Notes on OneFlow paper

Table of Contents


Distributed Deep Learning is bottlenecked by data movement. Translation from logical graph to physical graph is neither automatic nor optimal.

Proposed solution:

  • Data movement as first-class member like computations.
  • Manage the dependencies.
  • Actor-based runtime for asynchorous execution of physical graph.

Author: expye(Zihao Ye)



Last modified: 2022-12-27 Tue 07:18

Licensed under CC BY-NC 4.0