4 ms·
They did release the model [0]. A smaller-scale version of it. If you want the big one, all you have to do is scrape 45M web pages, and increase the parameter
by angel_j 8y ago
They did release the model [0]. A smaller-scale version of it. If you want the big one, all you have to do is scrape 45M web pages, and increase the parameters to scale.
What they didn't release is the training corpus and the trained filters. By not releasing these, and focusing on ethics, they are trying to avert attention from the mistake of training a massive cyclops on unfiltered internet text. They probably read a lot of embarrassing text they didn't share, imagine that.
0. https://github.com/openai/gpt-2/ https://github.com/openai/gpt-2/