How Many Visual Tokens Do Multimodal Language Models Need? Scaling Visual Token Pruning with F^3A

• URL: http://arxiv.org/abs/2605.16359v1
• Date: Sat, 09 May 2026 13:13:04 GMT
• Title: How Many Visual Tokens Do Multimodal Language Models Need? Scaling Visual Token Pruning with F^3A
• Abstract summary: We propose F3A, a training-free router for visual token pruning.<n>It builds lightweight question-conditioned cues, matches them to visual-grid tokens through frozen sparse sensing heads.<n>It allocates a fixed vision token budget via coarse evidence localization, local refinement, coverage-preserving competition, and recovery of under-covered regions.
• Score: 35.028071915491104
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Vision-language models improve perception by feeding increasingly long visual token sequences into language backbones, but the resulting inference cost raises a basic scaling question: as multimodal models grow, how many visual tokens are actually needed, and how should they be allocated under a fixed visual token budget? Existing training-free pruning methods typically answer this with one-shot proxies such as decoder attention, visual similarity, or conditional diversity. We argue that visual token pruning is better viewed as task-conditioned evidence search, especially under aggressive compression and across model scales. We propose F^3A, a training-free router for visual token pruning that operates before the language model consumes image tokens. F^3A builds lightweight question-conditioned cues, matches them to visual-grid tokens through frozen sparse sensing heads, and allocates a fixed vision token budget via coarse evidence localization, local refinement, coverage-preserving competition, and recovery of under-covered regions. It requires no model training, no extra LLM forward pass and preserves the original multimodal prompting and decoding pipeline.

Related papers

• Paying More Attention to Visual Tokens in Self-Evolving Large Multimodal Models [54.097381112138]
  Current self-evolving LMMs struggle on vision-language understanding tasks such as image captioning and visual question answering.<n>We propose VISE, a purely unsupervised self-evolving framework that directly regularizes the model's visual conditioning policy.<n>VISE operates within a single model without specialist roles, external reward models, or annotations.
  arXiv  Detail & Related papers  (2026-06-25T17:59:55Z)
• A More Word-like Image Tokenization for MLLMs [26.120899392740203]
  We propose a novel Disentangled Visual Tokenization (DiVT) that clusters patch embeddings into coherent semantic units.<n>Across diverse multimodal benchmarks, DiVT matches or surpasses baselines with significantly fewer visual tokens.
  arXiv  Detail & Related papers  (2026-05-18T07:09:46Z)
• TrimTokenator-LC: Towards Adaptive Visual Token Pruning for Large Multimodal Models with Long Contexts [6.465999214817427]
  Growing number of visual tokens greatly increases inference cost.<n>Visual token pruning has emerged as a promising solution.<n>Our approach can reduce up to 80% of visual tokens while maintaining performance in long context settings.
  arXiv  Detail & Related papers  (2025-12-28T02:40:56Z)
• See the Text: From Tokenization to Visual Reading [63.10220471118435]
  SeeTok renders text as images (visual-text) and leverages pretrained multimodal computations to interpret them.<n>Across three different language tasks, SeeTok matches or surpasses subword tokenizers while requiring 4.43 times fewer tokens and reducing FLOPs by 70.5%.<n>SeeTok signals a shift from symbolic tokenization to human-like visual reading, and takes a step toward more natural and cognitively inspired language models.
  arXiv  Detail & Related papers  (2025-10-21T17:34:48Z)
• ZSPAPrune: Zero-Shot Prompt-Aware Token Pruning for Vision-Language Models [7.7352936204066]
  We propose a novel, zero-shot method to model visual token pruning as a balance between task relevance and information diversity.<n>Our method achieves performance that matches or surpasses the state-of-the-art with only minimal accuracy loss.<n>These gains are accompanied by significant reductions in GPU memory footprint and inference latency.
  arXiv  Detail & Related papers  (2025-10-20T06:18:47Z)
• TrimTokenator: Towards Adaptive Visual Token Pruning for Large Multimodal Models [4.779482139419908]
  We introduce a mutual information-based token pruning strategy that removes visual tokens semantically with textual tokens.<n>Our method maintains strong performance while reducing textual tokens by 88.9% on models such as LLaVA-15-7B and LLaVA--7B.
  arXiv  Detail & Related papers  (2025-08-30T02:43:50Z)
• PoRe: Position-Reweighted Visual Token Pruning for Vision Language Models [12.189644988996022]
  We present an extremely simple yet effective approach to alleviate the recency bias in visual token pruning.<n>We propose a straightforward reweighting mechanism that adjusts the attention scores of visual tokens according to their spatial positions in the image.<n>Our method, termed Position-reweighted Visual Token Pruning, is a plug-and-play solution that can be seamlessly incorporated into existing visual token pruning frameworks.
  arXiv  Detail & Related papers  (2025-08-25T08:56:32Z)
• CoViPAL: Layer-wise Contextualized Visual Token Pruning for Large Vision-Language Models [75.88232735646018]
  Large Vision-Language Models (LVLMs) process multimodal inputs consisting of text tokens and vision tokens extracted from images or videos.<n>Existing methods attempt to prune redundant vision tokens, revealing substantial redundancy in visual representations.<n>We propose CoViPAL, a layer-wise contextualized visual token pruning method that employs a Plug-and-Play Pruning Module (PPM) to predict and remove redundant vision tokens before they are processed by the LVLM.
  arXiv  Detail & Related papers  (2025-08-24T07:47:00Z)
• VisionThink: Smart and Efficient Vision Language Model via Reinforcement Learning [95.89543460132413]
  Vision-language models (VLMs) have improved performance by increasing the number of visual tokens.<n>However, most real-world scenarios do not require such an extensive number of visual tokens.<n>We present a new paradigm for visual token compression, namely, VisionThink.
  arXiv  Detail & Related papers  (2025-07-17T17:59:55Z)
• VScan: Rethinking Visual Token Reduction for Efficient Large Vision-Language Models [57.2662376527586]
  VScan is a two-stage visual token reduction framework.<n>It addresses token redundancy by: (1) integrating complementary global and local scans with token merging during visual encoding, and (2) introducing pruning at intermediate layers of the language model.<n>VScan achieves a 2.91$times$ speedup in prefilling and a 10$times$ reduction in FLOPs, while retaining 95.4% of the original performance.
  arXiv  Detail & Related papers  (2025-05-28T17:59:08Z)
• ControlMLLM: Training-Free Visual Prompt Learning for Multimodal Large Language Models [73.34709921061928]
  We propose a training-free method to inject visual prompts into Multimodal Large Language Models (MLLMs)<n>We optimize a learnable latent variable based on an energy function, enhancing the strength of referring regions in the attention map.<n>Our method offers a promising direction for integrating referring abilities into MLLMs, and supports referring with box, mask, scribble and point.
  arXiv  Detail & Related papers  (2024-07-31T11:40:29Z)
• LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Models [35.88374542519597]
  Large Multimodal Models (LMMs) have shown significant visual reasoning capabilities by connecting a visual encoder and a large language model. Recent LMMs incorporate more complex visual inputs, such as high-resolution images and videos, which further increases the number of visual tokens significantly. We propose PruMerge, a novel adaptive visual token reduction strategy that significantly reduces the number of visual tokens without compromising the performance of LMMs.
  arXiv  Detail & Related papers  (2024-03-22T17:59:52Z)
• SA$^2$VP: Spatially Aligned-and-Adapted Visual Prompt [59.280491260635266]
  Methods for visual prompt tuning follow the sequential modeling paradigm stemming from NLP. Mymodel model learns a two-dimensional prompt token map with equal (or scaled) size to the image token map. Our model can conduct individual prompting for different image tokens in a fine-grained manner.
  arXiv  Detail & Related papers  (2023-12-16T08:23:43Z)

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.