Skip to content

The NVFP4-converted weights are invalid for inference. #877

@baihei-1122

Description

@baihei-1122

For NVFP4 inference, the input_global_scale and alpha parameters are required. These two values are derived from the existing weight_global_scale and the input_absmax parameter, yet the latter is currently unavailable.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    Status

    Todo

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions