SCHEDULING METHOD, DEVICE, EQUIPMENT AND MEDIUM OF DEEP LEARNING INFERENCE ENGINE

摘要

The present invention discloses a scheduling method, apparatus, device and medium of a deep learning inference engine, which relates to artificial intelligence, deep learning and chip technology. A specific implementation method includes: determining a type of a current inference task in response to a scheduling request of a current inference task derived from an application layer; calculate the total load after each inference engine has performed a current inference task of said type; compare the total load of each of the inference engines and determine a target inference engine that performs the current inference task in each of the inference engines based on the comparison resu An index of the target inference engine indicating a scheduling path of the inference engine is returned to the application layer. An embodiment of the present invention can improve system performance by distributing the reasoning engine according to the actual computing power of each reasoning engine. In addition, when the inference engine is applied to face recognition, the speed and execution efficiency of the face recognition can be improved.