Skip to content

Distributed data parallel: divide by model.BATCHES in allreduce? #9

@hzyjerry

Description

@hzyjerry

Hi, this might be a minor thing, but I'm wondering in distributed data parallel, when we aggregate grad_weights from all machines using model.allgather, since allgather performs sum operation, shouldn't we further divide grad_weights by model.BATCHES?

Thank you!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions