SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference

Yuan Zhang, Chun-Kai Fan, Junpeng Ma, Wenzhao Zheng, Tao Huang, Kuan Cheng, Denis A Gudovskiy, Tomoyuki Okuno, Yohei Nakata, Kurt Keutzer, Shanghang Zhang
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:74840-74857, 2025.

Abstract

In vision-language models (VLMs), visual tokens usually consume a significant amount of computational overhead, despite their sparser information density compared to text tokens. To address this, most existing methods learn a network to prune redundant visual tokens and require additional training data. Differently, we propose an efficient training-free token optimization mechanism dubbed SparseVLM without extra parameters or fine-tuning costs. Concretely, given that visual tokens complement text tokens in VLMs for linguistic reasoning, we select visual-relevant text tokens to rate the significance of vision tokens within the self-attention matrix extracted from the VLMs. Then we progressively prune irrelevant tokens. To maximize sparsity while retaining essential information, we introduce a rank-based strategy to adaptively determine the sparsification ratio for each layer, alongside a token recycling method that compresses pruned tokens into more compact representations. Experimental results show that our SparseVLM improves the efficiency of various VLMs across a range of image and video understanding tasks. In particular, when LLaVA is equipped with SparseVLM, it achieves a 54% reduction in FLOPs, lowers CUDA time by 37%, and maintains an accuracy rate of 97%. Our code is available at https://github.com/Gumpest/SparseVLMs.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-zhang25s, title = {{S}parse{VLM}: Visual Token Sparsification for Efficient Vision-Language Model Inference}, author = {Zhang, Yuan and Fan, Chun-Kai and Ma, Junpeng and Zheng, Wenzhao and Huang, Tao and Cheng, Kuan and Gudovskiy, Denis A and Okuno, Tomoyuki and Nakata, Yohei and Keutzer, Kurt and Zhang, Shanghang}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {74840--74857}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/zhang25s/zhang25s.pdf}, url = {https://proceedings.mlr.press/v267/zhang25s.html}, abstract = {In vision-language models (VLMs), visual tokens usually consume a significant amount of computational overhead, despite their sparser information density compared to text tokens. To address this, most existing methods learn a network to prune redundant visual tokens and require additional training data. Differently, we propose an efficient training-free token optimization mechanism dubbed SparseVLM without extra parameters or fine-tuning costs. Concretely, given that visual tokens complement text tokens in VLMs for linguistic reasoning, we select visual-relevant text tokens to rate the significance of vision tokens within the self-attention matrix extracted from the VLMs. Then we progressively prune irrelevant tokens. To maximize sparsity while retaining essential information, we introduce a rank-based strategy to adaptively determine the sparsification ratio for each layer, alongside a token recycling method that compresses pruned tokens into more compact representations. Experimental results show that our SparseVLM improves the efficiency of various VLMs across a range of image and video understanding tasks. In particular, when LLaVA is equipped with SparseVLM, it achieves a 54% reduction in FLOPs, lowers CUDA time by 37%, and maintains an accuracy rate of 97%. Our code is available at https://github.com/Gumpest/SparseVLMs.} }
Endnote
%0 Conference Paper %T SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference %A Yuan Zhang %A Chun-Kai Fan %A Junpeng Ma %A Wenzhao Zheng %A Tao Huang %A Kuan Cheng %A Denis A Gudovskiy %A Tomoyuki Okuno %A Yohei Nakata %A Kurt Keutzer %A Shanghang Zhang %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-zhang25s %I PMLR %P 74840--74857 %U https://proceedings.mlr.press/v267/zhang25s.html %V 267 %X In vision-language models (VLMs), visual tokens usually consume a significant amount of computational overhead, despite their sparser information density compared to text tokens. To address this, most existing methods learn a network to prune redundant visual tokens and require additional training data. Differently, we propose an efficient training-free token optimization mechanism dubbed SparseVLM without extra parameters or fine-tuning costs. Concretely, given that visual tokens complement text tokens in VLMs for linguistic reasoning, we select visual-relevant text tokens to rate the significance of vision tokens within the self-attention matrix extracted from the VLMs. Then we progressively prune irrelevant tokens. To maximize sparsity while retaining essential information, we introduce a rank-based strategy to adaptively determine the sparsification ratio for each layer, alongside a token recycling method that compresses pruned tokens into more compact representations. Experimental results show that our SparseVLM improves the efficiency of various VLMs across a range of image and video understanding tasks. In particular, when LLaVA is equipped with SparseVLM, it achieves a 54% reduction in FLOPs, lowers CUDA time by 37%, and maintains an accuracy rate of 97%. Our code is available at https://github.com/Gumpest/SparseVLMs.
APA
Zhang, Y., Fan, C., Ma, J., Zheng, W., Huang, T., Cheng, K., Gudovskiy, D.A., Okuno, T., Nakata, Y., Keutzer, K. & Zhang, S.. (2025). SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:74840-74857 Available from https://proceedings.mlr.press/v267/zhang25s.html.

Related Material