Study from Meta researchers suggests that training LLMs to predict multiple tokens at once, instead of just the next token, results in better and faster models (Ben Dickson/VentureBeat)

Ben Dickson / VentureBeat:
Study from Meta researchers suggests that training LLMs to predict multiple tokens at once, instead of just the next token, results in better and faster models  —  In a recent study, researchers at Meta, Ecole des Ponts ParisTech and Université Paris-Saclay suggest improving the accuracy …

from Techmeme https://ift.tt/0Im1xp3