UniRL-Zero: Reinforcement Learning on Unified Models with Joint Language Model and Diffusion Model Experts

TL;DR

UniRL-Zero introduces a unified RL framework that enhances multimodal understanding and generation by integrating language and diffusion models, establishing systematic baselines for reinforcement learning in unified models.

Contribution

It proposes a novel unified reinforcement learning framework combining language and diffusion models, with defined scenarios and systematic baselines for multimodal tasks.

Findings

Enhanced multimodal understanding and reasoning.

Improved multimedia generation capabilities.

Established systematic RL baselines for unified models.

Abstract

We present UniRL-Zero, a unified reinforcement learning (RL) framework that boosts, multimodal language model understanding and reasoning, diffusion model multimedia generation, and their beneficial interaction capabilities within a unified model. Our work defines six scenarios for unified model reinforcement learning, providing systematic baselines for reinforcement learning of unified understanding and generation model. Our code is available at https://github.com/G-U-N/UniRL.