REPRODUCING GPT-2 (124M) IN LLM.C IN 90 MINUTES FOR $20

Summary from the TLDR AI newsletter

Andrej Karpathy has released an update to LLM C, a single self contained GPT-2 implementation aimed at reproducing the 2019 suite of models. With this most recent update, the library is able to train the smallest of these models in just 90 minutes. It runs end to end and has minimal dependencies.