Mixed-R1: Unified Reward Perspective For Reasoning Capability in Multimodal Large Language Models

Understand

Recent works on large language models (LLMs) have successfully demonstrated the emergence of reasoning capabilities via reinforcement learning (RL).

• Although recent efforts leverage group relative policy optimization (GRPO) for MLLMs post-training, they constantly explore one specific aspect, such as grounding tasks, math problems, or chart analysis.
• There are no works that can leverage multi-source MLLM tasks for stable reinforcement learning.
• In this work, we present a unified perspective to solve this problem.