@voooooogel What’s the difference between “pivot tokens” and other forms of planning ahead that the model does? Is the idea that the output space becomes more collapsed after the token is sampled?
in reply to: 1997410683370283349
same thread: 1997410683370283349 1997413705760547155 1997441452612444240 1997444772668281098 1997518644734042134
Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.