Skip to content

[Feature]: RTN based INT8 W8A8 scheme support #1468

@thuang6

Description

@thuang6

Feature Description

INT8 W8A8 scheme support was requested. We plan to start from adding RTN based quantization scheme first. Smoothing algo is likely needed for accuracy, will track another issue seperately

considering deployment, compressed-tensor format should be supported

Motivation and Use Case

Target for current Xeon CPU, like GNR.

Alternatives Considered

No response

Definition of Done

No response

Additional Context

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions