Skip to content

[feature request] Core API for invertible/inplace and flow-like ops + memory-saving (hookless?) reversible sequential container for RevNets to allow for much larger batch-sizes in academic setting #23756

Open
Open
[feature request] Core API for invertible/inplace and flow-like ops + memory-saving (hookless?) reversible sequential container for RevNets to allow for much larger batch-sizes in academic setting#23756
Labels
featureA request for a proper, new feature.high prioritymodule: distributionsRelated to torch.distributionsmodule: nnRelated to torch.nnneeds designWe want to add this feature but we need to figure out how firsttriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
@vadimkantorov

Description

@vadimkantorov
vadimkantorov
opened on Aug 3, 2019

Inplace BatchNorm seems to be developed by Mapillary here: https://github.com/mapillary/inplace_abn

This would be a very nice addition to core PyTorch (for memory savings).

cc @ezyang @gchanan @zou3519 @bdhirsh @jbschlosser @fritzo @neerajprad @alicanb @nikitaved @albanD @mruberry @walterddr @anjali411 @vincentqb @vishwakftw @ssnl

Metadata

Metadata

Assignees

No one assigned

    Labels

    featureA request for a proper, new feature.high prioritymodule: distributionsRelated to torch.distributionsmodule: nnRelated to torch.nnneeds designWe want to add this feature but we need to figure out how firsttriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions