Implement a ChatGPT-like LLM in PyTorch from scratch, step by step
[NeurIPS'23 Oral] Visual Instruction Tuning (LLaVA) built towards GPT-4V level capabilities and beyond.