1.karpathy / llm.cLLM training in simple, raw C/CUDACudaMIT31K total3.8K forks+1,769 starsfrom 29,618 · score 7.48
2.HigherOrderCO / HVM1HVM1 (2022): a massively parallel, optimal functional runtime in RustCudaMIT11K total437 forks+179 starsfrom 11,172 · score 5.19