NanoEuler brings GPT-2-scale LLM training to bare metal C and CUDA

A from-scratch implementation of a GPT-2 sized language model in C with CUDA support, NanoEuler strips away frameworks to expose the raw mechanics of LLM training. For makers building AI infrastructure or learning how transformers actually work, this is a reference implementation worth studying—no PyTorch abstraction layer required.