LLaDA2.0-Uni: Unifying Multimodal Understanding and Generation with Diffusion Large Language Model

TL;DR

LLaDA2.0-Uni is a unified multimodal diffusion large language model that integrates understanding and generation of text and images, achieving high performance with efficient inference.

Contribution

It introduces a novel architecture combining discrete tokenization, MoE backbone, and diffusion decoding for unified multimodal processing.

Findings

Matches specialized VLMs in multimodal understanding

Delivers high-fidelity image generation and editing

Supports interleaved generation and reasoning

Abstract

We present LLaDA2.0-Uni, a unified discrete diffusion large language model (dLLM) that supports multimodal understanding and generation within a natively integrated framework. Its architecture combines a fully semantic discrete tokenizer, a MoE-based dLLM backbone, and a diffusion decoder. By discretizing continuous visual inputs via SigLIP-VQ, the model enables block-level masked diffusion for both text and vision inputs within the backbone, while the decoder reconstructs visual tokens into high-fidelity images. Inference efficiency is enhanced beyond parallel decoding through prefix-aware optimizations in the backbone and few-step distillation in the decoder. Supported by carefully curated large-scale data and a tailored multi-stage training pipeline, LLaDA2.0-Uni matches specialized VLMs in multimodal understanding while delivering strong performance in image generation and editing. Its native support for interleaved generation and reasoning establishes a promising and scalable paradigm for next-generation unified foundation models. Codes and models are available at https://github.com/inclusionAI/LLaDA2.0-Uni.