paper-with-me

홈 › Papers

Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference

2024-05-09 · Zhihang Lin, Mingbao Lin, Luxi Lin, Rongrong Ji

Multimodal large language models (MLLMs) demand considerable computations for inference due to the extensive parameters and the additional input tokens needed for visual information representation. Herein, we introduce Visual Tokens Withdrawal (VTW), a plug-and-play module to boost MLLMs for rapid inference. Our approach is inspired by two intriguing phenomena we have observed: (1) the attention sink phenomenon that is prevalent in LLMs also persists in MLLMs, suggesting that initial tokens and nearest tokens receive the majority of attention, while middle vision tokens garner minimal attention in deep layers; (2) the presence of information migration, which implies that visual information is transferred to subsequent text tokens within the first few layers of MLLMs. As per our findings, we conclude that vision tokens are unnecessary in the deep layers of MLLMs. Thus, we strategically withdraw them at a certain layer, enabling only text tokens to engage in subsequent layers. To pinpoint the ideal layer for VTW, we initially analyze a limited set of tiny datasets and choose the first layer that meets the Kullback-Leibler divergence criterion. Our VTW approach can cut computational overhead by over 40\% across diverse multimodal tasks while maintaining performance.

📄 PDF Abstract BibTeX arXiv:2405.05803

Code (1)

lzhxmu/vtw 공식 구현 pytorch

Methods 이 논문이 사용한 방법론

SET Dynamic Sparse Training method where weight mask is updated randomly periodically

Similar Papers 제목 키워드 기반

LEO: Boosting Mixture of Vision Encoders for Multimodal Large Language Models

2025-01-13 · Mozhgan Nasr Azadani, James Riddell, Sean Sedwards, Krzysztof Czarnecki

Enhanced visual understanding serves as a cornerstone for multimodal large language models (MLLMs). Recent hybrid MLLMs incorporate a mixture of vision experts to address the limitations of using a single vision encoder …

Autonomous Driving

Query-based Cross-Modal Projector Bolstering Mamba Multimodal LLM

2026-06-03 · SooHwan Eom, Jay Shim, Gwanhyeong Koo, Haebin Na 외 arxiv

The Transformer's quadratic complexity with input length imposes an unsustainable computational load on large language models (LLMs). In contrast, the Selective Scan Structured State-Space Model, or Mamba, addresses this…

Recoverable Compression: A Multimodal Vision Token Recovery Mechanism Guided by Text Information

2024-09-02 · Yi Chen, Jian Xu, Xu-Yao Zhang, Wen-Zhuo Liu 외

With the advancement of large-scale language modeling techniques, large multimodal models combining visual encoders with large language models have demonstrated exceptional performance in various visual tasks. Most of th…

Language ModelingLanguage ModellingLarge Language Model

Latent Noise Mask for Reducing Visual Redundancy in Multimodal Large Language Models

2026-06-29 · Kai Jiang, Ruishu Zhu, Siqi Huang, Hongyuan Zhang 외 arxiv

Multimodal large language models (MLLMs) often fail in fine-grained visual reasoning, as question-relevant visual cues are diluted by dense and redundant image tokens. Recent multimodal reasoning methods usually extend c…

Multimodal ReasoningVisual Reasoning

Visual-Guided Key-Token Regularization for Multimodal Large Language Model Unlearning

2026-01-29 · Chengyi Cai, Zesheng Ye, Peike Li, Bo Han 외 arxiv

Unlearning in Multimodal Large Language Models (MLLMs) prevents the model from revealing private information when queried about target images. Existing MLLM unlearning methods largely adopt approaches developed for LLMs.…

Question Answering