large Language Models with 3D Parallelism","created":"2023-12-08","cited_by_count":69,"area":"inference"},{"id":"2307.02764","title":"When Does Confidence-Based Cascade Deferral Suffice?","created":"2023-07-06","cited_by_count":60,"area":null},{"id":"2210.07795","title":"EfficientVLM: Fast and Accurate Vision-Language Models via Knowledge Distillation and Modal-adaptive Pruning","created":"2022-10-14","cited_by_count":56,"area":"mllm"},{"id":"2310.11877","title":"The Curious Case of Hallucinatory (Un)answerability: Finding Truths in the Hidden States of Over-Confident Large Language Models","created":"2023-10-18","cited_by_count":55,"area":null},{"id":"2305.17530","title":"PuMer: Pruning and Merging Tokens for Efficient Vision Language Models","created":"2023-05-27","cited_by_count":53,"area":"mllm"},{"id":"2405.01976","title":"Conformal Prediction for Natural Language Processing: A Survey","created":"2024-05-03","cited_by_count":53,"area":null},{"id":"2111.05193","title":"A Survey on Green Deep Learning","created":"2021-11-08","cited_by_count":52,"area":null},{"id":"2210.12714","title":"Generative Knowledge Graph Construction: A Review","created":"2022-10-23","cited_by_count":49,"area":null},{"id":"2102.01386","title":"AutoFreeze: Automatically Freezing Model Blocks to Accelerate Fine-tuning","created":"2021-02-02","cited_by_count":48,"area":null},{"id":"2402.05964","title":"A Survey on Transformer Compression","created":"2024-02-05","cited_by_count":46,"area":null},{"id":"2202.07105","title":"A Survey on Model Compression and Acceleration for Pretrained Language Models","created":"2022-02-15","cited_by_count":45,"area":null},{"id":"2110.07038","title":"Towards Efficient NLP: A Standard Evaluation and A Strong Baseline","created":"2021-10-13","cited_by_count":44,"area":null},{"id":"2306.04459","title":"Uncertainty in Natural Language Processing: Sources, Quantification, and Applications","created":"2023-06-05","cited_by_count":44,"area":null},{"id":"2406.12094","title":"Who's asking? User personas and the mechanics of latent misalignment","created":"2024-06-17","cited_by_count":40,"area":"safety"},{"id":"2109.03228","title":"Beyond Preserved Accuracy: Evaluating Loyalty and Robustness of BERT Compression","created":"2021-09-07","cited_by_count":38,"area":null},{"id":"2211.09961","title":"Path Independent Equilibrium Models Can Better Exploit Test-Time Computation","created":"2022-11-18","cited_by_count":38,"area":null},{"id":"2203.01670","title":"A Simple Hash-Based Early Exiting Approach For Language Understanding and Generation","created":"2022-03-03","cited_by_count":32,"area":"inference"},{"id":"2312.09299","title":"Weight subcloning: direct initialization of transformers using larger pretrained ones","created":"2023-12-14","cited_by_count":30,"area":null},{"id":"2105.13878","title":"Accelerating BERT Inference for Sequence Labeling via Early-Exit","created":"2021-05-28","cited_by_count":29,"area":"inference"},{"id":"2305.01628","title":"The Benefits of Bad Advice: Autocontrastive Decoding across Model Layers","created":"2023-05-02","cited_by_count":29,"area":null},{"id":"2409.03621","title":"Attend First, Consolidate Later: On the Importance of Attention in Different LLM Layers","created":"2024-09-05","cited_by_count":25,"area":null},{"id":"2411.02829","title":"CE-CoLLM: Efficient and Adaptive Large Language Models Through Cloud-Edge Collaboration","created":"2024-11-05","cited_by_count":25,"area":null},{"id":"2109.02071","title":"End-to-End Self-Debiasing Framework for Robust NLU Training","created":"2021-09-05","cited_by_count":24,"area":null},{"id":"2202.07101","title":"A Survey on Dynamic Neural Networks for Natural Language Processing","created":"2022-02-15","cited_by_count":24,"area":null},{"id":"2109.04144","title":"Avoiding Inference Heuristics in Few-shot Prompt-based Finetuning","created":"2021-09-09","cited_by_count":23,"area":null},{"id":"2301.13195","title":"Adaptive Computation with Elastic Input Sequence","created":"2023-01-30","cited_by_count":22,"area":null},{"id":"2310.03686","title":"DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformers","created":"2023-10-05","cited_by_count":22,"area":null},{"id":"2402.04513","title":"Online Cascade Learning for Efficient Inference over Streams","created":"2024-02-07","cited_by_count":20,"area":null},{"id":"2105.13792","title":"Early Exiting with Ensemble Internal Classifiers","created":"2021-05-28","cited_by_count":19,"area":null},{"id":"2203.08991","title":"AdapLeR: Speeding up Inference by Adaptive Length Reduction","created":"2022-03-16","cited_by_count":19,"area":null},{"id":"2302.06359","title":"Fixing Overconfidence in Dynamic Neural Networks","created":"2023-02-13","cited_by_count":19,"area":null},{"id":"2308.15452","title":"When Do Program-of-Thoughts Work for Reasoning?","created":"2023-08-29","cited_by_count":19,"area":"lr"},{"id":"2311.15436","title":"Learning to Skip for Language Modeling","created":"2023-11-26","cited_by_count":19,"area":null},{"id":"2305.11170","title":"Efficient Prompting via Dynamic In-Context Learning","created":"2023-05-18","cited_by_count":18,"area":null},{"id":"2305.11916","title":"F-PABEE: Flexible-patience-based Early Exiting for Single-label and Multi-label text Classification Tasks","created":"2023-05-21","cited_by_count":18,"area":null},{"id":"2111.00230","title":"Magic Pyramid: Accelerating Inference with Early Exiting and Token Pruning","created":"2021-10-30","cited_by_count":16,"area":"inference"},{"id":"2306.02652","title":"Towards Anytime Classification in Early-Exit Architectures by Enforcing Conditional Monotonicity","created":"2023-06-05","cited_by_count":16,"area":"inference"},{"id":"2604.10739","title":"When More Thinking Hurts: Overthinking in LLM Test-Time Compute Scaling","created":"2026-04-12","cited_by_count":16,"area":"lr"},{"id":"2105.13471","title":"Inspecting the concept knowledge graph encoded by modern language models","created":"2021-05-27","cited_by_count":15,"area":null},{"id":"2312.05385","title":"Apparate: Rethinking Early Exits to Tame Latency-Throughput Tensions in ML Serving","created":"2023-12-08","cited_by_count":15,"area":"inference"},{"id":"2406.00060","title":"Cascade-Aware Training of Language Models","created":"2024-05-29","cited_by_count":15,"area":null},{"id":"2107.00175","title":"Elbert: Fast Albert with Confidence-Window Based Early Exit","created":"2021-07-01","cited_by_count":14,"area":"inference"},{"id":"2101.00416","title":"Improving Sequence-to-Sequence Pre-training via Sequence Span Rewriting","created":"2021-01-02","cited_by_count":13,"area":null},{"id":"2204.06271","title":"TangoBERT: Reducing Inference Cost by using Cascaded Architecture","created":"2022-04-13","cited_by_count":13,"area":null},{"id":"2204.06328","title":"HuBERT-EE: Early Exiting HuBERT for Efficient Speech Recognition","created":"2022-04-13","cited_by_count":13,"area":"inference"},{"id":"2205.01549","title":"Adaptable Adapters","created":"2022-05-03","cited_by_count":13,"area":null},{"id":"2312.11882","title":"ConsistentEE: A Consistent and Hardness-Guided Early Exiting Method for Accelerating Language Models Inference","created":"2023-12-19","cited_by_count":13,"area":"inference"},{"id":"2402.00518","title":"EE-Tuning: An Economical yet Scalable Solution for Tuning Early-Exit Large Language Models","created":"2024-02-01","cited_by_count":13,"area":"inference"},{"id":"2603.08251","title":"Not All Queries Need Deep Thought: CoFiCot for Adaptive Coarse-to-fine Stateful Refinement","created":"2026-03-09","cited_by_count":13,"area":"lr"},{"id":"2405.15039","title":"CEEBERT: Cross-Domain Inference in Early Exit BERT","created":"2024-05-23","cited_by_count":12,"area":"inference"},{"id":"2410.20210","title":"Looking Beyond The Top-1: Transformers Determine Top Tokens In Order","created":"2024-10-26","cited_by_count":12,"area":null},{"id":"2205.15237","title":"VLUE: A Multi-Task Benchmark for Evaluating Vision-Language Models","created":"2022-05-30","cited_by_count":11,"area":"mllm"},{"id":"2310.13682","title":"Optimizing Retrieval-augmented Reader Models via Token Elimination","created":"2023-10-20","cited_by_count":10,"area":"rag"},{"id":"2603.01914","title":"AdaPonderLM: Gated Pondering Language Models with Token-Wise Adaptive Depth","created":"2026-03-02","cited_by_count":10,"area":"lr"},{"id":"2303.09266","title":"SmartBERT: A Promotion of Dynamic Early Exiting Mechanism for Accelerating BERT Inference","created":"2023-03-16","cited_by_count":9,"area":"inference"},{"id":"2110.10305","title":"When in Doubt, Summon the Titans: Efficient Inference with Large Models","created":"2021-10-19","cited_by_count":8,"area":null},{"id":"2311.08623","title":"DEED: Dynamic Early Exit on Decoder for Accelerating Encoder-Decoder Transformer Models","created":"2023-11-15","cited_by_count":7,"area":"inference"},{"id":"2403.14919","title":"Hierarchical Skip Decoding for Efficient Autoregressive Text Generation","created":"2024-03-22","cited_by_count":7,"area":"inference"},{"id":"2109.10060","title":"DS-Net++: Dynamic Weight Slicing for Efficient Inference in CNNs and Transformers","created":"2021-09-21","cited_by_count":6,"area":null},{"id":"2310.12126","title":"SHARCS: Efficient Transformers through Routing with Dynamic Width Sub-networks","created":"2023-10-18","cited_by_count":6,"area":null},{"id":"2311.05931","title":"Early-Exit Neural Networks with Nested Prediction Sets","created":"2023-11-10","cited_by_count":6,"area":null},{"id":"2502.00745","title":"BEEM: Boosting Performance of Early Exit DNNs using Multi-Exit Classifiers as Experts","created":"2025-02-02","cited_by_count":6,"area":"inference"},{"id":"2508.10175","title":"Estimating Machine Translation Difficulty","created":"2025-08-13","cited_by_count":6,"area":null},{"id":"2103.03457","title":"IOT: Instance-wise Layer Reordering for Transformer Structures","created":"2021-03-05","cited_by_count":5,"area":null},{"id":"2306.02307","title":"Finding the SWEET Spot: Analysis and Improvement of Adaptive Inference in Low Resource Settings","created":"2023-06-04","cited_by_count":5,"area":"inference"},{"id":"2407.20272","title":"An Efficient Inference Framework for Early-exit Large Language Models","created":"2024-07-25","cited_by_count":5,"area":"inference"},{"id":"2107.02102","title":"Training Adaptive Computation for Open-Domain Question Answering with Computational Constraints","created":"2021-07-05","cited_by_count":4,"area":null},{"id":"2111.09645","title":"Dynamic-TinyBERT: Boost TinyBERT's Inference Efficiency by Dynamic Sequence Length","created":"2021-11-18","cited_by_count":4,"area":null},{"id":"2301.02989","title":"Fair Multi-Exit Framework for Facial Attribute Classification","created":"2023-01-08","cited_by_count":4,"area":null},{"id":"2306.02379","title":"Modular Transformers: Compressing Transformers into Modularized Layers for Flexible Efficient Inference","created":"2023-06-04","cited_by_count":4,"area":"distill"},{"id":"2310.19152","title":"BERT Lost Patience Won't Be Robust to Adversarial Slowdown","created":"2023-10-29","cited_by_count":4,"area":null},{"id":"2503.20436","title":"Siformer: Feature-isolated Transformer for Efficient Skeleton-based Sign Language Recognition","created":"2025-03-26","cited_by_count":4,"area":null},{"id":"2204.02601","title":"Probing Structured Pruning on Multilingual Pre-trained Models: Settings, Algorithms, and Efficiency","created":"2022-04-06","cited_by_count":3,"area":null},{"id":"2305.15389","title":"Comparing Humans and Models on a Similar Scale: Towards Cognitive Gender Bias Evaluation in Coreference Resolution","created":"2023-05-24","cited_by_count":3,"area":null},{"id":"2305.19844","title":"Learning Task-preferred Inference Routes for Gradient De-conflict in Multi-output DNNs","created":"2023-05-31","cited_by_count":3,"area":null},{"id":"2308.03421","title":"RecycleGPT: An Autoregressive Language Model with Recyclable Module","created":"2023-08-07","cited_by_count":3,"area":null},{"id":"2402.05948","title":"DE$^3$-BERT: Distance-Enhanced Early Exiting for BERT based on Prototypical Networks","created":"2024-02-03","cited_by_count":3,"area":"inference"},{"id":"2509.23666","title":"Beyond Greedy Exits: Improved Early Exit Decisions for Risk Control and Reliability","created":"2025-09-28","cited_by_count":3,"area":"inference"},{"id":"2302.01440","title":"Generalized Uncertainty of Deep Neural Networks: Taxonomy and Applications","created":"2023-02-02","cited_by_count":2,"area":null},{"id":"2402.07052","title":"Understanding the Training Speedup from Sampling with Approximate Losses","created":"2024-02-10","cited_by_count":2,"area":null},{"id":"2403.13835","title":"SMART: Automatically Scaling Down Language Models with Accuracy Guarantees for Reduced Processing Fees","created":"2024-03-11","cited_by_count":2,"area":null},{"id":"2405.15208","title":"Decoding at the Speed of Thought: Harnessing Parallel Decoding of Lexical Units for LLMs","created":"2024-05-24","cited_by_count":2,"area":"lr"},{"id":"2506.17249","title":"Improving Prediction Certainty Estimation for Reliable Early Exiting via Null Space Projection","created":"2025-06-08","cited_by_count":2,"area":"inference"}],"next_cursor":"100"}