Skip to content

Investigate observed fp8 vs fp16 input projections performance difference #35

Open
Open
Investigate observed fp8 vs fp16 input projections performance difference#35
@garykbrixi

Description

@garykbrixi
garykbrixi
opened on Jan 30, 2025

40b, 1b models have significantly lower performance when use_fp8_input_projections is set to false

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions