Scaling Native Multimodal Pre-Training From Scratch

• URL: http://arxiv.org/abs/2607.22043v1
• Date: Fri, 24 Jul 2026 07:13:52 GMT
• Title: Scaling Native Multimodal Pre-Training From Scratch
• Abstract summary: We investigate the optimal model size and token count for training a vision-language model under a fixed computational budget.<n>We show minimal objective loss adheres to a predictable compute law, whereas compute-optimal model sizes and token counts scale as power laws.<n>Downstream evaluations reveal that native multimodal pre-training induces positive cross-modal transfer.
• Score: 15.178882099650174
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Although large language models (LLMs) exhibit remarkable reasoning capabilities, their reliance on text-only pre-training restricts the perception of the multimodal physical world. Native multimodal pre-training avoids this limitation by training models from scratch on multimodal inputs, thereby achieving deep cross-modal integration and mitigating optimization asymmetries inherent to traditional late-fusion architectures. Despite these advantages, the scaling properties of this paradigm remain systematically uncharacterized. To address this gap, we investigate the optimal model size and token count for training a transformer-based vision-language model under a fixed computational budget. We demonstrate that minimal objective loss adheres to a predictable compute law, whereas compute-optimal model sizes and token counts scale as power laws. Notably, language and multimodal objectives manifest distinct scaling behaviors. The language allocation law is largely invariant to the composition of the data, indicating stable language learning regardless of the multimodal data ratio. Conversely, the multimodal allocation law is highly sensitive to this composition. Specifically, text-heavy mixtures become compute-efficient only at larger model scales, shifting the optimal resource allocation toward greater model capacity. Additionally, by modeling the influence of data composition on compute laws and allocation exponents, we derive an efficiency frontier specifying precise configurations of model size, token count, and data mixture. Downstream evaluations further reveal that native multimodal pre-training induces positive cross-modal transfer, thereby enhancing pure-text spatial reasoning and enabling robust multimodal in-context learning. In summary, this empirical research establishes the essential groundwork for predictably scaling multimodal foundation models.

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)
• DeepInsert: Early Layer Bypass for Efficient and Performant Multimodal Understanding [26.39397960987363]
  We propose a simple modification to pretrained transformer models.<n>Instead of concatenation with the language prompt at the start, we insert multimodal tokens directly into the middle.<n>Our results indicate that our method reduces computational costs during both training and inference.
  arXiv  Detail & Related papers  (2025-04-27T18:56:26Z)
• Latent Thought Models with Variational Bayes Inference-Time Computation [52.63299874322121]
  Latent Thought Models (LTMs) incorporate explicit latent thought vectors that follow an explicit prior model in latent space.<n>LTMs demonstrate superior sample and parameter efficiency compared to autoregressive models and discrete diffusion models.
  arXiv  Detail & Related papers  (2025-02-03T17:50:34Z)
• Scaling Law Hypothesis for Multimodal Model [8.215178209572201]
  We propose a scaling law hypothesis for multimodal models processing text, audio, images, and video within a shared token and embedding space. Our framework predicts model performance based on modality-specific compression and tokenization efficiency, extending established scaling laws from text-based decoder models to mixed-modality systems.
  arXiv  Detail & Related papers  (2024-09-10T16:05:02Z)
• Advancing Multimodal Large Language Models with Quantization-Aware Scale Learning for Efficient Adaptation [70.22782550540714]
  Quantization-aware Scale LeArning method based on multimodal Warmup, termed QSLAW. We introduce a Quantization-aware Scale LeArning method based on multimodal Warmup, termed QSLAW.
  arXiv  Detail & Related papers  (2024-08-07T12:42:09Z)
• Diff-eRank: A Novel Rank-Based Metric for Evaluating Large Language Models [10.677971531050611]
  We introduce a rank-based metric, Diff-eRank, grounded in information theory and geometry principles. For language models, our results show that Diff-eRank increases with model size and correlates well with conventional metrics such as loss and accuracy. In the multi-modal context, we propose an alignment evaluation method based on the eRank, and verify that contemporary multi-modal LLMs exhibit strong alignment performance based on our method.
  arXiv  Detail & Related papers  (2024-01-30T16:19:55Z)
• Improving Discriminative Multi-Modal Learning with Large-Scale Pre-Trained Models [51.5543321122664]
  This paper investigates how to better leverage large-scale pre-trained uni-modal models to enhance discriminative multi-modal learning. We introduce Multi-Modal Low-Rank Adaptation learning (MMLoRA)
  arXiv  Detail & Related papers  (2023-10-08T15:01:54Z)
• Distributionally Robust Multilingual Machine Translation [94.51866646879337]
  We propose a new learning objective for Multilingual neural machine translation (MNMT) based on distributionally robust optimization. We show how to practically optimize this objective for large translation corpora using an iterated best response scheme. Our method consistently outperforms strong baseline methods in terms of average and per-language performance under both many-to-one and one-to-many translation settings.
  arXiv  Detail & Related papers  (2021-09-09T03:48:35Z)
• WenLan: Bridging Vision and Language by Large-Scale Multi-Modal Pre-Training [71.37731379031487]
  We propose a two-tower pre-training model called BriVL within the cross-modal contrastive learning framework. Unlike OpenAI CLIP that adopts a simple contrastive learning method, we devise a more advanced algorithm by adapting the latest method MoCo into the cross-modal scenario. By building a large queue-based dictionary, our BriVL can incorporate more negative samples in limited GPU resources.
  arXiv  Detail & Related papers  (2021-03-11T09:39:49Z)
• Diversity inducing Information Bottleneck in Model Ensembles [73.80615604822435]
  In this paper, we target the problem of generating effective ensembles of neural networks by encouraging diversity in prediction. We explicitly optimize a diversity inducing adversarial loss for learning latent variables and thereby obtain diversity in the output predictions necessary for modeling multi-modal data. Compared to the most competitive baselines, we show significant improvements in classification accuracy, under a shift in the data distribution.
  arXiv  Detail & Related papers  (2020-03-10T03:10:41Z)
• Scaling Laws for Neural Language Models [14.472857826717613]
  We study scaling laws for language model performance on the cross-entropy loss. The loss scales as a power-law with model size, dataset size, and the amount of compute used for training.
  arXiv  Detail & Related papers  (2020-01-23T03:59:20Z)

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.