How we made one of our largest inference workloads 4.7× more GPU-efficient

1 points | by aray07 5 hours ago