Charles Frye

Those optimizations plus Modal's autoscaling, global GPU fleet enabled us to serve as much as half a trillion tokens in a single day -- for just one deployment.