arxiv preprint - Mini-Gemini: Mining the Potential of Multi-modality Vision Language Models | AI Breakdown | Podwise