[fenomas]

TFA claims GPT-2 isn't open. Is it that the trained weights are open but the project that did the training isn't?

[alwayslikethis]

Realistically the most difficult part of training a LLM is curating a good data set. OpenAI never published their training data.