ViSpec: Accelerating Vision-Language Models with Vision-Aware Speculative Decoding

• URL: http://arxiv.org/abs/2509.15235v4
• Date: Sat, 27 Sep 2025 09:39:48 GMT
• Title: ViSpec: Accelerating Vision-Language Models with Vision-Aware Speculative Decoding
• Authors: Jialiang Kang, Han Shu, Wenshuo Li, Yingjie Zhai, Xinghao Chen,
• Abstract summary: We introduce Vision-Aware Speculative Decoding (ViSpec), a novel framework tailored for vision-language models (VLMs)<n>ViSpec employs a lightweight vision adaptor module to compress image tokens into a compact representation.<n>Our training strategy mitigates the risk of the draft model exploiting direct access to the target model's hidden states.
• Score: 13.295759874474767
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Speculative decoding is a widely adopted technique for accelerating inference in large language models (LLMs), yet its application to vision-language models (VLMs) remains underexplored, with existing methods achieving only modest speedups (<1.5x). This gap is increasingly significant as multimodal capabilities become central to large-scale models. We hypothesize that large VLMs can effectively filter redundant image information layer by layer without compromising textual comprehension, whereas smaller draft models struggle to do so. To address this, we introduce Vision-Aware Speculative Decoding (ViSpec), a novel framework tailored for VLMs. ViSpec employs a lightweight vision adaptor module to compress image tokens into a compact representation, which is seamlessly integrated into the draft model's attention mechanism while preserving original image positional information. Additionally, we extract a global feature vector for each input image and augment all subsequent text tokens with this feature to enhance multimodal coherence. To overcome the scarcity of multimodal datasets with long assistant responses, we curate a specialized training dataset by repurposing existing datasets and generating extended outputs using the target VLM with modified prompts. Our training strategy mitigates the risk of the draft model exploiting direct access to the target model's hidden states, which could otherwise lead to shortcut learning when training solely on target model outputs. Extensive experiments validate ViSpec, achieving, to our knowledge, the first substantial speedup in VLM speculative decoding. Code is available at https://github.com/KangJialiang/ViSpec.

Related papers

• Cross-Modal Attention Guided Unlearning in Vision-Language Models [16.460281156521646]
  Vision-Language Models (VLMs) have demonstrated immense capabilities in multi-modal understanding and inference tasks.<n>VLMs add a layer of complexity to this process, as the visual context in the query may also contain sensitive information in addition to the text.<n>We formulate Cross-Modal Attention Guided Unlearning (CAGUL), a lightweight and efficient VLM unlearning framework.
  arXiv  Detail & Related papers  (2025-10-08T21:21:59Z)
• Trust the Model: Compact VLMs as In-Context Judges for Image-Text Data Quality [5.750869893508341]
  Vision-language models (VLMs) extend the conventional large language models by integrating visual data, enabling richer multimodal reasoning.<n>We introduce a streamlined data filtration framework that employs a compact VLM, fine-tuned on a high-quality image-caption annotated dataset.<n>This model effectively evaluates and filters potential training samples based on caption and image quality and alignment.
  arXiv  Detail & Related papers  (2025-07-27T07:20:25Z)
• MASSV: Multimodal Adaptation and Self-Data Distillation for Speculative Decoding of Vision-Language Models [0.09895793818721334]
  We introduce Multimodal Adaptation and Self-Data Distillation for Speculative Decoding of Vision-Language Models (MASSV)<n>MASSV transforms existing small language models into effective multimodal drafters through a two-phase approach.<n>Experiments across the Qwen2.5-VL and Gemma3 model families demonstrate that MASSV increases accepted length by up to 30% and delivers end-to-end inference speedups of up to 1.46x on visually-grounded tasks.
  arXiv  Detail & Related papers  (2025-05-15T17:37:00Z)
• HoVLE: Unleashing the Power of Monolithic Vision-Language Models with Holistic Vision-Language Embedding [91.0552157725366]
  This paper presents a novel high-performance monolithic VLM named HoVLE.<n>It converts visual and textual inputs into a shared space, allowing LLMs to process images in the same way as texts.<n>Our experiments show that HoVLE achieves performance close to leading compositional models on various benchmarks.
  arXiv  Detail & Related papers  (2024-12-20T18:59:59Z)
• ADEM-VL: Adaptive and Embedded Fusion for Efficient Vision-Language Tuning [38.26304604660713]
  ADEM-VL is an efficient vision-language method that tunes models based on pretrained large language models. Our framework surpasses existing methods by an average accuracy of 0.77% on ScienceQA dataset.
  arXiv  Detail & Related papers  (2024-10-23T11:31:06Z)
• TWIST & SCOUT: Grounding Multimodal LLM-Experts by Forget-Free Tuning [54.033346088090674]
  We introduce TWIST & SCOUT, a framework that equips pre-trained MLLMs with visual grounding ability.<n>To fine-tune the model effectively, we generate a high-quality synthetic dataset we call SCOUT.<n>This dataset provides rich supervision signals, describing a step-by-step multimodal reasoning process.
  arXiv  Detail & Related papers  (2024-10-14T13:35:47Z)
• Response Wide Shut: Surprising Observations in Basic Vision Language Model Capabilities [30.176918208200604]
  Vision-Language Models (VLMs) have emerged as general purpose tools for addressing a variety of complex computer vision problems. These models have been shown to be highly capable, but also lacking some basic visual understanding skills. This paper sets out to understand the limitations of SoTA VLMs on fundamental visual tasks.
  arXiv  Detail & Related papers  (2024-08-13T08:26:32Z)
• Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs [61.143381152739046]
  We introduce Cambrian-1, a family of multimodal LLMs (MLLMs) designed with a vision-centric approach.<n>Our study uses LLMs and visual instruction tuning as an interface to evaluate various visual representations.<n>We provide model weights, code, supporting tools, datasets, and detailed instruction-tuning and evaluation recipes.
  arXiv  Detail & Related papers  (2024-06-24T17:59:42Z)
• VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks [89.24440488456405]
  VisionLLM v2 is an end-to-end generalist multimodal large model (MLLM)<n>It unifies visual perception, understanding, and generation within a single framework.
  arXiv  Detail & Related papers  (2024-06-12T16:44:50Z)
• Chain-of-Spot: Interactive Reasoning Improves Large Vision-Language Models [81.71651422951074]
  Chain-of-Spot (CoS) method is a novel approach that enhances feature extraction by focusing on key regions of interest. This technique allows LVLMs to access more detailed visual information without altering the original image resolution. Our empirical findings demonstrate a significant improvement in LVLMs' ability to understand and reason about visual content.
  arXiv  Detail & Related papers  (2024-03-19T17:59:52Z)
• PIN: Positional Insert Unlocks Object Localisation Abilities in VLMs [55.8550939439138]
  Vision-Language Models (VLMs) have shown immense potential by integrating large language models with vision systems. These models face challenges in the fundamental computer vision task of object localisation, due to their training on multimodal data containing mostly captions. We introduce an input-agnostic Positional Insert (PIN), a learnable spatial prompt, containing a minimal set of parameters that are slid inside the frozen VLM. Our PIN module is trained with a simple next-token prediction task on synthetic data without requiring the introduction of new output heads.
  arXiv  Detail & Related papers  (2024-02-13T18:39:18Z)
• MouSi: Poly-Visual-Expert Vision-Language Models [132.58949014605477]
  This paper proposes the use of ensemble experts technique to synergize the capabilities of individual visual encoders. This technique introduces a fusion network to unify the processing of outputs from different visual experts. In our implementation, this technique significantly reduces the positional occupancy in models like SAM, from a substantial 4096 to a more efficient and manageable 64 or even down to 1.
  arXiv  Detail & Related papers  (2024-01-30T18:09:11Z)
• GPT4Image: Large Pre-trained Models Help Vision Models Learn Better on Perception Task [47.1857510710807]
  We present a new learning framework, dubbed GPT4Image, where the knowledge of the large pre-trained models are extracted to help CNNs and ViTs learn better representations.<n>We conduct extensive experiments to verify the effectiveness of the proposed algorithm on various visual perception tasks.
  arXiv  Detail & Related papers  (2023-06-01T14:02:45Z)

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.