Let ViT Speak: Generative Language-Image Pre-training

TL;DR

GenLIP introduces a simple, scalable, and effective generative pretraining framework for Vision Transformers, aligning vision and language modeling for multimodal large language models.

Contribution

It proposes a minimalist, autoregressive pretraining method for ViTs that improves multimodal model performance with less data and complexity.

Findings

Achieves competitive results on multimodal benchmarks.

Matches or surpasses baselines with less pretraining data.

Improves OCR and chart understanding after further training.

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.