karpathy/llm.c

LLM training in simple, raw C/CUDA

1,050 best points · 347 comments · 8 Hacker News posts · 31,402 GitHub stars · View on GitHub

Hacker News discussions

• Llm.c – LLM training in simple, pure C/CUDA (1,050 points, 168 comments, by tosh, 2024-04-08)
• Reproducing GPT-2 in llm.c (618 points, 117 comments, by tosh, 2024-05-28)
• Karpathy: Let's reproduce GPT-2 (1.6B): one 8XH100 node 24h $672 in llm.c (182 points, 58 comments, by alecco, 2024-07-11)
• Full forward pass of GPT-2 in one file of pure CUDA (63 points, 4 comments, by tosh, 2024-04-10)
• Layernorm (3 points, 0 comments, by sva_, 2024-04-11)
• Reproducing GPT-2 (124M) in llm.c in 90 minutes for $20 (3 points, 0 comments, by georgehill, 2024-05-28)
• Reproducing GPT-2 (124M) in llm.c in 90 minutes for $20 (3 points, 0 comments, by gmays, 2025-09-17)
• Llm.c State of the Union (1 points, 0 comments, by neeleshs, 2024-05-04)