Understanding and Harnessing Sparsity in Unified Multimodal Models

• URL: http://arxiv.org/abs/2512.02351v1
• Date: Tue, 02 Dec 2025 02:47:29 GMT
• Title: Understanding and Harnessing Sparsity in Unified Multimodal Models
• Authors: Shwai He, Chaorui Deng, Ang Li, Shen Yan,
• Abstract summary: Large multimodal models have achieved remarkable progress in both understanding and generation.<n>Recent efforts pursue unified multimodal models that integrate heterogeneous components to support both capabilities within a single framework.<n>Yet, a systematic understanding of how these inefficiencies manifest across different components remains limited.
• Score: 32.09095929575726
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large multimodal models have achieved remarkable progress in both understanding and generation. Recent efforts pursue unified multimodal models that integrate heterogeneous components to support both capabilities within a single framework. However, such unification introduces inference inefficiencies, e.g., specific tasks or samples may not require the full knowledge or capacity of the unified model. Yet, a systematic understanding of how these inefficiencies manifest across different components remains limited. In this work, we first conduct a systematic analysis of unified multimodal model components using training-free pruning as a probing methodology, considering both depth pruning and width reduction. Our study reveals that the understanding component exhibits notable compressibility in both understanding and generation tasks, which is more pronounced in the latter. In contrast, the generation components are highly sensitive to compression, with performance deteriorating sharply even under moderate compression ratios. To address this limitation, we propose the Mixture-of-Experts (MoE) Adaptation, inspired by the dynamic activation patterns observed across different samples. This approach partitions the generation module into multiple experts and enables sparse activation to restore generation quality. We validate the effectiveness of sparse activation through expert-frozen tuning and further demonstrate that a fully trainable adaptation delivers additional gains. As a result, the adapted BAGEL model achieves performance comparable to the full model while activating only about half of its parameters. The code is released at \href{https://github.com/Shwai-He/SparseUnifiedModel}{this link}.

Related papers

• Plug, Play, and Fortify: A Low-Cost Module for Robust Multimodal Image Understanding Models [6.350443894942629]
  Multimodal Weight Allocation Module (MWAM) is a plug-and-play component that dynamically re-balances the contribution of each branch during training.<n>MWAM delivers consistent performance gains across a wide range of tasks and modality combinations.
  arXiv  Detail & Related papers  (2026-02-26T05:51:41Z)
• NExT-OMNI: Towards Any-to-Any Omnimodal Foundation Models with Discrete Flow Matching [64.10695425442164]
  We introduce NExT-OMNI, an open-source omnimodal foundation model that achieves unified modeling through discrete flow paradigms.<n>Trained on large-scale interleaved text, image, video, and audio data, NExT-OMNI delivers competitive performance on multimodal generation and understanding benchmarks.<n>To advance further research, we release training details, data protocols, and open-source both the code and model checkpoints.
  arXiv  Detail & Related papers  (2025-10-15T16:25:18Z)
• Filling the Gaps: A Multitask Hybrid Multiscale Generative Framework for Missing Modality in Remote Sensing Semantic Segmentation [28.992992584085787]
  Multimodal learning has shown significant performance boost compared to ordinary unimodal models.<n>In real-world scenarios, multimodal signals are susceptible to missing because of sensor failures and adverse weather conditions.<n>We propose a novel Generative-Enhanced MultiModal learning Network (GEMMNet) to tackle these limitations.
  arXiv  Detail & Related papers  (2025-09-14T05:40:35Z)
• OneCAT: Decoder-Only Auto-Regressive Model for Unified Understanding and Generation [91.45421429922506]
  OneCAT is a unified multimodal model that seamlessly integrates understanding, generation, and editing.<n>Our framework eliminates the need for external components such as Vision Transformers (ViT) or vision tokenizer during inference.
  arXiv  Detail & Related papers  (2025-09-03T17:29:50Z)
• Scaling Laws for Native Multimodal Models [53.490942903659565]
  We revisit the architectural design of native multimodal models and conduct an extensive scaling laws study.<n>Our investigation reveals no inherent advantage to late-fusion architectures over early-fusion ones.<n>We show that incorporating Mixture of Experts (MoEs) allows models to learn modality-specific weights, significantly benefiting performance.
  arXiv  Detail & Related papers  (2025-04-10T17:57:28Z)
• DiffMoE: Dynamic Token Selection for Scalable Diffusion Transformers [86.5541501589166]
  DiffMoE is a batch-level global token pool that enables experts to access global token distributions during training.<n>It achieves state-of-the-art performance among diffusion models on ImageNet benchmark.<n>The effectiveness of our approach extends beyond class-conditional generation to more challenging tasks such as text-to-image generation.
  arXiv  Detail & Related papers  (2025-03-18T17:57:07Z)
• Unified Multi-modal Unsupervised Representation Learning for Skeleton-based Action Understanding [62.70450216120704]
  Unsupervised pre-training has shown great success in skeleton-based action understanding. We propose a Unified Multimodal Unsupervised Representation Learning framework, called UmURL. UmURL exploits an efficient early-fusion strategy to jointly encode the multi-modal features in a single-stream manner.
  arXiv  Detail & Related papers  (2023-11-06T13:56:57Z)
• A Deep Unrolling Model with Hybrid Optimization Structure for Hyperspectral Image Deconvolution [50.13564338607482]
  We propose a novel optimization framework for the hyperspectral deconvolution problem, called DeepMix.<n>It consists of three distinct modules, namely, a data consistency module, a module that enforces the effect of the handcrafted regularizers, and a denoising module.<n>This work proposes a context aware denoising module designed to sustain the advancements achieved by the cooperative efforts of the other modules.
  arXiv  Detail & Related papers  (2023-06-10T08:25:16Z)
• Balanced Multimodal Learning via On-the-fly Gradient Modulation [10.5602074277814]
  Multimodal learning helps to comprehensively understand the world, by integrating different senses. We propose on-the-fly gradient modulation to adaptively control the optimization of each modality, via monitoring the discrepancy of their contribution towards the learning objective.
  arXiv  Detail & Related papers  (2022-03-29T08:26:38Z)

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.