🚀 Introducing OpenMMReasoner — a transparent, reproducible recipe for multimodal reasoning. We present a 2-stage pipeline uses 874K SFT samples with step-by-step validation and 74K high-quality RL samples. Paper: More in thread: