Looks similar
70B Llama 2 at 35tokens/second on 4090(github.com)
288 points by Palmik | 108 comments
Llm.c – LLM training in simple, pure C/CUDA(github.com)
810 points by tosh | 134 comments
GitHub repository for llm.c, Andrej Karpathy's project for training LLMs like GPT-2 in simple, raw C/CUDA.






