# Elastic MoE: Unlocking the Inference-Time Scalability of Mixture-of-Experts

arXiv: https://arxiv.org/abs/2509.21892 · HTML: https://mlsanity.com/paper/2509.21892/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Placed in Efficient MoE training and inference systems, one of 12 directions in Mixture-of-experts.
- Builds on 28 of its 40 references that are in the map.
- First posted on arXiv 26 Sep 2025.

## Placed in

- Mixture-of-experts / [Efficient MoE training and inference systems](https://mlsanity.com/moe/efficient-moe-training-and-inference-systems/)

## Builds on (most cited first, 28 of 28)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension](https://mlsanity.com/paper/1705.03551/) (2017)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding](https://mlsanity.com/paper/2006.16668/) (2020)
- [DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models](https://mlsanity.com/paper/2401.06066/) (2024)
- [MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models](https://mlsanity.com/paper/2309.12284/) (2023)
- [From Sparse to Soft Mixtures of Experts](https://mlsanity.com/paper/2308.00951/) (2023)
- [Auxiliary-Loss-Free Load Balancing Strategy for Mixture-of-Experts](https://mlsanity.com/paper/2408.15664/) (2024)
- [Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation](https://mlsanity.com/paper/2507.10524/) (2025)
- [MatFormer: Nested Transformer for Elastic Inference](https://mlsanity.com/paper/2310.07707/) (2023)
- [LoRAMoE: Alleviate World Knowledge Forgetting in Large Language Models via MoE-Style Plugin](https://mlsanity.com/paper/2312.09979/) (2023)
- [Inner Thinking Transformer: Leveraging Dynamic Depth Scaling to Foster Adaptive Internal Thinking](https://mlsanity.com/paper/2502.13842/) (2025)
- [MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts](https://mlsanity.com/paper/2410.07348/) (2024)
- [ReMoE: Fully Differentiable Mixture-of-Experts with ReLU Routing](https://mlsanity.com/paper/2412.14711/) (2024)
- [AdaMoE: Token-Adaptive Routing with Null Experts for Mixture-of-Experts Language Models](https://mlsanity.com/paper/2406.13233/) (2024)
- [HMoE: Heterogeneous Mixture of Experts for Language Modeling](https://mlsanity.com/paper/2408.10681/) (2024)
- [HydraViT: Stacking Heads for a Scalable ViT](https://mlsanity.com/paper/2409.17978/) (2024)
- [Upcycling Instruction Tuning from Dense to Mixture-of-Experts via Parameter Merging](https://mlsanity.com/paper/2410.01610/) (2024)

## Built on by (most cited first, 11 of 11)

- [ERNIE 5.0 Technical Report](https://mlsanity.com/paper/2602.04705/) (2026)
- [Mixture of Universal Experts: Scaling Virtual Width via Depth-Width Transformation](https://mlsanity.com/paper/2603.04971/) (2026)
- [Grouter: Decoupling Routing from Representation for Accelerated MoE Training](https://mlsanity.com/paper/2603.06626/) (2026)
- [MoSE: Mixture of Slimmable Experts for Efficient and Adaptive Language Models](https://mlsanity.com/paper/2602.06154/) (2026)
- [Foundry: Template-Based CUDA Graph Context Materialization for Fast LLM Serving Cold Start](https://mlsanity.com/paper/2604.06664/) (2026)
- [MoE-Prefill: Zero Redundancy Overheads in MoE Prefill Serving](https://mlsanity.com/paper/2605.02960/) (2026)
- [MoE-Prism: Disentangling Monolithic Experts for Elastic MoE Services via Model-System Co-Designs](https://mlsanity.com/paper/2510.19366/) (2025)
- [Mixture-of-Experts in Remote Sensing: A Survey](https://mlsanity.com/paper/2604.03342/) (2026)
- [Do Vision Transformers Need All-to-All Attention? Global Communication Through Elastic Learned Cores](https://mlsanity.com/paper/2605.12491/) (2026)
- [Training-Free Halving of Activated Experts in Fine-Grained Mixture-of-Experts Models](https://mlsanity.com/paper/2609.04575/) (2026)
- [One Model, Multiple Compute Budgets: Multi-Ratio ConceptMoE for Flexible Compute Across Turns](https://mlsanity.com/paper/2610.14074/) (2026)

## Abstract (arXiv)

Mixture-of-Experts (MoE) models typically fix the number of activated experts k at both training and inference. However, real-world deployments often face heterogeneous hardware, fluctuating workloads, and diverse quality-latency requirements, while training separate models for each scenario is costly. Considering that MoE models already operate with sparse activation, adjusting the number of activated experts offers a natural path to serving diverse budgets with a single model. Yet, we find that activating more experts k' (> k) at inference does not yield the expected gains. Instead, performance degrades rapidly after only a slight increase, a phenomenon we term the textitinference-time scaling wall. Further investigation reveals that this degradation stems from a lack of learned collaboration among experts. To address this, we introduce Elastic Mixture-of-Experts (EMoE), a novel training framework that enables MoE models to elastically vary the number of activated experts at inference. By simultaneously training experts to collaborate in diverse combinations and encouraging the router to make high-quality selections, EMoE ensures robust performance across inference budgets. Extensive experiments across four MoE architectures (7B--21B) and nine benchmarks show that EMoE significantly expands the effective scaling range to 2-3× the training-time k, while also achieving higher peak performance.