Skip to content

Document Triton's alternative to round() #118

Description

@lifengxiang1025

I traced a numerical error in INT8 Quantized MatMul to a misunderstanding of Python's rounding mechanism. Accordingly, it would be beneficial to document Triton's alternative to round()(using tl.extra.cuda.libdevice.rint() as mentioned in triton-lang/triton#4449.)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions