Base VLM model for GroundingPI-nano trained on general image-text pairs, specialized data, video captions, and text-only data.