Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

To add MPI allReduce distributed training to MXNet and make it cooperate with existing components, like dependency engine, execution engine, operators, NDArray, a new KVStore-MPI is implemented to extend the functionality of the existing KVStore in MXNet. In this section, we will introduce the design and APIs of KVStore-MPI.

Image Added

Implementation

1)      CPP

...