MM-Zero
MM-Zero: Self-Evolving Multi-Model Vision Language Models From Zero Data
Zongxia Li, Hongyang Du, Chengsong Huang et al. · arXiv 2026
Co-evolve proposer, renderer and multimodal solver.
What you will implement
MM-Zero Visual Curriculum Rewards; MM-Zero Coder and Solver Feedback
- 112. MM-Zero Visual Curriculum Rewards
MM-Zero Visual Curriculum Rewards
Aggregate successful and failed render feedback with proposal and diversity rewards.
- 113. MM-Zero Coder and Solver Feedback
MM-Zero Coder and Solver Feedback
Separate Coder execution feedback from Solver majority-agreement and format rewards.