VARGPT-v1.1: Improve Visual Autoregressive Large Unified Model via Iterative Instruction Tuning and Reinforcement Learning

2025-04-03

AI Summary

This work addresses the challenge of unifying architectural design and jointly improving performance across multimodal understanding, text-to-image generation, and image editing in vision autoregressive large models. To this end, we propose VARGPT-v1.1—the first end-to-end unified model supporting both “vision understanding (next-token prediction)” and “image generation/editing (next-scale synthesis)” paradigms. Methodologically, we introduce a novel integration of iterative vision-instruction fine-tuning and Direct Preference Optimization (DPO)-based reinforcement learning, alongside an 8.3M-sample high-quality vision-generation instruction dataset. Built upon the Qwen2 language backbone, the model incorporates a high-resolution image synthesis module. Experiments demonstrate state-of-the-art performance on multimodal understanding and text-to-image instruction-following benchmarks; notably, zero-shot image editing capability emerges without task-specific supervision. The code and model weights are publicly released.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

In this work, we present VARGPT-v1.1, an advanced unified visual autoregressive model that builds upon our previous framework VARGPT. The model preserves the dual paradigm of next-token prediction for visual understanding and next-scale generation for image synthesis. Specifically, VARGPT-v1.1 integrates: (1) a novel training strategy combining iterative visual instruction tuning with reinforcement learning through Direct Preference Optimization (DPO), (2) an expanded training corpus containing 8.3M visual-generative instruction pairs, (3) an upgraded language model backbone using Qwen2, (4) enhanced image generation resolution, and (5) emergent image editing capabilities without architectural modifications. These advancements enable VARGPT-v1.1 to achieve state-of-the-art performance in multimodal understanding and text-to-image instruction-following tasks, demonstrating significant improvements in both comprehension and generation metrics. Notably, through visual instruction tuning, the model acquires image editing functionality while maintaining architectural consistency with its predecessor, revealing the potential for unified visual understanding, generation, and editing. Our findings suggest that well-designed unified visual autoregressive models can effectively adopt flexible training strategies from large language models (LLMs), exhibiting promising scalability. The codebase and model weights are publicly available at https://github.com/VARGPT-family/VARGPT-v1.1.

Problem

• Enhancing visual autoregressive model via iterative instruction tuning
• Improving multimodal understanding and image generation resolution
• Enabling image editing without architectural changes

Innovation

• Iterative visual instruction tuning with DPO
• Expanded 8.3M visual-generative instruction pairs
• Upgraded Qwen2 language model backbone