Digging Deep into Model Performing using Across Stack Profiling

Abstract

This talk presents an Across-Stack Profiling (XSP) which is a leveled profiling design that leverages existing profiling tools to give a drill down view of model, system, and hardware bottlenecks. The design does so in spite of the profiling overheads incurred from the profiling. We coupled the profiling capability with an automatic analysis pipeline to systematically characterize over 150 state-of-the-art ML models. Through this characterization, we show that our across-stack profiling solution provides insights (which are difficult to discern otherwise) on the characteristics of ML models, ML frameworks, and GPU hardware.

Date
Apr 18, 2020 3:30 PM
Location
Lausanne, Switzerland