Bert Production intro Posted on 2020-10-12 In ML 原文链接 we needed to handle over 25,000 inferences per second (and over 1 billion inferences per day), at a latency of under 20ms benchmark的重要性