DSMoE: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense LLMs

• URL: http://arxiv.org/abs/2502.12455v1
• Date: Tue, 18 Feb 2025 02:37:26 GMT
• Title: DSMoE: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense LLMs
• Authors: Minxuan Lv, Zhenpeng Su, Leiyu Pan, Yizhe Xiong, Zijia Lin, Hui Chen, Wei Zhou, Jungong Han, Guiguang Ding, Cheng Luo, Di Zhang, Kun Gai, Songlin Hu,
• Abstract summary: This paper proposes DSMoE, a novel approach that achieves sparsification by partitioning pre-trained FFN layers into computational blocks. We implement adaptive expert routing using sigmoid activation and straight-through estimators, enabling tokens to flexibly access different aspects of model knowledge. Experiments on LLaMA models demonstrate that under equivalent computational constraints, DSMoE achieves superior performance compared to existing pruning and MoE approaches.
• Score: 70.91804882618243
• License:
• Abstract: As large language models continue to scale, computational costs and resource consumption have emerged as significant challenges. While existing sparsification methods like pruning reduce computational overhead, they risk losing model knowledge through parameter removal. This paper proposes DSMoE (Dynamic Sparse Mixture-of-Experts), a novel approach that achieves sparsification by partitioning pre-trained FFN layers into computational blocks. We implement adaptive expert routing using sigmoid activation and straight-through estimators, enabling tokens to flexibly access different aspects of model knowledge based on input complexity. Additionally, we introduce a sparsity loss term to balance performance and computational efficiency. Extensive experiments on LLaMA models demonstrate that under equivalent computational constraints, DSMoE achieves superior performance compared to existing pruning and MoE approaches across language modeling and downstream tasks, particularly excelling in generation tasks. Analysis reveals that DSMoE learns distinctive layerwise activation patterns, providing new insights for future MoE architecture design.

Related papers

• LESA: Learnable LLM Layer Scaling-Up [57.0510934286449]
  Training Large Language Models (LLMs) from scratch requires immense computational resources, making it prohibitively expensive. Model scaling-up offers a promising solution by leveraging the parameters of smaller models to create larger ones. We propose textbfLESA, a novel learnable method for depth scaling-up.
  arXiv  Detail & Related papers  (2025-02-19T14:58:48Z)
• Explore Activation Sparsity in Recurrent LLMs for Energy-Efficient Neuromorphic Computing [3.379854610429579]
  Recurrent Large Language Models (R-LLM) have proven effective in mitigating the complexity of self-attention. We propose a low-cost, training-free algorithm to sparsify R-LLMs' activations to enhance energy efficiency on neuromorphic hardware.
  arXiv  Detail & Related papers  (2025-01-09T19:13:03Z)
• Retraining-Free Merging of Sparse MoE via Hierarchical Clustering [14.858134039539697]
  This paper introduces Hierarchical Clustering for Sparsely activated Mixture of Experts (HC-SMoE) HC-SMoE is a task-agnostic expert merging framework for parameter reduction without retraining. We provide theoretical analysis and evaluations across multiple zero-shot language tasks to demonstrate HC-SMoE's effectiveness in state-of-the-art models including Qwen and Mixtral.
  arXiv  Detail & Related papers  (2024-10-11T07:36:14Z)
• Duo-LLM: A Framework for Studying Adaptive Computation in Large Language Models [16.16372459671255]
  Large Language Models (LLMs) typically generate outputs token by token using a fixed compute budget. We propose a novel framework that integrates smaller auxiliary modules within each Feed-Forward Network layer of the LLM. We show that trained routers operate differently from oracles and often yield suboptimal solutions.
  arXiv  Detail & Related papers  (2024-10-01T16:10:21Z)
• Revisiting SMoE Language Models by Evaluating Inefficiencies with Task Specific Expert Pruning [78.72226641279863]
  Sparse Mixture of Expert (SMoE) models have emerged as a scalable alternative to dense models in language modeling. Our research explores task-specific model pruning to inform decisions about designing SMoE architectures. We introduce an adaptive task-aware pruning technique UNCURL to reduce the number of experts per MoE layer in an offline manner post-training.
  arXiv  Detail & Related papers  (2024-09-02T22:35:03Z)
• FactorLLM: Factorizing Knowledge via Mixture of Experts for Large Language Models [50.331708897857574]
  We introduce FactorLLM, a novel approach that decomposes well-trained dense FFNs into sparse sub-networks without requiring any further modifications. FactorLLM achieves comparable performance to the source model securing up to 85% model performance while obtaining over a 30% increase in inference speed.
  arXiv  Detail & Related papers  (2024-08-15T16:45:16Z)
• 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)
• EsaCL: Efficient Continual Learning of Sparse Models [10.227171407348326]
  Key challenge in the continual learning setting is to efficiently learn a sequence of tasks without forgetting how to perform previously learned tasks. We propose a new method for efficient continual learning of sparse models (EsaCL) that can automatically prune redundant parameters without adversely impacting the model's predictive power.
  arXiv  Detail & Related papers  (2024-01-11T04:59:44Z)
• A Multi-Head Ensemble Multi-Task Learning Approach for Dynamical Computation Offloading [62.34538208323411]
  We propose a multi-head ensemble multi-task learning (MEMTL) approach with a shared backbone and multiple prediction heads (PHs) MEMTL outperforms benchmark methods in both the inference accuracy and mean square error without requiring additional training data.
  arXiv  Detail & Related papers  (2023-09-02T11:01:16Z)

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.