From Sparsity to Simplicity: Enabling Simpler Sequential Replacements via Sparse Attention Distillation

• URL: http://arxiv.org/abs/2605.18865v1
• Date: Fri, 15 May 2026 07:47:32 GMT
• Title: From Sparsity to Simplicity: Enabling Simpler Sequential Replacements via Sparse Attention Distillation
• Abstract summary: Self-attention serves as the core foundation of large-scale transformer pretraining, but its quadratic token interaction cost makes inference expensive.<n> Replacing attention with simpler sequential modules is appealing, yet naive substitution is often lossy, especially at larger scales.<n>This paper revisits attention replacement through the lens of sparsity.
• Score: 10.58477116337273
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Self-attention serves as the core foundation of large-scale transformer pretraining, but its quadratic token interaction cost makes inference expensive. Replacing attention with simpler sequential modules is appealing, yet naive substitution is often lossy, especially at larger scales. This paper revisits attention replacement through the lens of sparsity. Based on the observation of diverse sparsity patterns across transformer layers, we posit that pretrained transformers decompose the complex token dependency across tokens into various sequence-to-sequence mappings of diverse complexities, where some layer functionalities can be approximated and replaced with much simpler sequential modules without loss. We evaluate this premise using a plug-and-play layer-wise distillation framework to approximate and replace attention functionalities in pretrained vision transformer models. Controlled group-wise replacements under a fixed training budget reveal a clear pattern: substituting layers with sparser attention incurs substantially smaller accuracy drops than replacing denser ones. We further impose explicit attention sparsity on the pretrained ViT via AViT-style token retention and perform sparsity-guided distillation for sequential replacing models, where we see increasing teacher sparsity consistently reduces the student-teacher gap. The proposed method achieves efficient attention replacement for reduced parameter size and latency through the guidance of attention sparsity.

Related papers

• One-Step Distillation of Discrete Diffusion Image Generators via Fixed-Point Iteration [20.457786704615454]
  Fixed-Point Distillation (FPD) is an end-to-end framework that constructs local correction targets by partially corrupting the student's one-step draft and refining it with a single teacher step.<n>FPD achieves competitive visual fidelity and structural alignment within a single inference step, narrowing the gap to multi-step teachers.
  arXiv  Detail & Related papers  (2026-05-20T17:59:10Z)
• Learn to Rank: Visual Attribution by Learning Importance Ranking [58.69028273772474]
  We propose a new approach to generate visual attribution maps for computer vision models.<n>Our method produces dense, pixel-level attributions in a single forward pass with optional, few-step gradient refinement.<n>Our experiments demonstrate consistent quantitative improvements and sharper, boundary-aligned explanations.
  arXiv  Detail & Related papers  (2026-04-07T12:53:22Z)
• Sparse Attention Post-Training for Mechanistic Interpretability [55.030850996535776]
  We introduce a simple post-training method that makes transformer attention sparse without sacrificing performance.<n>Applying a flexible sparsity regularisation under a constrained-loss objective, we show on models up to 1B parameters that it is possible to retain the original pretraining loss while reducing attention connectivity to $approx 0.3 % of its edges.
  arXiv  Detail & Related papers  (2025-12-05T16:40:08Z)
• Integral Transformer: Denoising Attention, Not Too Much Not Too Little [22.670315809624466]
  Softmax self-attention assigns disproportionate weight to semantically uninformative tokens such as special tokens and punctuation.<n>We propose the Integral Transformer, a novel self-attention mechanism that denoises attention by integrating signals sampled from the logit distribution.<n>Our approach mitigates noise while preserving the contributions of special tokens critical for model performance.
  arXiv  Detail & Related papers  (2025-08-25T18:19:21Z)
• Is Attention Required for Transformer Inference? Explore Function-preserving Attention Replacement [13.38679135071682]
  We propose a Function-preserving Attention Replacement framework that replaces all attention blocks in pretrained transformers with learnable sequence-to-sequence modules.<n>We validate FAR on the DeiT vision transformer family and demonstrate that it matches the accuracy of the original models on ImageNet and multiple downstream tasks with reduced parameters and latency.
  arXiv  Detail & Related papers  (2025-05-24T02:23:46Z)
• Can We Achieve Efficient Diffusion without Self-Attention? Distilling Self-Attention into Convolutions [94.21989689001848]
  We propose (Delta)ConvFusion to replace conventional self-attention modules with Pyramid Convolution Blocks ((Delta)ConvBlocks)<n>By distilling attention patterns into localized convolutional operations while keeping other components frozen, (Delta)ConvFusion achieves performance comparable to transformer-based counterparts while reducing computational cost by 6929$times$ and surpassing LinFusion by 5.42$times$ in efficiency--all without compromising generative fidelity.
  arXiv  Detail & Related papers  (2025-04-30T03:57:28Z)
• Efficient Diffusion Transformer with Step-wise Dynamic Attention Mediators [83.48423407316713]
  We present a novel diffusion transformer framework incorporating an additional set of mediator tokens to engage with queries and keys separately. Our model initiates the denoising process with a precise, non-ambiguous stage and gradually transitions to a phase enriched with detail. Our method achieves a state-of-the-art FID score of 2.01 when integrated with the recent work SiT.
  arXiv  Detail & Related papers  (2024-08-11T07:01:39Z)
• SPION: Layer-Wise Sparse Training of Transformer via Convolutional Flood Filling [1.0128808054306186]
  We propose a novel sparsification scheme for the Transformer that integrates convolution filters and the flood filling method. Our sparsification approach reduces the computational complexity and memory footprint of the Transformer during training. New SPION achieves up to 3.08X speedup over existing state-of-the-art sparse Transformer models.
  arXiv  Detail & Related papers  (2023-09-22T02:14:46Z)
• Calibrating Undisciplined Over-Smoothing in Transformer for Weakly Supervised Semantic Segmentation [51.14107156747967]
  Weakly supervised semantic segmentation (WSSS) has attracted considerable attention because it requires fewer annotations than fully supervised approaches.<n>We propose an Adaptive Re-Activation Mechanism (AReAM) to control deep-level attention to undisciplined over-smoothing.<n>AReAM substantially improves segmentation performance compared with existing WSSS methods, reducing noise while sharpening focus on relevant semantic regions.
  arXiv  Detail & Related papers  (2023-05-04T19:11:33Z)
• Stabilizing Transformer Training by Preventing Attention Entropy Collapse [56.45313891694746]
  We investigate the training dynamics of Transformers by examining the evolution of the attention layers. We show that $sigma$Reparam successfully prevents entropy collapse in the attention layers, promoting more stable training. We conduct experiments with $sigma$Reparam on image classification, image self-supervised learning, machine translation, speech recognition, and language modeling tasks.
  arXiv  Detail & Related papers  (2023-03-11T03:30:47Z)

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.