Skip to content

What the dtype of input in Float8Linear backward? #1850

Open
@yh8899

Description

@yh8899

In Float8Linear forward input is saved in high precision,

Image Why not save input in float8? I don't know if I understand this correctly.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

questionFurther information is requested

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions