Skip to content

请问下int4 kvcache在做mma的时候是如何反量化的? #3320

@ZL1206

Description

@ZL1206

如题,我想知道int4 kvcache是如何做高效反量化的,做了kv cache的重排吗,还是先ldsm后做warp内的shuffle

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