https://blog.crazyai.uk/tags/llmc/Recent content in Llmc on Zero to One AIHugoenThu, 27 Aug 2026 10:30:00 +0800https://blog.crazyai.uk/posts/c-field-notes-02-there-is-no-autograd-here/Thu, 27 Aug 2026 10:30:00 +0800https://blog.crazyai.uk/posts/c-field-notes-02-there-is-no-autograd-here/llm.c's CPU reference does not build a graph. Each layer is a handwritten forward paired with a handwritten backward, and training is forty-one updates of floats.https://blog.crazyai.uk/posts/c-field-notes-01-python-writes-the-answer-sheet/Thu, 27 Aug 2026 10:20:00 +0800https://blog.crazyai.uk/posts/c-field-notes-01-python-writes-the-answer-sheet/I expected llm.c's Python file to be the real trainer. It is mostly a GPT-2 reference plus a machine that prints the answers C has to match.