Let ViT Speak: Generative Language-Image Pre-training

2026-05-01

AI Summary

Effectively aligning the visual encoder with the generative characteristics of autoregressive large language models remains a key challenge in building high-performance multimodal large language models. This work proposes GenLIP, a framework that enables the Vision Transformer to directly predict language tokens from visual tokens using a standard language modeling objective, without relying on contrastive learning or an additional text decoder. This approach unifies visual and linguistic token generation within a single generative paradigm, yielding a structurally simple and highly scalable architecture that significantly reduces dependence on massive pretraining datasets. Trained on only 8 billion samples, GenLIP surpasses strong baselines on detail-sensitive tasks such as OCR and chart understanding, demonstrating exceptional performance.

Abstract

In this paper, we present \textbf{Gen}erative \textbf{L}anguage-\textbf{I}mage \textbf{P}re-training (GenLIP), a minimalist generative pretraining framework for Vision Transformers (ViTs) designed for multimodal large language models (MLLMs). To better align vision encoders with the autoregressive nature of LLMs, GenLIP trains a ViT to predict language tokens directly from visual tokens using a standard language modeling objective, without contrastive batch construction or an additional text decoder. This design offers three key advantages: (1) \textbf{Simplicity}: a single transformer jointly models visual and textual tokens; (2) \textbf{Scalability}: it scales effectively with both data and model size; and (3) \textbf{Performance}: it achieves competitive or superior results across diverse multimodal benchmarks. Trained on 8B samples from Recap-DataComp-1B, GenLIP matches or surpasses strong baselines despite using substantially less pretraining data. After continued pretraining on multi-resolution images at native aspect ratios, GenLIP further improves on detail-sensitive tasks such as OCR and chart understanding, making it a strong foundation for vision encoders in MLLMs.

Problem

• Vision Transformers
• multimodal large language models
• generative pretraining
• language-image alignment
• vision encoders

Innovation

• Generative Pre-training
• Vision Transformer
• Multimodal LLM
• Language Modeling Objective
• Multi-resolution Training