Skip to content

Make RIB more memory efficient #324

@nix-apollo

Description

@nix-apollo

Unclear how many of these we will actually need to do.

  • Put hook output matricies on cpu
  • Optionally put models on cpu when possible (moving section to gpu when needed for gradients)
  • Use float32 when possible

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions