paper-with-me

Papers

Exploiting Activation Sparsity with Dense to Dynamic-k Mixture-of-Experts Conversion

2023-10-06 · Filip Szatkowski, Bartosz Wójcik, Mikołaj Piórczyński, Simone Scardapane

Transformer models can face practical limitations due to their high computational requirements. At the same time, such models exhibit significant activation sparsity, which can be leveraged to reduce the inference cost by converting parts of the network into equivalent Mixture-of-Experts (MoE) layers. Despite the crucial role played by activation sparsity, its impact on this process remains unexplored. We demonstrate that the efficiency of the conversion can be significantly enhanced by a proper regularization of the activation sparsity of the base model. Moreover, motivated by the high variance of the number of activated neurons for different inputs, we introduce a more effective dynamic-$k$ expert selection rule that adjusts the number of executed experts on a per-token basis. To achieve further savings, we extend this approach to multi-head attention projections. Finally, we develop an efficient implementation that translates these computational savings into actual wall-clock speedup. The proposed method, Dense to Dynamic-$k$ Mixture-of-Experts (D2DMoE), outperforms existing approaches on common NLP and vision tasks, reducing inference cost by up to 60% without significantly impacting performance.

📄 PDF Abstract BibTeX arXiv:2310.04361

Code (2)

bartwojcik/d2dmoe 공식 구현 pytorch
bartwojcik/sadmoe 공식 구현 pytorch

Tasks

Mixture-of-Experts

Methods 이 논문이 사용한 방법론

BASE 설명 없음
Attention 설명 없음
BPE Byte Pair Encoding, or BPE, is a subword segmentation algorithm that encodes rare and unknown words as sequences of subword units. The intuition is that various word…
Dense Connections Dense Connections, or Fully Connected Connections, are a type of layer in a deep neural network that use a linear operation where every input is connected to every output…
Label Smoothing Label Smoothing is a regularization technique that introduces noise for the labels. This accounts for the fact that datasets may have mistakes in them, so maximizing the…
Adam 설명 없음
Absolute Position Encodings Absolute Position Encodings are a type of position embeddings for [Transformer-based models] where positional encodings are…
Residual Connection 설명 없음

Similar Papers 제목 키워드 기반

Uncovering Intra-expert Activation Sparsity for Efficient Mixture-of-Expert Model Execution

2026-05-09 · Jongseok Park, Sunga Kim, Zhenyu Gu, Ion Stoica 외 arxiv

Mixture of Experts (MoE) architecture has become the standard for state-of-the-art large language models, owing to its computational efficiency through sparse expert activation. However, sparsity through finer expert gra…

Computational Efficiency

Turbo Sparse: Achieving LLM SOTA Performance with Minimal Activated Parameters

2024-06-10 · Yixin Song, Haotong Xie, Zhengyan Zhang, Bo Wen 외

Exploiting activation sparsity is a promising approach to significantly accelerating the inference process of large language models (LLMs) without compromising performance. However, activation sparsity is determined by a…

Mixture-of-Experts

The Impact of Activation Sparsity on Overfitting in Convolutional Neural Networks

2021-04-13 · Karim Huesmann, Luis Garcia Rodriguez, Lars Linsen, Benjamin Risse

Overfitting is one of the fundamental challenges when training convolutional neural networks and is usually identified by a diverging training and test loss. The underlying dynamics of how the flow of activations induce …

ExpertWeaver: Unlocking the Inherent MoE in Dense LLMs with GLU Activation Patterns

2026-02-17 · Ziyu Zhao, Tong Zhu, Zhi Zhang, Tiantian Fan 외 arxiv

Mixture-of-Experts (MoE) effectively scales model capacity while preserving computational efficiency through sparse expert activation. However, training high-quality MoEs from scratch is prohibitively expensive. A promis…

Computational Efficiency

SparseTrain:Leveraging Dynamic Sparsity in Training DNNs on General-Purpose SIMD Processors

2019-11-22 · Zhangxiaowen Gong, Houxiang Ji, Christopher Fletcher, Christopher Hughes 외

Our community has greatly improved the efficiency of deep learning applications, including by exploiting sparsity in inputs. Most of that work, though, is for inference, where weight sparsity is known statically, and/or …

CPU