DualSparse-MoE: Coordinating Tensor/Neuron-Level Sparsity with Expert Partition and Reconstruction

• URL: http://arxiv.org/abs/2508.18376v1
• Date: Mon, 25 Aug 2025 18:08:32 GMT
• Title: DualSparse-MoE: Coordinating Tensor/Neuron-Level Sparsity with Expert Partition and Reconstruction
• Authors: Weilin Cai, Le Qin, Shwai He, Junwei Cui, Ang Li, Jiayi Huang,
• Abstract summary: Mixture of Experts (MoE) has become a mainstream architecture for building Large Language Models (LLMs)<n>We identify dual sparsity at the tensor and neuron levels in pre-trained MoE modules as a key factor for both accuracy and efficiency.<n>We propose DualSparse-MoE, an inference system that integrates dynamic tensor-level dropping with static neuron-level reconstruction.
• Score: 15.261077484922616
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Mixture of Experts (MoE) has become a mainstream architecture for building Large Language Models (LLMs) by reducing per-token computation while enabling model scaling. It can be viewed as partitioning a large Feed-Forward Network (FFN) at the tensor level into fine-grained sub-FFNs, or experts, and activating only a sparse subset for each input. While this sparsity improves efficiency, MoE still faces substantial challenges due to their massive computational scale and unpredictable activation patterns. To enable efficient MoE deployment, we identify dual sparsity at the tensor and neuron levels in pre-trained MoE modules as a key factor for both accuracy and efficiency. Unlike prior work that increases tensor-level sparsity through finer-grained expert design during pre-training, we introduce post-training expert partitioning to induce such sparsity without retraining. This preserves the mathematical consistency of model transformations and enhances both efficiency and accuracy in subsequent fine-tuning and inference. Building upon this, we propose DualSparse-MoE, an inference system that integrates dynamic tensor-level computation dropping with static neuron-level reconstruction to deliver significant efficiency gains with minimal accuracy loss. Experimental results show that enforcing an approximate 25% drop rate with our approach reduces average accuracy by only 0.08%-0.28% across three prevailing MoE models, while nearly all degrees of computation dropping consistently yield proportional computational speedups. Furthermore, incorporating load-imbalance awareness into expert parallelism achieves a 1.41x MoE module speedup with just 0.5% average accuracy degradation.

Related papers

• YOLO-Master: MOE-Accelerated with Specialized Transformers for Enhanced Real-time Detection [26.013463778761317]
  YOLO-Master is a novel YOLO-like framework that introduces instance-conditional adaptive computation for Real-Time Object Detection.<n>Our model achieves 42.4% AP with 1.62ms latency, outperforming YOLOv13-N by +0.8% mAP and 17.8% faster inference.
  arXiv  Detail & Related papers  (2025-12-29T07:54:49Z)
• Tensor-Compressed and Fully-Quantized Training of Neural PDE Solvers [10.320585073024455]
  We present a framework that enables scalable and energy-efficient PINN training on edge devices.<n>This work enables real-time PDE solving on edge devices and paves the way for energy-efficient scientific computing at scale.
  arXiv  Detail & Related papers  (2025-12-10T00:00:34Z)
• MC#: Mixture Compressor for Mixture-of-Experts Large Models [86.64315380917827]
  Mixture-of-Experts (MoE) effectively scales large language models (LLMs) and vision-language models (VLMs) by increasing capacity through sparse activation.<n>We propose MC# (Mixture-Compressor-sharp), a framework that combines static quantization and dynamic expert pruning.
  arXiv  Detail & Related papers  (2025-10-13T03:12:46Z)
• PT$^2$-LLM: Post-Training Ternarization for Large Language Models [52.4629647715623]
  Large Language Models (LLMs) have shown impressive capabilities across diverse tasks, but their large memory and compute demands hinder deployment.<n>We propose PT$2$-LLM, a post-training ternarization framework tailored for LLMs.<n>At its core is an Asymmetric Ternary Quantizer equipped with a two-stage refinement pipeline.
  arXiv  Detail & Related papers  (2025-09-27T03:01:48Z)
• DSMoE: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense LLMs [70.91804882618243]
  This paper proposes DSMoE, a novel approach that achieves sparsification by partitioning pre-trained FFN layers into computational blocks.<n>We implement adaptive expert routing using sigmoid activation and straight-through estimators, enabling tokens to flexibly access different aspects of model knowledge.<n>Experiments on LLaMA models demonstrate that under equivalent computational constraints, DSMoE achieves superior performance compared to existing pruning and MoE approaches.
  arXiv  Detail & Related papers  (2025-02-18T02:37:26Z)
• HALO: Hadamard-Assisted Lower-Precision Optimization for LLMs [45.37278584462772]
  We present HALO, a novel quantization-aware training approach for Transformers.<n>Our approach ensures that all large matrix multiplications during the forward and backward passes are executed in lower precision.<n>Applying to LLAMA-family models, HALO achieves near-full-precision-equivalent results during fine-tuning on various tasks.
  arXiv  Detail & Related papers  (2025-01-05T18:41:54Z)
• TensorGRaD: Tensor Gradient Robust Decomposition for Memory-Efficient Neural Operator Training [91.8932638236073]
  We introduce textbfTensorGRaD, a novel method that directly addresses the memory challenges associated with large-structured weights.<n>We show that sparseGRaD reduces total memory usage by over $50%$ while maintaining and sometimes even improving accuracy.
  arXiv  Detail & Related papers  (2025-01-04T20:51:51Z)
• Learning Mixtures of Experts with EM: A Mirror Descent Perspective [28.48469221248906]
  Classical Mixtures of Experts (MoE) are Machine Learning models that involve the input space, with a separate "expert" model trained on each partition.<n>We study theoretical guarantees of the Expectation Maximization (EM) algorithm for the training of MoE models.
  arXiv  Detail & Related papers  (2024-11-09T03:44:09Z)
• SEER-MoE: Sparse Expert Efficiency through Regularization for Mixture-of-Experts [49.01990048827639]
  We introduce SEER-MoE, a framework for reducing both the memory footprint and compute requirements of pre-trained MoE models. The first stage involves pruning the total number of experts using a heavy-hitters counting guidance, while the second stage employs a regularization-based fine-tuning strategy to recover accuracy loss. Our empirical studies demonstrate the effectiveness of our method, resulting in a sparse MoEs model optimized for inference efficiency with minimal accuracy trade-offs.
  arXiv  Detail & Related papers  (2024-04-07T22:13:43Z)
• LUT-GEMM: Quantized Matrix Multiplication based on LUTs for Efficient Inference in Large-Scale Generative Language Models [9.727062803700264]
  We introduce LUT-GEMM, an efficient kernel for quantized matrix multiplication. LUT-GEMM eliminates the resource-intensive dequantization process and reduces computational costs. We show experimentally that when applied to the OPT-175B model with 3-bit quantization, LUT-GEMM substantially accelerates token generation latency.
  arXiv  Detail & Related papers  (2022-06-20T03:48:17Z)
• A Fast and Efficient Conditional Learning for Tunable Trade-Off between Accuracy and Robustness [11.35810118757863]
  Existing models that achieve state-of-the-art (SOTA) performance on both clean and adversarially-perturbed images rely on convolution operations conditioned with feature-wise linear modulation (FiLM) layers. We present a fast learnable once-for-all adversarial training (FLOAT) algorithm, which instead of the existing FiLM-based conditioning, presents a unique weight conditioned learning that requires no additional layer. In particular, we add scaled noise to the weight tensors that enables a trade-off between clean and adversarial performance.
  arXiv  Detail & Related papers  (2022-03-28T19:25:36Z)
• Parameter-Efficient Mixture-of-Experts Architecture for Pre-trained Language Models [68.9288651177564]
  We present a novel MoE architecture based on matrix product operators (MPO) from quantum many-body physics. With the decomposed MPO structure, we can reduce the parameters of the original MoE architecture. Experiments on the three well-known downstream natural language datasets based on GPT2 show improved performance and efficiency in increasing model capacity.
  arXiv  Detail & Related papers  (2022-03-02T13:44:49Z)
• MoEfication: Conditional Computation of Transformer Models for Efficient Inference [66.56994436947441]
  Transformer-based pre-trained language models can achieve superior performance on most NLP tasks due to large parameter capacity, but also lead to huge computation cost. We explore to accelerate large-model inference by conditional computation based on the sparse activation phenomenon. We propose to transform a large model into its mixture-of-experts (MoE) version with equal model size, namely MoEfication.
  arXiv  Detail & Related papers  (2021-10-05T02:14:38Z)
• Activation Density based Mixed-Precision Quantization for Energy Efficient Neural Networks [2.666640112616559]
  We propose an in-training quantization method for neural network models. Our method calculates bit-width for each layer during training a mixed precision model with competitive accuracy. We run experiments on benchmark datasets like CIFAR-10, CIFAR-100, TinyImagenet on VGG19/ResNet18 architectures.
  arXiv  Detail & Related papers  (2021-01-12T09:01:44Z)

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.