LLaVA-UHD v3: Progressive Visual Compression for Efficient Native-Resolution Encoding in MLLMs - View it on GitHub