VoCoT: Unleashing Visually Grounded Multi-Step Reasoning in Large Multi-Modal Models - View it on GitHub