[ifkakao2021] TensorFlow Serving보다 10배 빠르게 서빙하기
kakao tech · 13:45
Kakao's recommendation team took an ad click-through-rate (pCTR) model built in TensorFlow and served it with a custom C++ inference engine. It ran 10.29× faster than TensorFlow Serving, going from about 120,000 to 1,...