AIAny
Icon for item

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

Systematically studies how language and vision interact during unified multimodal pretraining, identifies mechanisms that enable modality synergy versus competition, demonstrates the benefit of early joint training, and derives efficient pretraining recipes validated at scale.

Introduction

Vision often gets added to language models as an afterthought, but the training dynamics reveal deeper asymmetries: this paper shows that when modalities are unified early, models avoid relying on language priors, enable productive knowledge transfer, and can be trained far more compute-efficiently than naive recipes suggest. Its core insight is that modality interactions follow regular patterns that can be shaped by data complexity and architectural choices to yield synergy rather than competition.

Key Findings
  • Knowledge Flow: Language, visual understanding, and visual generation transfer knowledge across modalities in distinct and asymmetric ways — knowing these directional effects lets designers target which capability to boost (e.g., better visual generation vs. language-grounded vision).

  • Synergy vs. Competition: Whether modalities help or hurt each other depends largely on data "complexity". Architectural choices that promote shared attention and normalization but preserve modality-specific feed-forward layers consistently encourage synergy across different visual tokenizer designs.

  • Early Unification: Training modalities together from the start outperforms late alignment or sequential schemes. Delaying visual integration produces a "vision laziness" effect where models over-rely on language priors and under-develop visual representations.

  • Practical Recipes and Scaling: The paper derives efficient pretraining recipes that reach strong generative performance using only about 5% of the compute of baseline approaches, and validates the core findings at scale with multiple 13.5B mixture-of-experts models trained on ~2T tokens.

Who it's for and tradeoffs

Great fit if you design or evaluate foundation models that combine vision and language and need principled guidance on when to unify modalities, which architecture choices encourage positive transfer, and how to reduce compute for generative visual capabilities. Look elsewhere if your application is narrowly supervised, small-scale vision-only work, or you cannot afford the dataset engineering needed for early joint training — the benefits rely on appropriate multimodal data and careful architectural choices. The large-scale validations use MoE models and substantial data, so reproducing scale effects may need similar resources.

Methodology (brief)

The conclusions come from controlled experiments on synthetic and large-scale real-world datasets, ablations over architectural variants (attention, normalization, modality-specific FFNs), and scaling runs that confirm the small-scale phenomena persist when training larger MoE models.

Information

  • Websitearxiv.org
  • AuthorsJunlin Han, Shengbang Tong, David Fan, Minghao Chen, Philip Torr, Filippos Kokkinos, Mike Lewis
  • Published date2026/08/05

More Items

Detects and filters spurious token-level teacher supervision in on-policy distillation by estimating input-groundedness and removing high-impact misleading updates, improving OPD on both LLM and VLM benchmarks.

Evaluates how large language models fabricate user attributes in personalization and whether model self-monitoring is a reliable signal. Introduces MirageBench (150 personas, 6 personalization tasks, judge-validated faithfulness taxonomy) and a 12-model leaderboard revealing pervasive over-inference and a 'Self-Monitoring Inversion'.

Orchestrates reasoning, external tool use, and native image generation under one unified multimodal agent policy via post-training. Introduces RAD-GRPO for agentic reinforcement fine-tuning and releases training data plus the full post-training infrastructure.