[xuanlin314]

The lesson-style README is a great approach. Breaking down LLM inference into digestible steps makes the codebase approachable even for people who haven't touched CUDA before.