基于英特尔架构的人工智能性能优化.pptxVIP

基于英特尔架构的人工智能性能优化;General Optimization Method AI Optimization on IA Case Sharing NLP (Bert) Speech Synthesis (WaveNet) Game AI (LSTM) Recommender (WDL) CNN based on VNNI Next Step;;;;Optimizations MKL for MatMul BatchMatMul Combine several MatMul into one Operations fusion Emit transpose op with strided MatMul Sgemm parameters exploration;;;Xt;Case 5: WDL Optimization (Feature Column);;Got 3.26x performance boost, after transforming FP32 solution to INT8 solution to leverage VNNI technology, while with similar accuracy.;Keras Model;Grappler: Default graph optimization system in the TF runtime