Predictive Auto Tuning
Luis discusses the innovative approach of using predictive models to determine the fastest code execution without running all alternatives. By leveraging features of the code and hardware, these models can significantly reduce evaluation time, making the tuning process millions of times faster. Additionally, the conversation touches on the potential for transfer learning, allowing insights gained from one hardware architecture to benefit others.In this clip
From this podcast
Software Engineering Radio - the podcast for professional software developers
Episode 479: Luis Ceze on the Apache TVM Machine Learning Compiler
Related Questions