# ReaLB: Real-Time Load Balancing for Multimodal MoE Inference

arXiv: https://arxiv.org/abs/2604.19503 · HTML: https://mlsanity.com/paper/2604.19503/

## Key facts

- Cited by 1 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 18 of its 40 references that are in the map.
- First posted on arXiv 21 Apr 2026.

## 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, 18 of 18)

- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI](https://mlsanity.com/paper/2311.16502/) (2023)
- [Towards VQA Models That Can Read](https://mlsanity.com/paper/1904.08920/) (2019)
- [MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts](https://mlsanity.com/paper/2310.02255/) (2023)
- [DocVQA: A Dataset for VQA on Document Images](https://mlsanity.com/paper/2007.00398/) (2020)
- [SGLang: Efficient Execution of Structured Language Model Programs](https://mlsanity.com/paper/2312.07104/) (2023)
- [DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model](https://mlsanity.com/paper/2405.04434/) (2024)
- [CogVLM: Visual Expert for Pretrained Language Models](https://mlsanity.com/paper/2311.03079/) (2023)
- [An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models](https://mlsanity.com/paper/2403.06764/) (2024)
- [DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding](https://mlsanity.com/paper/2412.10302/) (2024)
- [PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reduction](https://mlsanity.com/paper/2410.17247/) (2024)
- [FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving](https://mlsanity.com/paper/2501.01005/) (2025)
- [DynaMath: A Dynamic Visual Benchmark for Evaluating Mathematical Reasoning Robustness of Vision Language Models](https://mlsanity.com/paper/2411.00836/) (2024)
- [MoMa: Efficient Early-Fusion Pre-training with Mixture of Modality-Aware Experts](https://mlsanity.com/paper/2407.21770/) (2024)
- [FlexMoE: Scaling Large-scale Sparse Pre-trained Model Training via Dynamic Device Placement](https://mlsanity.com/paper/2304.03946/) (2023)
- [Towards MoE Deployment: Mitigating Inefficiencies in Mixture-of-Expert (MoE) Inference](https://mlsanity.com/paper/2303.06182/) (2023)
- [Shortcut-connected Expert Parallelism for Accelerating Mixture-of-Experts](https://mlsanity.com/paper/2404.05019/) (2024)
- [MegaScale-MoE: Large-Scale Communication-Efficient Training of Mixture-of-Experts Models in Production](https://mlsanity.com/paper/2505.11432/) (2025)

## Built on by (most cited first, 1 of 1)

- [X-Stage: Modeling Post-Issue Backpressure in GPU Communication--Computation Fusion](https://mlsanity.com/paper/2607.23264/) (2026)

## Abstract (arXiv)

Mixture-of-Experts (MoE) architectures are widely used in modern large language models and multimodal models. However, inference efficiency is often limited by highly dynamic and skewed expert workloads across different modalities. During the prefill stage with large batch sizes, vision tokens frequently dominate the input sequences. Under expert parallelism (EP), this leads to severe load imbalance, where a subset of devices becomes overloaded, reducing overall system throughput. We propose ReaLB, a real-time load balancing method for multimodal MoE (MMoE) inference that introduces zero scheduling overhead. ReaLB dynamically adjusts the computation precision of MoE experts at runtime on a per-EP-rank basis. For ranks dominated by vision-heavy experts, ReaLB assigns lower-precision computation to improve execution efficiency by exploiting FP4 Tensor Cores. ReaLB does not require redundant experts or additional memory allocation. Instead, it performs layer-wise expert precision transformation on the fly and hides the associated overhead within the dispatch phase before MoE computation. Experiments on representative MMoE models show that ReaLB achieves 1.10×-1.32× end-to-end speedup while limiting average accuracy degradation to within 1%.