paper-with-me

홈 › Papers

MGVQ: Synergizing Multi-dimensional Sensitivity-Aware and Gradient-Hessian Fusion for Vector Quantization

2026-05-20 · Zhong Wang, Zukang Xu, Xing Hu, Dawei Yang arxiv

Vision-Language Models (VLMs) achieve outstanding performance, yet their huge model size severely hinders deployment on edge devices with limited resources. As an efficient model compression technique, vector quantization (VQ) excels in ultra-low-bit representation, which maps model weights to discrete codewords in a compact codebook to cut memory consumption and transmission overhead while preserving model capability. Direct VQ application to VLMs still has two core limitations. First, cross-modality weight distribution differences brought by visual and textual inputs cannot be well fitted by a single unified codebook. Second, current second-order error compensation ignores first-order gradient information, causing weight deviation from pre-trained optimal states, gradient drift and biased compensation results. This work proposes MGVQ, a novel vector quantization framework integrating multi-dimensional sensitivity perception and gradient-Hessian fusion. It consists of two core modules: sensitivity-guided structured mixed-precision quantization dynamically assigns different bit-widths according to channel sensitivity via combined global and local sensitivity analysis for refined resource allocation; gradient-aware second-order error compensation embeds first-order gradients into error correction, and adopts Kronecker and Block-LDL decomposition to ensure low computational cost. Extensive experiments on mainstream VLMs including LLaVA-onevision, InternVL2 and Qwen2-VL verify the effectiveness of MGVQ. In 2-bit quantization settings, MGVQ surpasses existing advanced post-training quantization methods significantly, achieving a maximum accuracy improvement of 4.9 points (71.4% vs 67.0% on InternVL2-26B). The proposed method realizes stable and efficient ultra-low-bit VLM quantization, greatly promoting the practical deployment of multimodal large models in resource-limited environments.

📄 PDF Abstract BibTeX arXiv:2605.24019

Code (0)

등록된 구현이 없습니다.

Tasks

Model Compression

Similar Papers 제목 키워드 기반

MGVQ: Could VQ-VAE Beat VAE? A Generalizable Tokenizer with Multi-group Quantization

2025-07-14 · Mingkai Jia, Wei Yin, Xiaotao Hu, Jiaxin Guo 외

Vector Quantized Variational Autoencoders (VQ-VAEs) are fundamental models that compress continuous visual data into discrete tokens. Existing methods have tried to improve the quantization strategy for better reconstruc…

2kImage GenerationImage ReconstructionQuantization

MGVQ: Could VQ-VAE Beat VAE? A Generalizable Tokenizer with Multi-group Quantization

2025-07-10 · Mingkai Jia, Wei Yin, Xiaotao Hu, Jiaxin Guo 외

Vector Quantized Variational Autoencoders (VQ-VAEs) are fundamental models that compress continuous visual data into discrete tokens. Existing methods have tried to improve the quantization strategy for better reconstruc…

2kQuantization

M3G: Multi-Granular Gesture Generator for Audio-Driven Full-Body Human Motion Synthesis

2025-05-13 · Zhizhuo Yin, Yuk Hang Tsui, Pan Hui

Generating full-body human gestures encompassing face, body, hands, and global movements from audio is a valuable yet challenging task in virtual avatar creation. Previous systems focused on tokenizing the human gestures…

Gesture GenerationMotion Synthesis

Exploring Human-in-the-Loop Test-Time Adaptation by Synergizing Active Learning and Model Selection

2024-05-29 · Yushu Li, Yongyi Su, Xulei Yang, Kui Jia 외

Existing test-time adaptation (TTA) approaches often adapt models with the unlabeled testing data stream. A recent attempt relaxed the assumption by introducing limited human annotation, referred to as Human-In-the-Loop …

Active LearningModel SelectionTest-time Adaptation

Synergizing RAG and Reasoning: A Systematic Review

2025-04-22 · Yunfan Gao, Yun Xiong, Yijie Zhong, Yuxi Bi 외

Recent breakthroughs in large language models (LLMs), particularly in reasoning capabilities, have propelled Retrieval-Augmented Generation (RAG) to unprecedented levels. By synergizing retrieval mechanisms with advanced…

RAGRetrievalRetrieval-augmented Generation