4 ms·I think this points to future optimizing will be focused on making inference really inexpensive.by truelson 1y agoI think this points to future optimizing will be focused on making inference really inexpensive.