Outperform GPT-3 with Karpathy's llm.c using just 1/3 training tokens

null

Read more here: External Link