Thanks to visit codestin.com
Credit goes to github.com

Skip to content

bfloat16 support for stable half precision training #7

@KickCellarDoor

Description

@KickCellarDoor

Thanks for the awesome job! Currently nr3d_lib supports float64, float32 and float16, I think bfloat16 support can be introduced without too much effort and for my experience it can stabilize NeRF training with half precision. Hope you can consider this option later, thanks!

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