4 ms·
Mainly it will help improve latency at inference time. For preprocessing training one can always add more threads, but not at inference time.
by sergeio76 7y ago
Mainly it will help improve latency at inference time. For preprocessing training one can always add more threads, but not at inference time.