Hacker News new | ask | show | jobs
by calaphos 336 days ago
Inference throughout scales really well with larger batch sizes (at the cost of latency) due to rising arithmetic intensity and the fact that it's almost always memory BW limited.