Mixture-of-Experts Language Models

Overview

Mixture-of-Experts language models divide parts of a transformer into expert networks and use a router to select experts for each token, so a model can have many parameters without activating them all at once. Research explores top-k and other routing rules, load-balancing methods, and training strategies such as pretraining sparse models or converting dense models into experts; evaluations commonly include language-model loss and reasoning, coding, and multilingual benchmarks. A large body of work also aims to make these models easier to fine-tune, compress, interpret, and serve through pruning, quantization, expert sharing, and routing-aware systems. A central tension is how to encourage useful expert specialization while keeping routing reliable and balanced, since communication and memory costs can limit the practical gains from sparse computation.

Latest papers

248 papers23 papers in the last four weeks, down 39% on the four weeks before. 0.2% of all new papers.

Jan 20, 2026

ButterflyMoE: Compression-Scalable Ternary Experts via Structured Butterfly Orbits

In current Mixture of Experts (MoE) architectures, linear memory scaling is present, the memory grows as the number of experts increases. NN independent expert weight matrices require O(N⋅d2)\mathcal{O}(N \cdot d^2) memory which exceeds the memory budget of edge devices. Current compression methods like quantization, pruning, and low-rank factorization reduce constant factors, but the scaling bottleneck is still unresolved. We introduce ButterflyMoE, a method which treats experts not as independent matrices but as geometric reorientations of a shared quantized substrate. Diversity amongst the experts arises from viewing different angles of the shared capacity and not from redundant storage. Learned rotations are applied to a shared ternary prototype. With this, each expert yields O(d2+N⋅dlog⁡d)\mathcal{O}(d^2 + N \cdot d \log d) memory-reducing per-expert cost from O(d2)\mathcal{O}(d^2) to O(dlog⁡d)\mathcal{O}(d \log d). The key insight is that training these rotations with quantization reduces activation outliers and stabilizes extreme low-bit training where other static methods collapse. Across language modeling benchmarks, ButterflyMoE achieves 80×\times memory reduction at 8 experts with a highly favorable memory-accuracy tradeoff.At this 80x compression ButterflyMoE outperforms an equal memory dense baseline, showing that orbital parameterization extracts fundamentally more utility per byte. When scaled up to 256 experts, ButterflyMoE asymptotically compresses the memory by 150 ×\times. ButterflyMoE reduces the constant factor of linear scaling with compression ratio growing with the expert count.
Jan 8, 2026

DR-LoRA: Dynamic Rank LoRA for Fine-Tuning Mixture-of-Experts Models

Mixture-of-Experts (MoE) has become a prominent paradigm for scaling Large Language Models (LLMs). Parameter-efficient fine-tuning methods, such as LoRA, are widely adopted to adapt pretrained MoE LLMs to downstream tasks. However, existing approaches typically assign identical LoRA ranks to all expert modules, ignoring the heterogeneous specialization of pretrained experts. This uniform allocation leads to a resource mismatch: task-relevant experts are under-provisioned, while less relevant ones receive redundant parameters. To address this, we propose DR-LoRA, a Dynamic Rank LoRA framework for fine-tuning pretrained MoE models. Specifically, DR-LoRA initializes all expert LoRA modules with a small active rank and uses an expert saliency score, which combines routing frequency and gradient-based rank importance, to identify which experts would benefit most from additional capacity. It then periodically expands the active ranks of the task-critical expert LoRA, progressively constructing a heterogeneous rank distribution tailored to the target task. Experiments on three MoE models across six tasks show that DR-LoRA consistently outperforms LoRA and other strong baselines, demonstrating that task-adaptive heterogeneous rank allocation is an effective strategy to improve active capacity utilization in MoE fine-tuning.
Nov 9, 2025

Mixtures of SubExperts for Large Language Continual Learning

Enabling lifelong learning in LLMs demands resolving the stability-plasticity dilemma (i.e., models must incorporate new knowledge without overwriting prior representations) while maintaining scalability under bounded parameter growth. Existing PEFT methods fail to satisfy this triad; shared-parameter approaches suffer from catastrophic interference, while task-isolated expansions preclude knowledge transfer and scale linearly. We propose Mixtures of SubExperts (MoSEs), a modular and sparse framework that factorizes model capacity into reusable, compositional primitives. MoSEs augment transformer layers with lightweight SubExperts and a learned sub-routing function that dynamically selects and composes a sparse subset of modules conditioned on task inputs. This induces a structured decomposition of the parameter space where knowledge is localized yet accessible, mitigating interference while preserving reuse. Specifically, MoSEs balance the dilemma via three pillars: (i) stability by isolating knowledge within sparsely activated modules, (ii) plasticity through routing-driven recombination and selective expansion, and (iii) scalability via sublinear growth in effective capacity. Notably, the routing mechanism enables compositional generalization, allowing new tasks to be represented as combinations of previously acquired sub-functions. We empirically validate MoSEs on TRACE and SuperNI, showing reduced forgetting, improved forward transfer, and better parameter efficiency over strong PEFT baselines. MoSEs establish a new Pareto frontier, achieving state-of-the-art performance while maintaining strict parameter budgets. Our results suggest that modular sparsity and compositional routing are key inductive biases for building foundation models that continually learn without saturation.
Oct 13, 2025

Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers

Reinforcement learning (RL) has emerged as a crucial approach for enhancing the capabilities of large language models. However, in Mixture-of-Experts (MoE) models, the routing mechanism often introduces instability, even leading to catastrophic RL training collapse. We analyze the training-inference consistency of MoE models and identify a notable discrepancy in routing behaviors between the two phases. Moreover, even under identical conditions, the routing framework can yield divergent expert selections across repeated forward passes. To address this foundational inconsistency, we propose Rollout Routing Replay (R3), a method that records routing distributions from the inference engine and replays them during training. R3 significantly reduces training-inference policy KL divergence and mitigates extreme discrepancies without compromising training speed. Extensive experiments on various settings confirm that R3 succeeds in stabilizing RL training, preventing collapse and outperforming methods such as GSPO and TIS. We believe this work can offer a new solution for stabilizing RL in MoE models.
Sep 30, 2025

Training Matryoshka Mixture-of-Experts for Elastic Inference-Time Expert Utilization

Mixture-of-Experts (MoE) has emerged as a promising paradigm for efficiently scaling large language models without a proportional increase in computational cost. However, the standard training strategy of Top-K router prevents MoE models from realizing their full potential for elastic inference. When the number of activated experts is altered at inference time, these models exhibit precipitous performance degradation. In this work, we introduce Matryoshka MoE (M-MoE), a training framework that instills a coarse-to-fine structure directly into the expert ensemble. By systematically varying the number of activated experts during training, M-MoE compels the model to learn a meaningful ranking: top-ranked experts collaborate to provide essential, coarse-grained capabilities, while subsequent experts add progressively finer-grained detail. We explore this principle at multiple granularities, identifying a layer-wise randomization strategy as the most effective. Our experiments demonstrate that a single M-MoE model achieves remarkable elasticity, with its performance at various expert counts closely matching that of an entire suite of specialist models, but at only a fraction of the total training cost. This flexibility not only unlocks elastic inference but also enables optimizing performance by allocating different computational budgets to different model layers. Our work paves the way for more practical and adaptable deployments of large-scale MoE models.
Apr 24, 2025

BadMoE: Backdooring Mixture-of-Experts LLMs via Routing-Aware Triggers and Task-Underutilized Experts

Mixture-of-Experts (MoE) architectures are rapidly becoming the standard for building scalable and efficient large language models (LLMs), thanks to their routing mechanism that activates only a subset of experts for each input. However, the modular structure and sparse activation of experts also introduce new security risks, as manipulating a small portion of experts may influence the behavior. In this work, we study a white-box, task-specific model-supply-chain setting and propose BadMoE, a routing-aware backdoor attack against MoE LLMs. We motivate BadMoE through a routing-aware decomposition of the MoE-layer output, which separates the contributions of the selected target experts from those of the remaining experts. This analysis highlights two measurable factors underlying the attack: the routing mass assigned to the target experts and their relative contribution to the layer output. Guided by this insight, BadMoE selects task-underutilized experts and utilizes routing-aware triggers to activate them, enabling reliable activation of task-specific, predefined backdoor behaviors. Specifically, BadMoE involves three steps: 1) identifying experts with low usage on a target-task calibration set, 2) optimizing a routing-aware trigger toward these experts, and 3) associating the selected-expert activations with attacker-specified behavior. We evaluate BadMoE on Mixtral, OLMoE, and Deepseek across sentiment, topic, and emotion classification, sentiment steering, and targeted refusal.
Mar 13, 2025

Ensemble Learning for Large Language Models in Text and Code Generation: A Survey

Generative Pretrained Transformers (GPTs) are foundational Large Language Models (LLMs) for text generation. However, individual LLMs often produce inconsistent outputs and exhibit biases, limiting their representation of diverse language patterns. The closed-source nature of many powerful LLMs further restricts industry applications due to data privacy concerns. Inspired by successes in text generation, LLM ensemble techniques are now increasingly explored for code generation. This article reviews these emerging ensemble approaches to enhance understanding, encourage further research, and promote practical implementation in both text and code generation. We categorize LLM ensembles into seven main methods - weight merging, knowledge fusion, mixture-of-experts, reward ensemble, output ensemble, routing, and cascading - analyzing capabilities of those approaches. Our findings highlight key benefits such as improved diversity representation, enhanced output quality, and greater application flexibility. These insights aid model selection for real-world tasks and crucially, lay groundwork for extending ensemble strategies to multimodal LLMs.
Apr 25, 2024

Prediction Is All MoE Needs: Expert Load Distribution Goes from Fluctuating to Stabilizing

MoE facilitates the development of large models by making the computational complexity of the model no longer scale linearly with increasing parameters. The learning sparse gating network selects a set of experts for each token to be processed; however, this may lead to differences in the number of tokens processed by each expert over several successive iterations, i.e., the expert load fluctuations, which reduces computational parallelization and resource utilization. To this end, we traced and analyzed loads of each expert in the training iterations for several large language models in this work, and defined the transient state with "obvious load fluctuation" and the stable state with "temporal locality". Moreover, given the characteristics of these two states and the computational overhead, we deployed three classical prediction algorithms that achieve accurate expert load prediction results. For the GPT3 350M model, the average error rates for predicting the expert load proportion over the next 1,000 and 2,000 steps are approximately 1.3% and 1.8%, respectively. This work can provide valuable guidance for expert placement or resource allocation for MoE model training. Based on this work, we will propose an expert placement scheme for transient and stable states in our coming work.