karpathy/llm.c: LLM training in simple, raw C/CUDA

LLM training in simple, pure C/CUDA. The post provides instructions on downloading and tokenizing datasets, initializing the model with GPT-2 weights, and decoding token ids back to text.