LaViDa-R1: Advancing Reasoning for Unified Multimodal Diffusion Language Models

2026-02-15

AI Summary

This work proposes LaViDa-R1, the first diffusion language model capable of unified multimodal understanding and generation. Addressing the limitation of existing approaches that rely on task-specific reinforcement learning and struggle to handle diverse reasoning tasks cohesively, LaViDa-R1 introduces a unified post-training framework that integrates supervised fine-tuning (SFT) with multi-task reinforcement learning (RL). The framework incorporates several novel strategies, including answer forcing, tree search, and complementary likelihood estimation, to enhance reasoning fidelity and coherence. Evaluated across a broad spectrum of tasks—such as visual mathematical reasoning, complex referential grounding, and image editing—LaViDa-R1 demonstrates consistently superior performance, significantly advancing the model’s generalization and multimodal reasoning capabilities.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Diffusion language models (dLLMs) recently emerged as a promising alternative to auto-regressive LLMs. The latest works further extended it to multimodal understanding and generation tasks. In this work, we propose LaViDa-R1, a multimodal, general-purpose reasoning dLLM. Unlike existing works that build reasoning dLLMs through task-specific reinforcement learning, LaViDa-R1 incorporates diverse multimodal understanding and generation tasks in a unified manner. In particular, LaViDa-R1 is built with a novel unified post-training framework that seamlessly integrates supervised finetuning (SFT) and multi-task reinforcement learning (RL). It employs several novel training techniques, including answer-forcing, tree search, and complementary likelihood estimation, to enhance effectiveness and scalability. Extensive experiments demonstrate LaViDa-R1's strong performance on a wide range of multimodal tasks, including visual math reasoning, reason-intensive grounding, and image editing.

Problem

• diffusion language models
• multimodal reasoning
• unified framework
• multimodal understanding and generation
• reasoning

Innovation

• diffusion language models
• multimodal reasoning
• unified post-training
• multi-task reinforcement learning
• answer-forcing