News for hackers. Stuff that matters.

A comment on GPT-J-6B – A 6 billion parameter, autoregressive text generation model.

Re:GPT-J-6B – A 6 billion parameter, autoregressive text gen... by trianx on Monday July 05, 2021 @03:58AM (#27735573)
We tested it for knowledge extraction against GPT-3. Unfortunately, it's only at the level of GPT-3's smallest ada engine (for which we have found no use). Nowhere near GPT3's strongest model davinci (for which we have a lot of use cases; for their second strongest model curie as well).

Having said that, I am very happy that eleuther.ai works on open-sourcing GPT-x. Eventually, it will be successful.

[ Parent | on HN ]

There are no comments on this comment yet.