1 comment

[ 2.9 ms ] story [ 13.7 ms ] thread
LLM training in simple, pure C/CUDA. There is no need for 245MB of PyTorch or 107MB of cPython.