LaVIT: Unified Language-Vision Pretraining in LLM with Dynamic Discrete Visual Tokenization - View it on GitHub