paper-with-me

홈 › Papers

PowerAttention: Exponentially Scaling of Receptive Fields for Effective Sparse Attention

2025-03-05 · Lida Chen, Dong Xu, Chenxin An, Xintao Wang, Yikai Zhang, Jiangjie Chen, Zujie Liang, Feng Wei, Jiaqing Liang, Yanghua Xiao, Wei Wang

Large Language Models (LLMs) face efficiency bottlenecks due to the quadratic complexity of the attention mechanism when processing long contexts. Sparse attention methods offer a promising solution, but existing approaches often suffer from incomplete effective context and/or require complex implementation of pipeline. We present a comprehensive analysis of sparse attention for autoregressive LLMs from the respective of receptive field, recognize the suboptimal nature of existing methods for expanding the receptive field, and introduce PowerAttention, a novel sparse attention design that facilitates effective and complete context extension through the theoretical analysis. PowerAttention achieves exponential receptive field growth in $d$-layer LLMs, allowing each output token to attend to $2^d$ tokens, ensuring completeness and continuity of the receptive field. Experiments demonstrate that PowerAttention outperforms existing static sparse attention methods by $5\sim 40\%$, especially on tasks demanding long-range dependencies like Passkey Retrieval and RULER, while maintaining a comparable time complexity to sliding window attention. Efficiency evaluations further highlight PowerAttention's superior speedup in both prefilling and decoding phases compared with dynamic sparse attentions and full attention ($3.0\times$ faster on 128K context), making it a highly effective and user-friendly solution for processing long sequences in LLMs.

📄 PDF Abstract BibTeX arXiv:2503.03588

Code (0)

등록된 구현이 없습니다.

Methods 이 논문이 사용한 방법론

Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…
Attention 설명 없음

Similar Papers 제목 키워드 기반

Covariance properties under natural image transformations for the generalized Gaussian derivative model for visual receptive fields

2023-03-17 · Tony Lindeberg

This paper presents a theory for how geometric image transformations can be handled by a first layer of linear receptive fields, in terms of true covariance properties, which, in turn, enable geometric invariance propert…

Log-Polar Space Convolution

2021-09-29 · Bing Su, Ji-Rong Wen

Convolutional neural networks use regular quadrilateral convolution kernels to extract features. Since the number of parameters increases quadratically with the size of the convolution kernel, many popular models use sma…

Unified theory for joint covariance properties under geometric image transformations for spatio-temporal receptive fields according to the generalized Gaussian derivative model for visual receptive fields

2023-11-17 · Tony Lindeberg

The influence of natural image transformations on receptive field responses is crucial for modelling visual operations in computer vision and biological vision. In this regard, covariance properties with respect to geome…

Log-Polar Space Convolution for Convolutional Neural Networks

2021-07-26 · Bing Su, Ji-Rong Wen

Convolutional neural networks use regular quadrilateral convolution kernels to extract features. Since the number of parameters increases quadratically with the size of the convolution kernel, many popular models use sma…

Scaling Up Graph Neural Networks Via Graph Coarsening

2021-06-09 · Zengfeng Huang, Shengzhong Zhang, Chong Xi, Tang Liu 외

Scalability of graph neural networks remains one of the major challenges in graph machine learning. Since the representation of a node is computed by recursively aggregating and transforming representation vectors of its…

Stochastic Optimization