[11/1] 김태수, DietCode: Automatic optimization for dynamic tensor program

[tae-su-kim]

When : 11/1
Who : 김태수 (SqueezeBits Inc., [email protected])
What
Title: DietCode: Automatic optimization for dynamic tensor program
Links: https://www.amazon.science/publications/dietcode-automatic-optimization-for-dynamic-tensor-program
Keywords: Auto-scheduler, deep learning compiler, dynamic-shaped tensors
Preliminaries:

• Papers
  Learning to Optimize Tensor Programs (https://arxiv.org/pdf/1805.08166.pdf)
  Ansor: Generating High-Performance Tensor Programs for Deep Learning (https://www.usenix.org/conference/osdi20/presentation/zheng)

• Additional Materials
  https://github.com/merrymercy/awesome-tensor-compilers (관련 논문 list-up)
  https://www.youtube.com/watch?v=QXp5ebZzLuE (TVM 세미나, Tianqi Chen)

주로 TVM - Ansor - DietCode 계보를 따라 설명을 진행할 생각이어서 관심이 있으시다면 Learning to Optimize Tensor Programs 페이퍼를 읽고 autoTVM 관련 내용을 먼저 공부 해 오시면 발표를 수월하게 들으실 수 있습니다!

발표자료
221101_DietCode_taesukim.pdf