Unrolling thread...
How LLM Inference Works