Method and system for training predictor and computer readable storage medium

• Main Authors: CHEN YILIN, HUANG YALIN, HONG SUNGUL, CAI YIMIN, LI HUAITING, ZHAN CHENGSHENG, CAI RENJIE
• Format: Patent
• Language: Chinese; English
• Published: 01.03.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

Aspects of the present application provide a method for training a predictor that predicts performance of a plurality of machine learning (ML) models on a plurality of platforms. For example, the method may include converting each ML model of the plurality of ML models into a plurality of instructions or the plurality of instructions with a plurality of intermediate representations (IRs). The method may also include simulating execution of a plurality of instructions corresponding to each of the plurality of ML models on a platform, and generating a plurality of commanding performance reports. Each of the instruction performance reports is associated with the performance of a plurality of instructions corresponding to one of the plurality of ML models executed on the platform. The method may also include training the predictor with the plurality of instructions or the plurality of IRs as learning features and the plurality of commanded performance reports as learning tags, compiling the predictor into a libra