The COUNT: Counting N-grams Unlocks Scaling on Host Memory and Training-Free Adaptation
Paper • 2610.20456 • Published • 1
N-gram tables and checkpoints for The COUNT: Counting N-grams Unlocks Scaling on Host Memory and Training-Free Adaptation
Note The paper
Note Packed stupid-backoff n-gram tables COUNT reads from CPU RAM
Note COUNT models, 60M-1B at D/N 10, 20, 40
Note Learned n-gram embedding baseline
Note Standard transformer baseline