NanoEuler: a GPT-2-scale LLM built from scratch in pure C/CUDA

vforno built NanoEuler, a GPT-2-scale language model implemented entirely in C and CUDA without deep learning frameworks. For indie makers and solo researchers, it's a rare from-scratch reference that strips away abstraction layers — useful for anyone who wants to understand or tinker with transformer internals without PyTorch or JAX in the way.