The weights of Meta's latest large language model, LLaMA, have been leaked via a torrent link posted as part of a pull request on their GitHub repo for the model's inference code.

Approved researchers are able to get hold of the model weights (if approved after filling out a Google Form, an instance of the structured access approach) but they are not supposed to be publicly available.

Public reports should be trustworthy; I will resolve to my judgement (and I won't trade in this market).