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

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.

• 1

  LaViDa-R1 enables reasoning in diffusion language models by unifying supervised finetuning, reinforcement learning, and self-distillation into one framework.

• 2

  Replacing KL divergence regularization with supervised finetuning loss prevents training collapse in diffusion models while maintaining exploration.

• 3

  Guided rollout generation solves the vanishing training signal problem when diffusion models fail to generate quality samples for difficult prompts.