Let's reproduce the GPT-2 (124M) in llm.c (~4,000 lines of C/CUDA) in 90 minutes for $20. The 124M model is the smallest model in the GPT-2 series released by OpenAI in 2019, and is actually quite ...