Pruning and Distilling Mixture-of-Experts into Dense Language Models

• URL: http://arxiv.org/abs/2605.28207v1
• Date: Wed, 27 May 2026 09:27:36 GMT
• Title: Pruning and Distilling Mixture-of-Experts into Dense Language Models
• Abstract summary: Mixture-of-Experts (MoE) is now the dominant architecture for frontier language models.<n>We present the first systematic framework for converting a trained MoE into a standard fully dense architecture.<n>We evaluate 7 scoring, 5 grouping, and 2 magnitude scaling methods across a range of selected expert counts on Qwen3-30B-A3B, yielding 350 configurations.
• Score: 13.03679207895992
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Mixture-of-Experts (MoE) is now the dominant architecture for frontier language models, yet it requires all expert parameters to be loaded in memory, making it less preferable for memory-constrained deployment. Existing compression methods reduce the number of experts but the output remains an MoE model with the same fundamental limitation. We present the first systematic framework for converting a trained MoE into a standard fully dense architecture: experts are scored, selected, and grouped, then concatenated into a dense FFN and refined by knowledge distillation from the MoE teacher. We evaluate 7 scoring, 5 grouping, and 2 magnitude scaling methods across a range of selected expert counts on Qwen3-30B-A3B, yielding 350 configurations. We find that the choice of scoring method is the most impactful, with our novel diversity-aware scoring consistently outperforming prior methods on Qwen3-30B-A3B, DeepSeek-V2-Lite, and GPT-OSS-20B. Under a controlled comparison at matched parameter count, MoE-to-dense outperforms dense-to-dense pruning by +6.3 pp in average downstream accuracy after ~4B-token distillation at 1.6x faster training wall-clock speed.

Related papers

• LLaDA MoE v2: Scaling Mixture-of-Experts Diffusion Language Models [129.64558694024126]
  Diffusion language models (dLLMs) offer an alternative to autoregressive (AR) language modeling.<n>We characterize how optimization hyper parameters, compute allocation, and architecture scale for MoE dLLMs.<n>We train LLaDA MoE v2, a 30B-A3B dLLM, from scratch on 23.5T tokens.
  arXiv  Detail & Related papers  (2026-08-04T10:53:02Z)
• Attribution-Guided and Coverage-Maximized Pruning for Structural MoE Compression [74.00650541246374]
  We propose a structural pruning framework tailored for MoE models.<n>Our approach reduces memory footprint by 5.27$times and consistently outperforms state-of-the-art baselines.
  arXiv  Detail & Related papers  (2026-06-16T06:53:27Z)
• Turning the TIDE: Cross-Architecture Distillation for Diffusion Large Language Models [25.171094765557996]
  Distilling 8B dense and 16B MoE teachers into a 0.6B student via two heterogeneous pipelines outperforms the baseline by an average of 1.53 points across eight benchmarks.<n>HumanEval scores reach 48.78 compared to 32.3 for the AR baseline.
  arXiv  Detail & Related papers  (2026-04-29T17:59:01Z)
• MoNE: Replacing Redundant Experts with Lightweight Novices for Structured Pruning of MoE [12.498106165046233]
  Mixture-of-Experts (MoE) enables efficient scaling of large language models by activating only a subset of experts per input token.<n>MoNE replaces redundant experts with lightweight novices to achieve effective and robust model compression.
  arXiv  Detail & Related papers  (2025-07-01T03:02:59Z)
• Union of Experts: Adapting Hierarchical Routing to Equivalently Decomposed Transformer [7.230514235208748]
  We propose Union-of-Experts (UoE), which decomposes the transformer model into an equivalent group of experts.<n>In language modeling tasks, UoE achieves an average reduction of 2.38 in perplexity compared to the best-performing MoE method.<n>In image classification, it yields an average accuracy improvement of 1.75% over the best model.
  arXiv  Detail & Related papers  (2025-03-04T11:01:25Z)
• FlexiGPT: Pruning and Extending Large Language Models with Low-Rank Weight Sharing [59.12511498024836]
  We present a method to prune large language models (LLMs) that selectively prunes model blocks based on an importance score.<n>We propose a principled metric to replace each pruned block using a weight-sharing mechanism.<n> Empirical evaluations demonstrate substantial performance gains over existing methods.
  arXiv  Detail & Related papers  (2025-01-24T18:46:37Z)
• HOBBIT: A Mixed Precision Expert Offloading System for Fast MoE Inference [54.40808356999408]
  We present HOBBIT, a mixed precision expert offloading system to enable flexible and efficient MoE inference. Our key insight is that dynamically replacing less critical cache-miss experts with low precision versions can substantially reduce expert-loading latency. HOBBIT achieves up to a 9.93x speedup in decoding compared to state-of-the-art MoE offloading systems.
  arXiv  Detail & Related papers  (2024-11-03T04:25:46Z)
• QuantMoE-Bench: Examining Post-Training Quantization for Mixture-of-Experts [47.01697456105496]
  Mixture-of-Experts (MoE) is a promising way to scale up the learning capacity of large language models.<n>MoE suffers from significant memory overheads due to the vast parameter size.<n>Post-training quantization offers a powerful approach for model compression.
  arXiv  Detail & Related papers  (2024-06-12T12:44:48Z)
• A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts [49.394145046409044]
  This paper provides the first provably efficient technique for pruning experts in finetuned MoE models. We theoretically prove that prioritizing the pruning of the experts with a smaller change of the routers l2 norm from the pretrained model guarantees the preservation of test accuracy. Although our theoretical analysis is centered on binary classification tasks on simplified MoE architecture, our expert pruning method is verified on large vision MoE models.
  arXiv  Detail & Related papers  (2024-05-26T17:52:58Z)

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.