[–] TSltd 6mo ago ↗ Reducing dropped tokens could also improve model training by reducing gradient noise
2 comments
[ 1542 ms ] story [ 760 ms ] thread