Training-Free Dynamic Upcycling of Expert Language Models

• URL: http://arxiv.org/abs/2603.29765v1
• Date: Tue, 31 Mar 2026 14:05:10 GMT
• Title: Training-Free Dynamic Upcycling of Expert Language Models
• Abstract summary: We introduce Dynamic Upcycling MoE, a novel approach that reuses dense experts trained on different domains to construct a unified MoE model.<n>Our method builds a single multitask model that preserves the capabilities of the original dense experts without requiring additional training.
• Score: 2.1485350418225244
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) have achieved remarkable performance on a wide range of specialized tasks, exhibiting strong problem-solving capabilities. However, training these models is prohibitively expensive, and they often lack domain-specific expertise because they rely on general knowledge datasets. Expertise finetuning can address this issue; however, it often leads to overspecialization, and developing a single multi-domain expert remains difficult due to diverging objectives. Furthermore, multitask training is challenging due to interference and catastrophic forgetting. Existing work proposes combining the expertise of dense models within a Mixture of Experts (MoE) architecture, although this approach still requires multitask finetuning. To address these issues, we introduce Dynamic Upcycling MoE (DUME), a novel approach that reuses dense experts trained on different domains to construct a unified MoE model. Our method builds a single multitask model that preserves the capabilities of the original dense experts without requiring additional training. DUME is both cost-efficient and scalable: by leveraging the closed-form solution of ridge regression, it eliminates the need for further optimization and enables experts to be added dynamically while maintaining the model's original performance. We demonstrate that DUME consistently outperforms baseline approaches in both causal language modeling and reasoning settings. Finally, we also show that the DUME model can be fine-tuned to further improve performance. We show that, in the causal language modeling setting, DUME can retain up to 97.6% of a dense expert model specialized in one particular domain, and that it can also surpass it in the reasoning setting, where it can achieve 102.1% of the dense expert performance. Our code is available at: github.com/gensyn-ai/dume.

Related papers

• Eigenvectors of Experts are Training-free Non-collapsing Routers [14.774596844618396]
  Sparse Mixture of Experts (SMoE) architectures improve the training efficiency of Large Language Models (LLMs) by routing input tokens to a subset of specialized experts.<n>Despite their remarkable success, both training and inference in SMoE models suffer from the expert collapse issue.<n>We propose Singular Value Decomposition SMoE (SSMoE), a novel and training-free framework that leverages spectral properties of the expert weights to address the collapse issue.
  arXiv  Detail & Related papers  (2026-05-29T08:27:10Z)
• Expert Divergence Learning for MoE-based Language Models [21.857507958048696]
  We introduce Expert Divergence Learning, a novel pre-training strategy that explicitly encourages functional specialization among experts.<n>Our method incorporates a label-driven auxiliary loss that leverages domain labels inherent in pre-training corpora.<n>We validate our approach by pre-training MoE models of up to 15 billion parameters from scratch.
  arXiv  Detail & Related papers  (2026-02-10T06:58:38Z)
• ERNIE 5.0 Technical Report [244.36480708815316]
  ERNIE 5.0 is a unified autoregressive foundation model for unified multimodal understanding and generation across text, image, video, and audio.<n>To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm.<n>We show that ERNIE 5.0 achieves strong and balanced performance across multiple modalities.
  arXiv  Detail & Related papers  (2026-02-04T16:18:15Z)
• Expertise need not monopolize: Action-Specialized Mixture of Experts for Vision-Language-Action Learning [56.129822832095726]
  AdaMoE is a Mixture-of-Experts (MoE) architecture that inherits pretrained weights from dense VLA models.<n>A substantial 21.5% improvement in real-world experiments validates its practical effectiveness for robotic manipulation tasks.
  arXiv  Detail & Related papers  (2025-10-16T04:52:57Z)
• Training Matryoshka Mixture-of-Experts for Elastic Inference-Time Expert Utilization [60.309915093470416]
  Matryoshka MoE (M-MoE) is a training framework that instills a coarse-to-fine structure directly into the expert ensemble.<n>Our work paves the way for more practical and adaptable deployments of large-scale MoE models.
  arXiv  Detail & Related papers  (2025-09-30T16:56:44Z)
• Symphony-MoE: Harmonizing Disparate Pre-trained Models into a Coherent Mixture-of-Experts [18.18231276284727]
  Mixture-of-Experts (MoE) models enable scalable performance by activating large parameter sets sparsely.<n>Recent work employs upcycling, reusing a single pre-trained dense model by replicating its feed-forward network (FFN) layers into experts.<n>This paper addresses this limitation by constructing powerful MoE models using experts sourced from multiple identically-architected but disparate pre-trained models.
  arXiv  Detail & Related papers  (2025-09-23T02:07:14Z)
• SpectR: Dynamically Composing LM Experts with Spectral Routing [37.969478059005574]
  This paper introduces SPECTR, an approach for dynamically composing expert models at each time step during inference.<n>We show that SPECTR improves routing accuracy over alternative training-free methods, increasing task performance across expert domains.
  arXiv  Detail & Related papers  (2025-04-04T13:58:44Z)
• Scalable Multi-Domain Adaptation of Language Models using Modular Experts [10.393155077703653]
  MoDE is a mixture-of-experts architecture that augments a general PLM with modular, domain-specialized experts. MoDE achieves comparable target performances to full parameter fine-tuning while achieving 1.65% better retention performance.
  arXiv  Detail & Related papers  (2024-10-14T06:02:56Z)
• Diversifying the Expert Knowledge for Task-Agnostic Pruning in Sparse Mixture-of-Experts [75.85448576746373]
  We propose a method of grouping and pruning similar experts to improve the model's parameter efficiency.<n>We validate the effectiveness of our method by pruning three state-of-the-art MoE architectures.<n>The evaluation shows that our method outperforms other model pruning methods on a range of natural language tasks.
  arXiv  Detail & Related papers  (2024-07-12T17:25:02Z)
• Harder Tasks Need More Experts: Dynamic Routing in MoE Models [58.18526590138739]
  We introduce a novel dynamic expert selection framework for Mixture of Experts (MoE) models. Our method dynamically selects experts based on the confidence level in expert selection for each input.
  arXiv  Detail & Related papers  (2024-03-12T13:41:15Z)
• Diversifying the Mixture-of-Experts Representation for Language Models with Orthogonal Optimizer [59.43462055143123]
  The Mixture of Experts (MoE) has emerged as a highly successful technique in deep learning. In this study, we shed light on the homogeneous representation problem, wherein experts in the MoE fail to specialize and lack diversity. We propose an alternating training strategy that encourages each expert to update in a direction to the subspace spanned by other experts.
  arXiv  Detail & Related papers  (2023-10-15T07:20:28Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.