> Why write in CUDA and not just use PyTorch etc? “LLM training in simple, pure C/CUDA. There is no need for 245MB of PyTorch or 107MB of cPython. […] A few mo

> Why write in CUDA and not just use PyTorch etc? “LLM training in simple, pure C/CUDA. There is no need for 245MB of PyTorch or 107MB of cPython. […] A few more words on what I want this repo to be: First, I want llm.c to be a place for education.”