Saved in:
Bibliographic Details
Main Authors: Zhong, Yi, Qin, Haotong, Zhang, Xindong, Zhang, Lei, Sun, Guolei
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.19929
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918511693004800
author Zhong, Yi
Qin, Haotong
Zhang, Xindong
Zhang, Lei
Sun, Guolei
author_facet Zhong, Yi
Qin, Haotong
Zhang, Xindong
Zhang, Lei
Sun, Guolei
contents Low-bit post-training quantization (PTQ) is a pivotal technique for deploying Vision-Language Models (VLMs) on resource-constrained devices. However, existing PTQ methods often degrade VLMs' accuracy due to the heterogeneous activation distributions of text and vision modalities during quantization. We find that this cross-modal heterogeneity is distributed unevenly across channels: a small subset of channels contains most modality-specific outliers, and these outliers typically reside in different channels for each modality. Motivated by this, we propose SplitQ, a channel-Splitting-driven post-training Quantization framework. At its core, SplitQ introduces a novel Modality-specific Outlier Channel Decoupling (MOCD) module that effectively isolates salient modality-specific outlier channels with minimal overhead. To further address the remaining cross-modal distribution discrepancies, we design an Adaptive Cross-Modal Calibration (ACC) module that employs dual lightweight learnable branches to dynamically mitigate modality-induced quantization errors. Extensive experiments on popular VLMs demonstrate that SplitQ significantly outperforms existing approaches across 6 popular multi-modal datasets under all evaluated quantization settings, including W4A8, W4A4, W3A3, and W3A2. Notably, SplitQ preserves 93.5% of FP16 performance under the challenging W3A3 setting (69.5 vs. 74.3), pushing the efficiency frontier for deploying advanced VLMs. Our code is available at https://github.com/EMVision-NK/SplitQ
format Preprint
id arxiv_https___arxiv_org_abs_2605_19929
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Breaking Modality Heterogeneity in Low-Bit Quantization for Large Vision-Language Models
Zhong, Yi
Qin, Haotong
Zhang, Xindong
Zhang, Lei
Sun, Guolei
Computer Vision and Pattern Recognition
Artificial Intelligence
Low-bit post-training quantization (PTQ) is a pivotal technique for deploying Vision-Language Models (VLMs) on resource-constrained devices. However, existing PTQ methods often degrade VLMs' accuracy due to the heterogeneous activation distributions of text and vision modalities during quantization. We find that this cross-modal heterogeneity is distributed unevenly across channels: a small subset of channels contains most modality-specific outliers, and these outliers typically reside in different channels for each modality. Motivated by this, we propose SplitQ, a channel-Splitting-driven post-training Quantization framework. At its core, SplitQ introduces a novel Modality-specific Outlier Channel Decoupling (MOCD) module that effectively isolates salient modality-specific outlier channels with minimal overhead. To further address the remaining cross-modal distribution discrepancies, we design an Adaptive Cross-Modal Calibration (ACC) module that employs dual lightweight learnable branches to dynamically mitigate modality-induced quantization errors. Extensive experiments on popular VLMs demonstrate that SplitQ significantly outperforms existing approaches across 6 popular multi-modal datasets under all evaluated quantization settings, including W4A8, W4A4, W3A3, and W3A2. Notably, SplitQ preserves 93.5% of FP16 performance under the challenging W3A3 setting (69.5 vs. 74.3), pushing the efficiency frontier for deploying advanced VLMs. Our code is available at https://github.com/EMVision-NK/SplitQ
title Breaking Modality Heterogeneity in Low-Bit Quantization for Large Vision-Language Models
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2605.19929