# Early exit and edge inference

Area: [Efficient LLM inference and serving](https://mlsanity.com/inference/index.md). HTML: https://mlsanity.com/inference/early-exit-and-edge-inference/

## Key facts

- Early exit and edge inference: 215 papers on arXiv since 2021 (data through 14 Oct 2026).
- 29 of its 215 mapped papers (13%) cite “Confident Adaptive Language Modeling” (2022, arXiv 2207.07061).
- 29 of its 215 mapped papers (13%) cite “LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding” (2024, arXiv 2404.16710).
- 7 new papers 24 Sep–14 Oct.

## Most cited here

- [Confident Adaptive Language Modeling](https://mlsanity.com/paper/2207.07061/) (2022), cited by 29 papers here
- [LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding](https://mlsanity.com/paper/2404.16710/) (2024), cited by 29 papers here
- [Adaptive Inference through Early-Exit Networks: Design, Challenges and Directions](https://mlsanity.com/paper/2106.05022/) (2021), cited by 17 papers here
- [Split Computing and Early Exiting for Deep Learning Applications: Survey and Research Challenges](https://mlsanity.com/paper/2103.04505/) (2021), cited by 16 papers here
- [EE-LLM: Large-Scale Training and Inference of Early-Exit Large Language Models with 3D Parallelism](https://mlsanity.com/paper/2312.04916/) (2023), cited by 12 papers here
- [Zero Time Waste: Recycling Predictions in Early Exit Neural Networks](https://mlsanity.com/paper/2106.05409/) (2021), cited by 11 papers here
- [You Need Multiple Exiting: Dynamic Early Exiting for Accelerating Unified Vision Language Model](https://mlsanity.com/paper/2211.11152/) (2022), cited by 11 papers here
- [Accelerating BERT Inference for Sequence Labeling via Early-Exit](https://mlsanity.com/paper/2105.13878/) (2021), cited by 10 papers here
- [A Simple Hash-Based Early Exiting Approach For Language Understanding and Generation](https://mlsanity.com/paper/2203.01670/) (2022), cited by 10 papers here
- [EdgeShard: Efficient LLM Inference via Collaborative Edge Computing](https://mlsanity.com/paper/2405.14371/) (2024), cited by 10 papers here

## New (24 Sep–14 Oct)

- [Selection-Safe Conformal Early Exit: Distribution-Free Coverage under Adaptive Depth](https://mlsanity.com/paper/2610.20773/) (2026-10-14)
- [SparseDecoding: Decoding-Aware Pruning for Accurate and Efficient LLM Inference](https://mlsanity.com/paper/2610.12327/) (2026-10-08)
- [TopK-Guided: Adaptive, Budget-Aware Activation Sparsity for Efficient LLM Inference](https://mlsanity.com/paper/2610.01763/) (2026-10-01)
- [Sol-H3: Recursive Self-Improvement for MiniMax-H3 Inference Acceleration on Sol-Engine across Cloud and Edge](https://mlsanity.com/paper/2609.35110/) (2026-09-28)
- [SlimWise: Decoupling Expert Pruning Across Prefill and Decode for Efficient MoE Serving](https://mlsanity.com/paper/2609.34117/) (2026-09-28)
- [When Less Compute Is More: Adaptive Early Exit Improves Pretrained Outlier Detection](https://mlsanity.com/paper/2609.32898/) (2026-09-26)
- [Decoupled Early Exits for Task-Dependent Compute Allocation in Flow-Matching VLAs](https://mlsanity.com/paper/2609.29382/) (2026-09-24)