Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes

• URL: http://arxiv.org/abs/2608.05000v2
• Date: Thu, 06 Aug 2026 17:18:02 GMT
• Title: Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes
• Abstract summary: We provide empirical clarity through a systematic exploration of multimodal pretraining.<n>Our experiments on both synthetic and large-scale real-world datasets yield four key insights into the physics of multimodal pretraining.
• Score: 57.55143519213221
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Vision offers a critical axis for advancing foundation models, driving a shift towards natively unified multimodal pretraining. Despite this momentum, the design space and the fundamental mechanisms of how modalities interact during unified training remain underexplored. We provide empirical clarity through a systematic exploration of multimodal pretraining. Our controlled experiments on both synthetic and large-scale real-world datasets yield four key insights into the physics of multimodal pretraining: (i) Knowledge Flow: We disentangle how language, visual understanding, and visual generation transfer knowledge across modalities, revealing distinct patterns of influence and asymmetry; (ii) Synergy vs. Competition: We show that data "complexity" largely determines whether modalities are synergistic, identify architectural choices that promote synergy: such as shared attention and normalization with modality-specific feed-forward layers, and find that these behaviors generalize across different visual tokenizer designs; (iii) Early Unification: Unifying modalities from the very early stages and training them jointly is shown to be more effective than late alignment or sequential training. This process uncovers a vision laziness phenomenon, where delayed integration leads models to rely on language priors; (iv) Recipes: We derive efficient pretraining recipes that achieve strong generative performance using only 5% of the compute budget. These core findings are subsequently validated at scale by training multiple 13.5B MoE models on 2T tokens. We hope this study provides a principled foundation for understanding and scaling multimodal pretraining.

Related papers

• Beyond Language Modeling: An Exploration of Multimodal Pretraining [125.34714978184638]
  We provide empirical clarity through controlled, from-scratch pretraining experiments.<n>We adopt the Transfusion framework, using next-token prediction for language and diffusion for vision.<n>We demonstrate that the MoE architecture harmonizes this scaling asymmetry by providing the high model capacity required by language.
  arXiv  Detail & Related papers  (2026-03-03T18:58:00Z)
• MoCa: Modality-aware Continual Pre-training Makes Better Bidirectional Multimodal Embeddings [75.0617088717528]
  MoCa is a framework for transforming pre-trained VLM backbones into effective bidirectional embedding models.<n>MoCa consistently improves performance across MMEB and ViDoRe-v2 benchmarks, achieving new state-of-the-art results.
  arXiv  Detail & Related papers  (2025-06-29T06:41:00Z)
• Will Pre-Training Ever End? A First Step Toward Next-Generation Foundation MLLMs via Self-Improving Systematic Cognition [89.50068130832635]
  Self-Improving cognition (SIcog) is a self-learning framework for constructing next-generation foundation MLLMs by multimodal knowledge.<n>We propose Chain-of-Description for step-by-step visual understanding and integrate structured Chain-of-Thought (CoT) reasoning to support in-depth multimodal reasoning.<n>Experiments demonstrate SIcog's effectiveness in developing MLLMs with enhanced multimodal cognition.
  arXiv  Detail & Related papers  (2025-03-16T00:25:13Z)
• A Statistical Theory of Contrastive Pre-training and Multimodal Generative AI [35.067337756433965]
  Multi-modal generative AI systems rely on contrastive pre-training to learn representations across different modalities.<n>This paper develops a theoretical framework to explain the success of contrastive pre-training in downstream tasks.
  arXiv  Detail & Related papers  (2025-01-08T17:47:06Z)
• Diving into Self-Evolving Training for Multimodal Reasoning [36.70979791148913]
  Self-evolving trainin has emerged as a key approach for complex reasoning tasks.<n>This paper reframes self-evolving training for multimodal reasoning through the lens of reinforcement learning.<n>We propose M-STAR, a framework that achieves consistent performance gains across models of varying sizes and diverse benchmarks.
  arXiv  Detail & Related papers  (2024-12-23T10:18:41Z)
• Unified View of Grokking, Double Descent and Emergent Abilities: A Perspective from Circuits Competition [83.13280812128411]
  Recent studies have uncovered intriguing phenomena in deep learning, such as grokking, double descent, and emergent abilities in large language models. We present a comprehensive framework that provides a unified view of these three phenomena, focusing on the competition between memorization and generalization circuits.
  arXiv  Detail & Related papers  (2024-02-23T08:14:36Z)
• Multimodal Contrastive Learning via Uni-Modal Coding and Cross-Modal Prediction for Multimodal Sentiment Analysis [19.07020276666615]
  We propose a novel framework named MultiModal Contrastive Learning (MMCL) for multimodal representation to capture intra- and inter-modality dynamics simultaneously. We also design two contrastive learning tasks, instance- and sentiment-based contrastive learning, to promote the process of prediction and learn more interactive information related to sentiment.
  arXiv  Detail & Related papers  (2022-10-26T08:24:15Z)
• i-Code: An Integrative and Composable Multimodal Learning Framework [99.56065789066027]
  i-Code is a self-supervised pretraining framework where users may flexibly combine the modalities of vision, speech, and language into unified and general-purpose vector representations. The entire system is pretrained end-to-end with new objectives including masked modality unit modeling and cross-modality contrastive learning. Experimental results demonstrate how i-Code can outperform state-of-the-art techniques on five video understanding tasks and the GLUE NLP benchmark, improving by as much as 11%.
  arXiv  Detail & Related papers  (2022-05-03T23:38:50Z)

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.